Process Analytical Chemistry - ACS Publications - American Chemical

Process Analytical Chemistry - ACS Publications - American Chemical...

3 downloads 112 Views 425KB Size

Anal. Chem. 1999, 71, 121R-180R

Process Analytical Chemistry Jerome Workman, Jr.,*,† David J. Veltkamp,‡ Steve Doherty,§ Brian B. Anderson,| Ken E. Creasy,⊥ Mel Koch,‡ James F. Tatera,I Alex L. Robinson,‡ Leonard Bond,+ Lloyd W. Burgess,‡ Gary N. Bokerman,# Alan H. Ullman,∇ Gary P. Darsey,¶ Foad Mozayeni,¶ Judith Ann Bamberger,+ and Margaret Stautberg Greenwood+

Analytical Science & Technology, Kimberly-Clark Corporation, Neenah, Wisconsin 54956, Center for Process Analytical Chemistry (CPAC), University of Washington, Seattle, Washington 98195-1700, Chemical Sciences Group, Monsanto Company/Searle, Skokie, Illinois 60077, Savannah River Technology Center, Westinghouse Savannah River Company, Akine, South Carolina 29808, On-Line Instrumentation Skill Center, AlliedSignal, Inc., Morristown, New Jersey 07962-1021, Process Analysis Expertise Center, Dow Corning Corporation, Carrollton, Kentucky 42701, Pacific Northwest National Laboratory, Richland, Washington 99352, Analytical and Testing Technology Center, Dow Corning Corporation, Midland, Michigan 48686-0995, The Procter & Gamble Company, Center Hill Road, Cincinnati, Ohio 43224, and Analytical Department, Akzo Nobel Chemicals Inc., Dobbs Ferry, New York 10522-3401 Ultrasonic Analysis Ultrasonic Systems and Measurements Velocity and Attenuation to Characterize Media and Monitor Processes Monitoring Solidification (Interface Sensing) Acoustic Time Domain Reflectometry Three-Phase Reactors Process Tomography Using Ultrasonic Methods Ultrasonic Holography Ultrasonic Transducers Density Measurement Ultrasonic Characterization of Multiphase Fluids and Flow Cavitation, Sonochemistry, and Sonoluminescence Miscellaneous Techniques Nuclear Magnetic Resonance Microwave Spectroscopy Future Needs and Directions Literature Cited

Review Contents General Survey of Review Articles


General Reviews Specific Reviews Sampling Systems Chromatography Gas Chromatography Liquid Chromatography Other Chromatographic Techniques Near-Infrared and Infrared Spectroscopy and Imaging Imaging Applications Biotechnology Applications

122R 123R 123R 125R 126R 127R 127R 128R 128R 129R

Earth Sciences and Mineralogy Fine Chemicals and Chemical Production Food and Beverage Applications Medicine and Clinical Chemistry Applications Petroleum, Natural Gas, and Fuel Applications

129R 130R 131R 133R 135R

Pharmaceutical Applications Polymer Applications Process Raman Spectroscopy Process Electronic Spectroscopy (UV-Visible and Fluorescence)

136R 137R 139R 140R

Mass Spectrometry Reviews Compact Instrumentation Fieldable Instrumentation

142R 143R 143R 143R

Inlet Technology Ionization Modes Process Monitoring and Control Environmental Monitoring Aerosol and Particulate Monitoring Process Chemometrics General Information Process Modeling Process Monitoring Process Control Neural Networks Wavelets Flow and Sequential Injection Analysis Chemical Process Monitoring Bioprocess Monitoring

143R 144R 145R 146R 147R 147R 148R 149R 150R 152R 153R 155R 156R 156R 158R

10.1021/a1990007s CCC: $18.00 Published on Web 05/01/1999

© 1999 American Chemical Society

158R 158R 158R 159R 159R 159R 159R 160R 160R 160R 161R 162R 163R 163R 163R 163R 165R

This review of process analytical chemistry is an update to the previous review on this subject published in 1995 (A2). The time period covered for this review includes publications written or published from late 1994 until early 1999, with the addition of a few classic references pointing to background information critical to an understanding of a specific topic area. These older references have been critically included as established fundamental works. New topics covered in this review not previously treated as separate subjects in past reviews include sampling systems, imaging (via optical spectroscopy), and ultrasonic analysis. The individual review subjects are organized into their most obvious subsection. The purpose of this review is to include the more critical work from each topic area in brief summaries, rather than as protracted abstracts. For this review, an expanded †

Kimberly-Clark Corp. University of Washington. § Monsanto Co./Searle. | Westinghouse Savannah River Co. ⊥ AlliedSignal Inc. I Dow Corning Corp., Carrollton, KY. + Pacific Northwest National Laboratory. # Dow Corning Corp., Midland, MI. ∇ The Procter & Gamble Co. ¶ Akzo Nobel Chemicals Inc. ‡

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999 121R

definition of the eras for the implementation of process analyzers has been included. The earlier definition for process analyzers was encompassed by the terms off-line, at-line, on-line, in-line, and noninvasive (A1, A2). A more expanded descriptive list of applicable process analyzer conditions, or eras, as extracted from the various titles of published research papers, includes (in alphabetical order) the following: airborne, at-line, automatic/ automated, fieldable, hyperspectral, imaging/image analysis, in situ, in-line, near-line, noninvasive/nondestructive, on-line, open path, portable/hand-held, process monitoring/production control, quality control/quality assurance, quality monitoring, rapid, real-time, and remote. These types or eras for process analyzers are included under each review topic. Analytical chemistry is a key component of measurement science that plays a valuable role in supporting the characterization of products resulting from research, product and process development, and manufacturing. Process analytical chemistry is now an established field of Analytical Chemistry, as evidenced by the growing number of publications, by sections of symposiums (e.g., FACSS, Pittsburgh Conference, ACS meetings, etc.), and by dedicated meetings such as IFPAC. The field of process analytical chemistry has developed over the past 50 years due to the growing appreciation for having process data gathered close to the production operation. This has occurred for a variety of reasons including time and cost savings, sampling concerns, sample transport, process efficiency, and safety. These reasons contributed to process efficiency and improved process control. With the globalization of industries, productivity and quality have added to the need for effective process control while minimizing the environmental impact. This approach requires a certain level of dedication and resources from the corresponding internal research operations within the organization. This has become more difficult recently as a result of cost reduction activities driven by corporate re-engineering and subsequent resource reduction of research, production maintenance, and plant instrumentation groups. To be most meaningful, process measurement science must tie into the needs in the engineering disciplines related to process control. There is a strong need for cross discipline appreciation, understanding, and cooperation in order to effectively incorporate the new developments in analytical measurement science with the advances in process modeling, monitoring, and control. These new developments are facilitated by continued improvements in the computer-related fields (semiconductor, automation, software, etc.). An important step is not only to incorporate these computerrelated developments into analytical instrumentation but to explore these computing advances for ideas in microinstrumentation. Traditional laboratory-based analytical instrumentation, as well as the process analytical technologies described in this review, will be evaluated as to whether there is value to process control in pursuing them at the microscale. There will be advantages in ruggedness, replication, and subsequent cost where multiple units are desired for useful implementation. In addition, nontraditional measurement and characterization techniques (such as imaging, acoustics, thermal, and rheology approaches) will take on additional importance. This is due to global interest in marketing, technical service and development, and sales of products where 122R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

the measurement and prediction of final product properties becomes key to a successful business. Although process measurements have traditionally included temperature, pressure, and flow rate, more efficiency can be added to processes by measuring composition or structural properties in a manner allowing real-time control during a manufacturing process. This review considers the essential elements to this compositional measurement scheme. These key aspects for process analytical chemistry include chemometrics and process control algorithms, sensors, optical spectroscopy, chromatography, mass spectrometry, flow injection analysis and automated wet chemical methods, ultrasonic analysis, NMR, and other techniques. GENERAL SURVEY OF REVIEW ARTICLES This topic covers published review articles in the period 1994 through 1998, with the intent of extending the review series (A1, A2). The first part will address publications that cover the field of process analysis from a broad perspective. The second part will address reviews focused on a particular industry or product segment and specific techniques. The growing need for industry to thrive in a global economy by cost-competitive, high-quality, and timely products and services provides a key driver for process analytical chemistry. The value of discriminating information from on-line complex process analyzers is being recognized as an important and, more extensively, an integral part of process control systems. The technology base continues to expand with new additions (nuclear magnetic resonance (NMR), Raman, etc.) and more robust designs of existing technology as well as increasingly powerful data handling. Expansion of process analytical chemistry into the mainstream education of chemists and engineers promises to breakdown many of the limitations to broader utility. General Reviews. Clevett (A3) draws on 30 years of experience in the field with references to two market studies to review the trends in application and technology. He cites sources for improved equipment reliability and performance with discussion on the trends in 10 practiced technologies in several industrial applications. The ingredients for successful analyzer installations, including the sampling challenge, and sources of aid are also addressed. Podkulski (A4) focuses on the promising newer techniques of near-infrared (NIR), Raman, NMR, and neural networks. The status of each technology is addressed along with the challenges to broader utility. The universal requirements are improvements in sample systems and information processing of both the analyzer output and maintenance diagnostics. Analytical Instrumentation, edited by Sherman and Rhodes (A5), is in The ISA Practical Guide Series which bridges theory with actual industrial practices. This publication addresses all aspects of on-line analyzer use, from application justification in specific industries, interfaces, sampling, effective operation, and maintenance in the first nine chapters. The subsequent 27 chapters address the measurement principles, common applications with possible case histories, limitations, and key operational insights of the widely used process analyzers. Physical property analyzers comprise 14 of these chapters, electrochemical 4, spectroscopic analyzers (infrared (IR), ultraviolet (UV), visible) 5, and compositional analyzers (gas chromatography (GC), liquid

chromatography (LC), mass spectrometry (MS)) the final 4 chapters. The recent edition of the Instrument Engineers’ Handbook addresses analytical instrumentation in the volume on Process Measurement and Control, edited by Liptak (A6). This compendium of 63 chapter segments with multiple authors presents both laboratory and on-line approaches to a long list of specific analysis needs, including air quality and environmental needs in addition to process control and safety. All the broadly used analyzer methods are included with varying level of detail. Probe-type analyzers receive a specific focus because of their simplified sample systems. Process Analytical Chemistry, edited by McLennan and Kowalski (A7), is a highly useful book for both the practitioners of process analytical chemistry and students in the curriculum. The text starts with the terminology and value of on-line analysis as well as support issues and mechanisms to stay technically current. The following chapters, by multiple authors, address sampling systems, physical property analyzers, process chromatography, molecular spectroscopy, and electrochemical methods. Typically each topic is covered with fundamental principles, instrumentation, and process applications. Two topics not routinely covered are complete chapters on flow injection analysis and process chemometrics. The latter deals with both calibration methods and process modeling. This book also looks to the future with a chapter on various sensor systems and sensor arrays. Kowalski (A8) reviews the growing recognition on the value of real-time process analysis and the implications this has in academic research and the chemistry and chemical engineering education process. A number of the global research consortiums focused in this area are reviewed by Ryan and Frey (A9). Special focus is given to the Center for Process Analytical Chemistry, University of Washington, with discussion of the center operating structure with the industrial sponsors and research success stories. Specific Reviews. (a) Applications and Industries. Realtime process analysis as applied to polymer processing is reviewed by Dumounlin et al. (A10). On-line rheometry is well established and several commercial options are discussed. Optical spectroscopy (IR, NIR, Raman) applications are growing rapidly. Other techniques such as ultrasonics, optical imaging, and NMR offer promise but require further development. Hool et al. (A11) review current practiced technology in gas-phase sensing in the chemical industry. At this time, more conventional analysis methods, such as gas chromatography, mass spectrometry, and optical spectroscopy, are preferred over the emerging area of specific gas “sensors”. Examples of several enhanced sampling systems that increase the utility of conventional analysis methods are presented. Several applications for gas “sensors” and sensor arrays are included. An advanced testing line for nuclear waste streams is reviewed by Day (A12). On-line analysis utilizing NIR, XRF, and a high acid sensor reveals the required process parameters. (b) Analytical Technique. Yalvac and Bredeweg (A13) review the current status and trends in flow injection analysis (FIA). While the versatility of the technique would suggest broad-based use, the maintenance issue of reagent preparation and consumption has limited applicability. Miniaturization of the FIA unit offers a viable solution. Christian (A14) reviews sequential injection analysis, which offers greater simplicity than standard FIA. A

series of laboratory as well as on-line applications and techniques are discussed. Kotiaho completes an extensive review (A15) of both the operational principles and key applications of on-site environmental and process analysis via mass spectrometry. The capabilities of various sample induction methods, analyzer types and ionization methods, portability, and data handling are addressed. A list of environmental and process analysis applications is given with more information on specific applications. Future prospects are good for increased sensitivity, lower costs, and smaller sizes. Walsh and LaPack (A16) also review mass spectrometry principles but focus heavily on the comparative advantages with respect to IR and GC in speed, maintenance dynamic range, concentration, precision, etc. A brief series of applications are discussed. Scrivens (A17) reviews mass spectrometry hardware, typical process applications, and advantages and disadvantages in on-line use. Workman reviews briefly (A18) and extensively (A19) process NIR from 1980 through 1994. The advantages/disadvantages of these five NIR process analyzer technologies are presented together with the broad range of applications. The latter are categorized by specific measuring characteristic of the analyzer. Doyle (A20) compares and contrasts the fundamentals of NIR and midrange IR with the conclusion that they represent a continuum of analytical possibilities. Spectral characteristics of the material often determine the choice of fundamental or overtone/combination vibrational frequency region. Peters (A21) also reviews the topic with an emphasis on several representative applications. Adar et al. (A22) briefly present components of a Raman process analyzer and advantages, with application examples, of the techniques. Beebe et al. (A23) completed a practical guide to chemometrics for the design, analysis, and interpretation of experimental data. This text is useful for the novice but also serves as a reference for appropriate selection of technique for a given situation. Schonkopf (A24) presents an overview of the various empirical modeling techniques. Traditional regression models are contrasted to the bilinear projection methods (principal component analysis (PCA), principal component regression (PCR), partial least squares (PLS)) with an application example. Workman et al. (A25) review the concept and design of an expert calibration system for process analytical spectroscopic methods to select the most appropriate method without an in-depth understanding of chemometrics. SAMPLING SYSTEMS Virtually every process analysis technique requires a sampling system and obtaining a truly representative sample is not a trivial matter. Appropriate and reliable sample transport and conditioning are key elements to obtaining good analytical results. Even those analytical techniques that observe the process sample in an unmodified state (in situ and open path) often require monitoring the sampling conditions and compensating the analytical results for variations in those conditions. A majority of process analytical problems can be attributed to problems in the sample system, and consequently, a lot of effort has been going into improving the state of sampling system technology and design. These are exciting times in the history of process analytical sampling technology. Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


Over the past few years many technology advancements have been applied to the science of on-line sampling. Design aids, modular designs, proving tests, and sample validation techniques; new materials of construction; fiber-optics probes; membrane technologies; improved valves and filters; smarter analyzers that can assist the analyst by monitoring, controlling, and/or compensating for aspects of the sample system that they are interfaced to; and more have been added to the sampling tool box of the process analyzer practitioner. A few books have been totally dedicated to the subject of online process analysis sampling. Though a little dated, the classic Houser (B1) is probably still one of the best theoretical and practical books on on-line sampling. Within the time frame covered by this review, a couple of new sampling books have been published. Carr-Brion and Clarke (B2) is a very fine revision (new edition) that is dedicated to sampling for process analysis. It contains an extensive treatment on the sampling of heterogeneous materials. Sherman (B3) is being written at this time and is treating the sample system as a part of the total process analytical system, with a special focus on component selection, new materials of construction (various polymers, alloys, ceramics, etc.), and calibration/validation of the total analysis system. Most process analysis books include a section on sampling. Those books that are focusing only on specific process analytical techniques, usually have a sampling discussion that is particularly focused on issues that are most relevant to that given technique. Please refer to references cited under the particular analytical technique you are using for additional and somewhat specific sampling information. The references/chapters cited here have very good general sampling sections. Gunnell (B4) appears to be oriented toward enlightening the lab analytical chemist about the many problems, concerns, and solutions that are relevant to online process sampling. Liptak and Liu (B5) take more of an engineer’s perspective and treat the sample conditioning system as a small chemical/mechanical process. Sherman (B6) is written to address the concerns of the practitioner (plant chemist and maintenance technician). Converse (B7) appears to be communicating more toward the laboratory chemist and maintenance technician and is quite extensive, considering that it is only a section and not a whole sampling book. It includes treatments of some of the newer sampling approaches (like remote discrete sampling) that are primarily panel and paper topics and are just beginning to appear in books and chapters. The IEC SC65D standards committee (Analyzing EquipmentsIndustrial Process Measurement and Control) has produced a large technical report (B8) that includes a significant amount of guidance and assistance for some one designing a process analysis sample system. Though primarily written to help the less experienced, it can also serve as a reminder for the experienced process analyzer engineer. The purpose of a sample system is to present an appropriate and representative sample to a process analyzer. Essentially, this amounts to making the sample compatible with the analytical technique that is being employed and representative of the material being sampled. Historically, sample transport and sample conditioning have involved some of the most complex and least reliable aspects of a process analysis system. Because of this, they have been receiving a lot of attention. Improved component 124R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

designs, new materials of construction, modular designs, standard designs, remote discrete sampling, and in situ sampling have all contributed to the advancement of the art/science of sampling. The importance and need for advancements in on-line analysis and sampling was recently emphasized in the chemical measurements section of the U.S. chemical industries vision for the future (B9). Clevett (B10) estimates that 80-90% of all problems with process analyzer systems concern sample handling and calls for improved systems designs, improved components, and more in situ systems to help improve process analysis performance. Podkulski (B11) is disappointed with the fact that in most cases we are applying 30-year-old sampling technology to new hightech process analyzers and called for large improvements in sampling technology. He would like to see more smart sampling systems, more reliable systems designs, and new and more reliable components. Too many sampling systems rely on local indicators and a visit by the friendly maintenance man to tell if they are working. They should be fitted with transmitters that report/alarm remotely to operating and maintenance personnel. Safety is an issue that must be considered for all systems handling hazardous materials. Greenham and Bird (B12) expressed concerns about the potential toxic and flammability hazards that are associated with many gas-sampling systems. Barben (B13) addressed the selection and operation of gas sample conditioning equipment for use in hazardous areas. Fiber optics are frequently being incorporated in spectroscopy sampling approaches. Dubaniewicz et al. (B14, B15) have reported the results of some methane in air and H2 in air ignition tests with fiber-optic-powered instruments. Mobley (B16) discusses efforts to develop a safety standard for the use of fiber optics in hazardous areas. For years it was assumed that fiber optics were intrinsically safe (as far as ignition hazards were concerned). With the increased power of laser-based sources and a better understanding of the technology, it is now recognized that these systems offer hazards beyond eye safety and the traditional concerns that existed when they were primarily confined to laboratory environments. Technological advances in optical sample probes and cells as well as general sampling technology have enabled spectroscopic process analytical measurements to be applied to samples that were previously outside the reach of these methods. Doyle and Jennings (B17) do a good job of discussing several IR and NIR cell and probe concepts, including transmission cells, attenuated total reflectance (ATR) cells, transmission probes, ATR probes, multipath and folded path concepts, the use of light guides, and more. Hansen and Khettry (B18) are working with an NIR probe design that was used to measure the composition of molten polymers. Advancements in several on-line sampling techniques have generated renewed interest and capabilities. On-line membrane extraction microtrap sampling was used by Mitra et al. (B19) to assist in the monitoring of organic pollutants in air and water. On-line headspace sampling was discussed by Pevoto et al. (B20), Schieck and Brown (B21), and Creasy and Capuano (B22). Villalobos (B23) developed an approach to predict the performance of remote discrete sampling, and Converse (B24) recommended using it and on-line GC to simplify CEMS sampling. Wilde (B25) provides a good overview of the various sampling ap-

proaches that are typically applied to CEMS (wet and dry extractive, in situ and ex situ dilution, and in situ sensors). On-line calibration used to be limited to simply switching analyzer sample system to a calibration stream that contained a known sample and adjusting the analyzer output to provide the desired result. Today we have a much larger on-line calibration tool box at our disposal. Converse (B26) discusses several methods of introducing internal standards into on-line analyzer systems. McKinley (B27) proposes using permeation tubes for the calibration and validation of process analyzers for lowconcentration components. Workman et al. (B28) discuss the notion of the expert calibration system, the calibration system that utilizes several known samples and a mathematical model instead of a traditional calibration sample. Some of the open-path optically based analytical techniques can be viewed as being either sample systemless or as employing the uncontrolled open atmosphere as the sample system. For the purposes of this article, we will view it as the later. These systems do have sampling issues and problems. Stray radiation, dew, fog, snow, rain, steam leaks, wind, temperature, and other atmospheric and environmental issues can cause measurement problems and can be viewed as influencing the sample presentation to the analyzer. This technique often requires compensations for variations in these uncontrolled conditions. It is worth noting that this sampling approach even affects the way the data can be interpreted and the results are normally reported (path averaged concentration, e.g., ppm‚meters). Open-path sampling approaches include passive (measuring the radiant energy) and active (measuring the absorbed energy), and the active systems can be further subdivided into bistatic (single pass) and monostatic (multipasssusually two). I was not able to find any articles that were primarily dedicated to these types of sampling, but I will mention a few where the different types of open-path sampling were discussed and utilized. Chaffin (B29) compared the SO2 results of a passive open-path Fourier transform infrared (FT-IR) system and an instack CEMS system. Craig (B30) validated the use of an active monostatic open-path filter IR system for measuring the vehicle emissions of passing vehicles on a highway. Roczko (B31) discussed the use of bistatic open-path IR and UV systems to monitor for LEL levels of combustible hydrocarbons and ppm levels of toxic compounds. For those of us who still have to work with extractive sample systems, there is good news. Design aids, new concepts, and new and improved components are being produced. We seem to be addressing the situation where problems the $5000 sample system have made the results generated by a $200 000 analyzer unusable/ undependable. Small and Saltzman (B32) describe a knowledgebased computer program to aid in the design of extractive sample systems. In discussing how the program helps the user design a system, they also help educate the practitioner with regard to several sampling issues and solutions. Strawn (B33) discusses a sampling system proving test that can be used to verify that one is getting a representative sample from an extractive sampling system. Mayeaux (B34) proposes an innovative change to modular component designs and sample system construction techniques. This would be an extension of the highly successful Semiconductor Electronics Manufacturers Institute gas sample system standardization effort. He believes a broader but similar concept can

be applied to more general process analysis sample systems designs. Gerhab and Schuyler (B35) propose inserting sample pathways with a thin layer of fused-silica-like material bonded to the metal surface of tubing, valves, fittings, etc., to prevent the adsorption and breakdown of highly active components in a sample system. In addition to the more general conceptual and componentrelated related sampling technology advances previously reviewed, there are several more specific sampling component design and utilization advances that are worth reviewing. Munn (B36) examined optimum design requirements for flue gas sample coolers and also discusses alternate sample-drying techniques, such as membrane dryers. Verhappen (B37) described how to design a sample sparger. Peterson (B38) discussed advances in liquid fixed-volume sampling technology and a new multistream ambient air sample valve that can replace a panel of solenoid valves and tubing. Mayeaux (39) discussed the application of membrane filters to remove entrained aerosol liquid droplets from gas samples and techniques to prevent condensation from occurring after the initial separation was completed. Fries (B40) reviewed the basic theory and design of pressure regulators and provided some guidelines/criteria that could be used to help the analyst select an appropriate regulator for a given application. Gavin and Valentine (B41) discussed configuration and performance improvements in a line of coalescing and particle filters that were designed to help meet the needs of process sample systems. McDaniel et al. (B42) described several advantages and precautions for the use of Titan series valves on analyzer sample system applications in oil refineries and chemical plants. Saltzman (B43) discusses several valve design modifications and new valve designs that have been implimented to help improve overall sample system performance and cost. The previously reviewed sampling techniques, designs, and components are generic to many industries, processes, and types of process analytical measurements. Recent advances in bioprocessing and bioprocess monitoring have created some special needs and techniques to meet those needs. Van de Merbel et al. (B44) review several on-line bioprocess sampling systems and techniques, including ultrafiltration, dialysis filtration, dialysis probes, and nonmembrane dialysis filtration. Freitag (B45) also discussed a large number of sampling modules and components for assistance in monitoring typical boitechnological processes by a variety of analytical procedures. In addition to assisting with the measurement process, important system/component issues such as sterilizability and special materials of construction were discussed. Bioprocessing technologists have their own language and have developed terms to describe sampling components that may seem strange to those who are not from their industry. CHROMATOGRAPHY Chromatography continues to enjoy wide usage in the field of process analysis. Although gas chromatography is the most mature of the multicomponent process analyzers and dominates the field of separation techniques used for monitoring environmental and production streams, other speciating methodologies are becoming more widely practiced. The development of more rugged and reliable sampling equipment is largely responsible for this diversification. Ralf and Janusz (C1) described new devices Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


based on solid-phase extraction techniques, and Knutsson et al. (C2) reported the use of supported organic liquid membranes for aqueous sampling. Ashraf-Khorassani et al. (C3) reviewed supercritical fluid extraction (SFE) for sample introduction to liquid and ion (IC) chromatographies. An extensive review on sampling and analytical strategies for on-line bioprocess monitoring was produced by van de Merbel and co-workers (C4). Lewis et al. (C5) described multidimensional methods for sampling and analysis of polycyclic aromatic compounds (PAHs) in different matrixes. The need for more information and faster response from process analyzers has led to work on chromatographic data treatment. Shao and co-workers (C6) obtained quantitative information from overlapping chromatographic peaks using wavelet transform. Mizrotsky et al. (C7) demonstrated that system peaks could be used to continuously monitor streams with complex matrixes such as wastewaters and physiological fluids. Lavine et al. (C8) used pattern recognition to identify the source of underground fuel spills from high-speed gas chromatograms. Beauford (C9) developed an algorithm for dynamic correction of distillation analyzers that essentially eliminates the inherent dead time associated with chromatographic analyses. Gas Chromatography. The field of process gas chromatography itself is prohibitively large to be defined comprehensibly by a single review article. DeBoer and Kenter (C10) reviewed the statistical process control methods for on-line chromatographic analyzers. Akard and Sacks (C11) explored the optimization of multicolumn, high-speed GC using window diagrams. Twodimensional, comprehensive, high-speed, gas chromatographic analyses were reported by the Synovec group (C12, C13). In this method, a polar second column performs separations on portions of the effluent from a nonpolar first column with a periodicity on the order of seconds. Fang and Wang (C14) introduced an online method for calibrating process gas chromatographs. (a) Sampling. Unless the portion of the process stream to be analyzed is pristine, sample capture and introduction to the process monitor is a critical step for proper component identification. Goosens et al. (C15) reviewed the field of sample pretreatment for capillary GC, while Brinkman and co-workers (C16) focused their review on solid-phase extraction (SPE) for capillary GC. Luque de Castro and Fernandez-Romero (C17) examined the work performed using continuous separation techniques (SPE and pervaporation). Many process and environmental samples are liquid; much of the recent gas chromatography sampling literature describes means of mitigating the effects of solvents to determine minor and trace components in process samples. Liu and Dasgupta (C18) described using a liquid droplet as a renewable gas-sampling interface. In this technique, gas-soluble components transition from the liquid drop to a flowing gas stream, where a sample loop is used to introduce the analytes on column. Collier and Thompson (C19) designed a concentrator apparatus for detecting trace organic components in aqueous samples. Guo and Mitra (C20) developed a pulsed introduction membrane extraction (PIME) method for similar applications. The PIME technique yielded subppb detection limits and high precision under controlled conditions. Grob and Munari (C21) developed a device for separating solvents from the analytes of interest, effectively concentrating the sample before injection. Panda et al. (C22) describe a closed126R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

loop vaporization chamber with a precisely timed, gas-sampling valve to introduce cleaner samples to the GC. Headspace sampling techniques may also be used for liquid process streams. In this technique, a liquid is typically trapped and heated, after a period of time the gas phase above the liquid is introduced using a gas-sampling valve. Creasy and Capuano (C23) reported using a stopped-flow headspace device for monitoring volatile species in organic and aqueous process streams. Yang and Pawliszyn (C24) combined the headspace technique with membrane extraction to yield clean sample introduction into GC columns. Chai and Zhu (C25) performed vapor/liquid phase studies to more fully explain the headspace sampling phenomenon. Gas sampling for GC analysis typically requires a concentration step. Castello et al. (C26) describe an automated sampling and analysis system for monitoring volatile organic compounds in air. Feng and Mitra (C27) developed a two-stage microtrap injection device for continuous, on-line monitoring. After sample collection, the first, larger trap is desorbed and the analytes are refocused on the second trap, which, when desorbed, provides a sharp band injection for GC separations. An in-line separator to remove condensable hydrocarbons from natural gas was reported by Ting (C28); the device protected the GC system without changing observed values significantly. Rapid sample preparation of complex solid matrixes by SFE for introduction into portable and automated GC equipment is gaining popularity. Bowadt et al. (C29) report using SFE with GC analysis to rapidly sample and measure polychlorinated biphenyls and polynuclear aromatic hydrocarbons in soil. Francis and coworkers (C30) reported analysis of nitro and nitroso compounds in soil using an SFE-GC system with a thermal desorption modulator interface. The total analysis was completed in less than 10 min and achieved a detection limit of 2.6 ppb for one compound. (b) Applications. David and Pauls (C31) described the comparison of sampling and analysis methods for airborne aromatic hydrocarbons in petrochemical plants. Clemons (C32) described a new GC method to measure paraffins, naphthalenes, and aromatics, which greatly reduces analysis times. The separation technique relies on molecular size and polarity differences instead of the classical boiling point methodology. Farber (C33) developed GC techniques for on-line testing of transformer faults; using headspace sampling, Farber was able to calculate oildissolved concentrations of fault gases. Ryan and co-workers (C34) described the development of a continuous emission monitor for combustion processes. Hou et al. (C35) studied outgassing of electronic wafers within different containers. The apparatus comprised thermal desorption equipment, a GC, and either a mass spectrometer or a nitrogen-phosphorus detector. Ballesteros and co-workers (C36) described a GC method whereby cholesterol and tocopherols in edible oils and fats are determined. Automatic removal of interfering triglycerides is achieved by postinjection derivatization. Giacometti et al. (C37) monitored the esterification of sorbitol and fatty acids by GC determination of lauric acid concentration in the process. Ma et al. (C38) reported the on-line monitoring of the catalytic conversion of methyl formate to methanol and carbon monoxide by GC. This high-temperature process was followed by using an isothermal method with a six-way gas-sampling valve. Fernandes

and co-workers (C39) similarly reported monitoring the catalytic degradation of polyethylene using a GC. The GC results mirrored the thermal gravimetric analysis data. Snavely and Subramaniam (C40) analyzed Fischer-Tropsch synthesis products formed in a supercritical fluid reaction medium using an on-line GC. (c) Hardware. Liu and co-workers (C41) described a GC designed for gas analysis at subatmospheric pressures. Melda (C42) reported stable and precise pressure control for process GC. This feature allows for pressure programming effects of component elution. Smith and Sacks (C43) described pressuretunable GC columns with electronic pressure control. This fast GC method demonstrated greatly improved tuning resolution and repeatability compared to previous reports. Strunk and Bechtold (C44) developed a two-dimensional interface for capillary GC, which resulted in column-switching efficiencies with minimal separation disturbance. Quimby et al. (C45) reported using a GC equipped with an atomic emission detector for improved measurement of nitrogen and sulfur compounds in refinery liquids. (d) Future Technology. Henry (C46) reviewed portable GC instruments. Overton and Carney (C47) reported a new portable GC and sampling system that operates with a minimal consumption of utilities. Payne and co-workers (C48) developed a probe chromatograph, capable of being installed in the source pipeline. Sittler et al. (C49) produced a modular GC; based on careful design of manifolds and ports, all valves, and detectors and pluggably engagable. GC miniaturization continues to proceed. NASA researchers (C50) reported a silicon-micromachined GC system for determination of planetary surface composition. Loux et al. (C51) designed a miniature GC with heated inlet, tubing, and microvalve assemble for more robust analyses. Nowak and co-workers (C52) developed a gas microvalve inlet system capable of millisecond-time scale operation. Tuan and co-workers (C53, C54) describe preconcentration techniques for portable micro GC systems. This system was demonstrated to have enrichment factors as high as 100 for on-line coupling to high-speed methods. Liquid Chromatography. Difficulties in sample capture and introduction, along with waste disposal, have inhibited the development of process LC. Accordingly, much of the recent work in this field has been focused on solving these problems. (a) Sampling. There has been much work reported using short guard columns to clean process and environmental samples while preconcentrating analytes before injection onto a larger column. Pocurull and co-workers (C55) reported the determination of trace phenols from environmental waters using this technique. Harvey and Clauss (C56) using a similar system were able to produce analyses of trace-level munition identifiers in a matter of minutes. Hogenboom et al. (C57) used a single short column to rapidly analyze organic microcontaminants in real environmental water samples. This technique employed a number of solvent flushes to obtain enrichment prior to analysis. Lacorte and coworkers (C58) reported a fully automated, solid-phase extraction device for sample capture and preparation prior to LC analysis. A new, simple in-line filtering device was also employed in this work. Verette et al. (C59) described an on-line, dialysis-LC system for food and beverage monitoring. This completely automated system was used to analyze sugar and organic acids with minimal interference. Ortiz-Boyer and co-workers (C60) reported a con-

tinuous cleanup and preconcentration system for monitoring metabolites in plasma. The method offers a simpler alternative to solid-phase extraction. (b) Applications. As they do in the laboratory, biological investigations dominate the process LC literature. Ge et al. (C61) described an on-line LC monitor for following a deprotecting process as part of the synthesis of a protease inhibitor. These researchers experienced much better control of the batch process as a result of the analyzer implementation. Thomas and co-workers (C62) presented a series of in-process LC methods for the synthesis of dorzolamide hydrochloride with similar results. Liden et al. (C63) followed a fermentation process using microdialysis sampling, LC, and amperometric biosensors as detectors. This instrument system accurately tracked monosaccharide and ethanol levels during the process. Musolino and co-workers (C64) used LC to determine the intermediates formed during the catalytic hydrogenation of 2,4-dinitrotoluene. These isomers have historically been difficult to identify and quantify. (c) Future Technology. Strancar et al. (C65) described fast separations of biopolymers using compact porous disks. This group demonstrated in-process monitoring of several biological processes. Prazen and co-workers (C66) discussed the standardization of second-order LC-spectroscopic data toward optimizing chemical analyses. The methodology described has a large effect on reducing errors associated with poor chromatographic resolution of hyphenated techniques. Kanazawa et al. (C67) described temperature-responsive LC using polymer-modified silica. In this system, the tradeoff between peak resolution and elution time is optimized. Foster and Synovec (C68) demonstrated reversedphase LC of hydrocarbons using water as the mobile phase. This led to relatively fast chromatography with minimal waste considerations. Other Chromatographic Techniques. Because of the similarity of hardware, as developments are made for process LC, other separation methods will also become more popular. IC and supercritical fluid chromatography (SFC) in particular share commonalities with LC. Other techniques will likely be driven by niche applications, solving very specific process analytical problems. (a) Ion Chromatography. The use of IC for the determination of actinides and fission products in nuclear applications was reviewed by Betti (C69). Montgomery et al. (C70) described a number of automated sample preparation techniques for IC. Rey and co-workers (C71) used column-switching techniques to obtain difficult cation separations. Cavotta et al. (C72) monitored ion residue in extraction tanks for circuit board production using a process IC. Laikhtman and co-workers (C73) determined cation impurities in 30% brine using IC equipped with on-line matrix elimination. Simon and Dasgupta (C74) used a continuous automated IC system to measure atmospheric particulate matter. Boring et al. (C75) developed a field-portable capillary IC. Nonomura and Hobo (C76) designed an automated IC system for the determination of sulfur oxides, nitrogen oxides, and hydrogen chloride in flue gases. (b) Supercritical Fluid Chromatography. Richter and coworkers (C77) described a SFC instrument optimized for the analysis of petroleum fractions. This instrument performed the analyses rapidly with very high precision. Shen and Lee (C78) Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


described a high-speed, solvating GC using packed capillary columns. At the inlet and through much of the column, the mobile phase was a supercritical fluid, while at the outlet it was a gas. This technique resulted in fast separations with high column efficiencies. Adams et al. (C79) built and tested a portable thermal pump for supercritical fluid formation and delivery. The simple and inexpensive pump can produce continuous fluid delivery without refilling. (c) Miscellaneous Techniques. Buscher and co-workers (C80) developed an electrodialysis device for sample cleanup and enrichment prior to capillary electrophoresis analysis. The threecompartment device was effective in removing complex matrix effects. Arce et al. (C81) described an automated flow injection (FI) interface for CE. The net effect was to provide sample pretreatment and preconcentration in the FI system prior to separation. Sandstrom and co-workers (C82) detailed applications of thin-layer chromatography to process control in the pulp and paper industry. Valuable process information was obtained from the separation and quantitation of fatty acids, resin acids, and lignins. NEAR-INFRARED AND INFRARED SPECTROSCOPY AND IMAGING A proliferation of work involving near-infrared and infrared spectroscopy in process and image analysis has rapidly occurred over the past few years. Today more papers are being written about the application of the near-infrared spectral region to all types of analyses than ever before. This critical review includes the aspects of near-infrared and infrared spectroscopic measurements for the analysis of materials categorized into distinct applications areas. These application categories include (in alphabetical order) the following: biotechnology, earth sciences/ atmospheric science/mineralogy, environmental monitoring, fine chemicals/chemical production, food and beverages, medical/ clinical chemistry, petroleum/natural gas/fuel research, pharmaceutical production, polymer science, and surface analysis. The basic interest and growth in two-dimensional imaging spectroscopy, including airborne imaging, hyperspectral imaging, remote monitoring, microspectroscopic imaging, and basic applications of standard spectroscopic techniques modified to produce a spectral image deserves special recognition and description within this review. Thus, the early section of this review will be devoted to imaging. The second portion will present an applications perspective. This review includes a brief look at the recent scientific literature directly associated with imaging spectroscopy using the near-infrared and infrared spectral regions. Key words used for this literature search included NIR, FT-NIR, IR, and FTIR. These measurement technique search words were used in combination with image(s), imaging, and mapping terms. The papers included here represent some of the most recent published work on the subject. Imaging Applications. A descriptive article surveying the field of fast FT-IR imaging including theory and applications is presented by Koenig and Snively (D1). This basic review describes instrumental aspects, imaging methods, data processing, and applications with 20 references. An Applied Spectroscopy focal point article describing infrared spectroscopic imaging in terms of focal plane arrays, remote sensing and astronomy, and terrestrial and medical applications is described in detail with 35 references by 128R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

Colarusso et al. (D2). Workman (D3) described the use of nearinfrared and infrared spectroscopy for process analysis in a review paper with 483 references. van den Broek and co-workers have completed work on the identification of plastics by remote sensing spectroscopic NIR imaging using kernel partial least squares (KPLS) (D4). The authors describe an identification system for sorting plastics from nonplastics in waste recycling. Partial least squares is used for data reduction in the classification of NIR spectral images. Given multidimensional NIR images, PLS projects the high-dimensional space into a low-dimensional latent space using the coded class information of the sample objects. PLS can be considered as a supervised latent variable analysis, and data reduction by PLS increases the speed of on-line classification useful for process control. To apply this method in near real time, a rapid PLS version, KPLS, was investigated. The method was successfully demonstrated for discrimination between plastics, nonplastics, and image backgrounds. Wienke et al. have presented and published work using nearinfrared imaging spectroscopy (NIRIS) and image rank analysis for remote identification of plastics in mixed waste (D5). This paper describes a method for the discrimination of plastics from nonplastics in household waste on the basis of a set of images taken at six wavelength ranges between 1100 and 2500 nm. The authors explained the use of multivariate image rank analysis (MIRA) and reported that the use of this technique provided correct classification in 80% or better of the test cases. The NIRIS experiments were performed using a traditional NIR source, with an InSb focal plane diode-array camera. The wavelength selection was performed with a filter wheel equipped with narrow-bandpass filters at 1600, 1700, 1562, and 2200-2300 nm, as well as broad-band-pass filters at 1700-2150 and 2115-2550 nm. Samples of five different plastics and cotton, glass, paper, wood, metal, and ceramics were examined at distances from up to 2 m. The technique of MIRA is reported for 51 samples in this study. In a third publication, the group from Catholic University, Nijmegen, The Netherlands, published work on the identification of plastics among nonplastics in mixed waste by remote sensing near-infrared imaging spectroscopy using multivariate image rank analysis for rapid classification (D6). Household waste was characterized by a sequence of images taken in four wavelength regions using NIR imaging spectrometry. Each sample was represented by a threedimensional stack of NIR images. A rapid data compression method, followed by an abstract factor rotation of the stack into an intermediate four-element vector, was accomplished using MIRA. These data reduction provided a single number used as a decision limit for the classification of the plastics. The method is insensitive to sample size and relative position within the camera cone of focus. The method was useful even with slight sample movement. Ning et al. published five novel applications of imaging visible and short near- infrared spectrophotometry and fluorometry in the plant sciences (D7). Noninvasive, in vivo applications were performed using a charge-coupled device (CCD) instrument for imaging spectrophotometry and fluorometry (D8). Three potential applications described include the following: (1) in vivo effects of a fungal pathogen; (2) following changes in water in vivo; and (3) estimation of quantum yield of photosynthetic fluorescence in variegated leaves. Sowa et al. completed work on noninvasive

assessment of regional and temporal variations in tissue oxygenation by near-infrared spectroscopy with imaging (D9). The work shows how the techniques can be used to map regional variations in the tissue oxygen saturation in the human forearm under conditions of interrupted or restricted blood flow. The data processing involves both multivariate analysis of temporal imaging and spectral data and fuzzy C-means clustering of temporal imaging data. Villringer and Chance published work on noninvasive optical spectroscopy and imaging of the human brain function (D10). Crowley and Zimbelman completed work for mapping hydrothermally altered rocks on Mount Rainier, WA, using airborne visible/infrared imaging spectrometer (AVIRIS) data (D11). Makipaa et al. have described an infrared-based imaging method for copper electrolysis short circuit detection (D12). Rowlands and Neville published work on calcite and dolomite discrimination using airborne SWIR imaging spectrometer data (D13). Otten measured performance of an airborne Fourier transform hyperspectral imager (D14). Valdez et al. reported on the selection of spectral bands for interpretation of hyperspectral remotely sensed images (D15). Silk and Schildkraut completed work describing imaging Fourier transform spectroscopy for remote chemical sensing (D16). Rowan et al. completed analysis of AVIRIS data from a carbonatite-alkalic igneous rock complex (D17). Mapping technologies provide a detail of data analysis where individual spatial images are combined across a broad two- or three-dimensional landscape to produce an actual map of the spectrally resolved data across a spatially resolved surface. Images are a part of the map, but images alone do not comprise a map until they are directly related to the spatial dimension. Heekeren et al. completed work on noninvasive optical human brain mapping reporting improvements of the spectral, temporal, and spatial resolution of near-infrared spectroscopic imaging (D18). Kruse reported on methods for geologic mapping using combined analysis of AVIRIS and SIR-C/X-SAR data (D19). Watanabe et al. described a technique for noninvasive functional mapping with multichannel near-infrared spectroscopic topography in humans (D20). Biotechnology Applications. The use of near-infrared measurement for process analysis in fermentation processes is discussed in work by Brookes et al. (D21). Near-infrared spectroscopy has shown promise in identification of various strains of yeast in Halasz et al. (D22). Zakim and Diem have shown that spectral cytometry has value for the rapid analysis of cells (D23). Macaloney et al. have reported that near-infrared spectroscopy has shown itself to be valuable for the simultaneous measurement of several constituents essential for production control of a highcell-density recombinant Escherichia coli process (D24). Nearinfrared spectroscopy is proposed as a method for estimating acetate, biomass, glycerol, and ammonium in an unmodified whole broth by Hall et al. (D25). The method is reported to measure and compute the concentrations of these components in 1 min (i.e., nearly in real time). The method could potentially be used on-line for process monitoring and control. Diffuse reflectance infrared Fourier transform spectroscopy (DRIFTS) and artificial neural networks have been reportedly combined by Goodacre et al. for the rapid identification of bacterial species (D26). NIR spectroscopy has been shown by Macaloney et al. to be valuable in control and fault analysis during high-cell-density E. coli

fermentation (D27). A general discussion by Dosi et al. of the use of on-line near-infrared analysis for control of fermentation processes is given (D28). FT-IR is used by Szalontai et al. to investigate fatty acyl chains in model and biological membrane structures (D29). The air/water interface has been studied using polarization modulated Fourier transform infrared spectroscopy by Cornut et al. in order to understand the basic structure and orientation of lipids and amphipathic peptides (D30). This work is undertaken to gain a more fundamental understanding of the structure of biological membranes. Raman spectroscopy has been used by Berger et al. to determine the concentration of biological materials in aqueous solutions (D31). FT-IR spectroscopy has been used by Henderson et al. to identify Candida at the species level using various spectroscopic discrimination techniques (D32). Polarization modulation has been described by Fournier et al. for in situ measurements of monolayer films in the study of biological membranes at the air/water interface (D33). Lipid monolayers have been studied by Axelsen et al. in situ using internal reflectance FT-IR (D34). NIR has been used by Carlsson et al. for quality control testing of hyaluronan as an alternative to biological testing (D35). FT-IR spectroscopy has been used to study biological membranes on solid surfaces as well as at the air/water interface by Pezolet an co-workers (D36). Earth Sciences and Mineralogy. A review by Aldstadt and Martin containing 64 references describes the use of a cone penetrometer for in situ analysis of contaminants in soil and groundwater (D37). Infrared analysis of coals by Matuszewska is described in ref D38. Speciation of water types in haplogranitic glasses and melts was determined by Nowak and Behrens using in situ near-infrared spectroscopy (D39). Airborne infrared spectroscopy of wildfires in the western United States is described by Worden and co-workers (D40). An FT-IR open-path airmonitoring system was used to monitor volcanic fumarole gases by Chaffin et al. (D41). Of the potential analytes in volcanic plume gases, SO2 and HCl exhibited the highest detectivity. Diode laserbased NIR spectrometry is described as an ideal and inexpensive method for gas monitoring by Martin and Feher (D42). NIR diode lasers for use in a transmission spectrometer are described for the analysis of atmospheric or waste gases. FT-IR spectrometers were used by Watson and co-workers for environmental analyses (D43). Instrumentation was carried on tethered and free-flying manned hot-air balloons. These instruments were configured to include the following : mid- and far-IR cameras, multispectral imaging spectrometers, radiometers, forward-looking IR cameras, and a variety of other sensors. Atmospheric volatile organic compounds have been monitored by Hammaker et al. using remote FT-NIR (D44). The use of far-infrared spectroscopy for studies of terrestrial and planetary or extraterrestrial atmospheres is described by Griffin (D45). Forestry studies conducted in North Carolina by Zwicker detail using the remote-sensing capabilities of open-path FT-IR spectrometry (D46). NIR has been used by Malley et al. for rapid assessment and quantitative measurements of suspended C, N, and P from Precambrian shield lakes (D47). Near-infrared spectroscopy and visible spectroscopy have been used to characterize asteroid composition via surface mineralogy as described in a dissertation by Howell (D48). FT-IR spectroscopy has been used by Yalamanchili et al. to characterize interfacial water species in situ at hydrophobic and Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


hydrophilic surfaces in mineralogy (D49). Remote FT-IR has been mounted on balloon-borne platforms for remote sensing of various analytes in the atmosphere as described by Camy-Peyret et al. (D50). Atmospheric free radicals have been studied in situ by Blake using a mid-infrared magnetic rotation spectrometer (D51). Short-wave infrared spectrometry has been reported for use in airborne prospecting for gas and oil resources by Yang (D52). Infrared microspectroscopy has been shown by Gao et al. to be useful in rapid identification of gemstones (D53). Multispectral instrumentation has been used to record the burning of biomass in West Africa by De A. Franca et al. (D54). Visible and nearinfrared spectroscopy have been used to study the exposure levels and iron concentration of lunar soils by Fischer and Pieters (D55). Infrared and near-infrared continue to play a valuable role in environmental monitoring. Open-path FT-IR spectrometry is reviewed for field use in remote sensing of airborne gas and vapor contaminants by Levine and Russwurm (D56). The review includes 27 references covering open-path FT-IR applications and describing where additional work is required for further development of the technique. The advantages of open-path FT-IR spectroscopy are described by Marshall et al. (D57). These benefits include (1) its versatility; (2) remote, long-path measurement; (3) in situ applications; and (4) near-real-time measurement. FT-IR is combined with column liquid chromatography for the identification of herbicide contamination in samples of 50-100 mL of river water by Somsen et al. (D58). Enough analyte was obtained for good-quality spectra down to water concentrations of 1-2 µg/L. FT-IR emission spectroscopy has been employed by Wang et al. for the study of Freon-12 in an alcohol plus air flame (D59). Infrared spectroscopy has been reported by Haschberger for use as a remote method for monitoring trace gases from aircraft emissions (D60). Environmental monitoring applications using fiber-optic sensing, and FT-IR spectroscopy is described by Druy et al. (D61). Ground-based FT-IR spectroscopy was used for remote sensing of pollution emission sources for application in environmental compliance monitoring by Schaefer (D62). Marshall et al. describe the effects of resolution on the performance of classical least-squares (CLS) spectral interpretation when applied to volatile organic compounds (VOCs) of interest in remote sensing using open-air long-path FT-IR spectrometry (D63). A 1995 patent describes the method and apparatus for remote infrared sensing and analysis of motor vehicle exhaust gases (D64). A real-time background correction algorithm is described for passive FT-IR spectroscopy of chemical plumes for environmental monitoring applications by Polak et al. (D65). Organic wastewater contaminants are measured by Eilert et al. using acoustooptical tunable filter (AOTF)-based NIR spectroscopy (D66). Various trace gases are measured by Toci and co-workers in situ using airborne infrared diode laser spectrometry (D67). Soil contamination is rapidly determined by Adams and Bennett using reflectance Fourier transform infrared spectroscopy (D68). Theriault et al. report on the use of ground-based FT-IR for monitoring weather cloud parameters (D69). Remote FT-IR is used by Davies et al. to remotely monitor motor vehicle exhaust gases (D70). Fine Chemicals and Chemical Production. An on-line NIR method is presented by Norris and Aldridge for determination of the steady-state end point of homogeneous and heterogeneous 130R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

organic reactions for chemical and pharmaceutical production (D71). The method uses periodic NIR monitoring over the progress of the reactions. The steady-state point is detected when the NIR do not change significantly over time. May et al. report the use of an FT-IR spectrophotometer equipped with an 8-m gas cell having solid nickel mirrors. The system is used for the determination of H2O in anhydrous HCl samples. An on-line determination of impurities in gaseous Cl2 using a temporary system is described. In addition, the on-line monitoring of impurities in anhydrous HCl as it passed through a carbon bed using a permanently installed system is described. Conventional methods for the analysis of chlorine and hydrogen chloride are also reviewed (D72). The combination of chemometrics with FTIR spectrometry was used by Lindblom et al. to monitor diphenylamine stabilizer concentrations in the presence of nitrocellulose, the main component of rocket propellants (D73). The shelf life of these propellants is normally determined by HPLC analysis of the diphenylamine stabilizer and its derivatives. FT-IR was used by Bandekar and co-workers to quantitatively measure sulfur oxide in white liquor. Liquid samples of ∼1.5 mL were placed into cylindrical internal reflectance cell for liquid evaluation (Circle) cells comprising an open-boat arrangement with a zinc selenide crystal at 45°. The infrared beam was configured for 12 internal reflections. The method was suitable for quantitative analysis calibration ranges to 70, 45, and 42 g/L for sulfate, thiosulfate, and sulfite species, respectively. Corresponding minimum detection limits were 0.01, 0.02, and 0.02 g/L (D74). IR spectroelectrochemical measurements are performed by Faguy and Marinkovic on single-crystal Pt electrodes for a variety of electrochemical systems. Use of a ZnSe hemispherical window in the dual roles as a lens and as an IR-transparent wall of the electrochemical cell allowed beam collimation, near-critical angle reflection, and low first-surface reflection losses. Derivation of the expected figure of merit for in situ reflection/absorption spectroscopy is presented. Examples are given of adsorption processes from H2SO4 solutions and for the oxidation of glucose on Pt(111) and Pt(100) electrode surfaces (D75). A FT-IR spectroscopic method was developed by Ge et al. using an ATR sample boat with a zinc selenide or zinc sulfide crystal (D76). Methanol/ acetonitrile (1:1) or air was used as the background spectra for the two types of ATR crystals, respectively. The method was used to monitor the epoxidation step in the synthesis of cis-1-amino-2indanol when indene is oxidized to indene oxide. For the ZnSe ATR, a portion of the reaction mixture was analyzed by FT-IR spectroscopy without sample preparation, and for the ZnS ATR, the sample was centrifuged to separate the organic and aqueous layers with FT-IR spectrometry being performed on the organic layer. Results were calibrated and cross-validated by HPLC with details described. Spectral data were collected from 1500 to 900 cm-1 and processed using multiple linear regression (MLR) and PLS. An FT-IR method is described by Ruyken et al., with resultant data analysis techniques presented, for the prediction of organic vapor concentrations (D77). The method described includes techniques for identifying interference present within the vapor mixtures. Basic principles of multivariate calibration using PCR are outlined, and the effect of interfering species on the predicted concentration is discussed. The infrared spectra of organic vapors

at 60 °C were recorded at 4-cm-1 resolution. A simple twocomponent calibration model was constructed for hexane and ethanol, with additions of dichloroethane, acetone, methyl acetate, and toluene as interference. The prediction errors depended on the concentration of the interference and the overlap of each interference’s spectrum onto the analyte regression vector. A residual library search method was used to identify each interference. The FT-IR spectrum of a ternary aqueous solution of chlorate, perchlorate, and dichromate was measured from 900 to 1250 cm-1 by Kargosha et al. (D78). After subtraction of the water spectrum and standard baseline correction, the spectrum was resolved using multicomponent analysis (i.e., partial least-squares regression and multiple linear regression). The infrared bands at 973, 1110, and 950 cm-1 were used to quantify chlorate, perchlorate, and dichromate, respectively. Calibrations were linear from 0.2 to 63.8, from 0 to 73.47, and from 0 to 149 g/L chlorate, perchlorate, and dichromate, respectively. The corresponding detection limits for each of the analytes were 1.064, 0.64, and 0.95 g/L. The relative standard deviation (RSD) as 1 σ for eight samples were from 1.19 to 2.7% relative. The method was applied to electrolytic solutions of chlorate and perchlorate. In situ FT-IR spectroscopy was used by Woelki and Salzer to study the structures of powdered humic acid salts (D79). Samples were prepared by placing the salt samples with KBr in a DRIFTS accessory. The sample reaction chamber was flushed for 15 min with dry N2 and the sample was heated under stopped- and continuous-flow N2 at 10 and 65 °C/min, respectively. The spectra were obtained at 4-cm-1 resolution by coadding 50 scans at slow heating and 200 scans at fast heating. The structural changes caused by heating were recorded with results presented. Remote high-resolution FT-IR was used by Wang et al. for spectral characterization of infrared flare material combustion (D80). The flare materials were pressed into a 30-g pellet and positioned 15 m from a remote-sensing (open-path) FT-IR instrument with a combustion flame. The emission spectra were recorded from 4700 to 740 cm-1 with a resolution of 0.24 cm-1. Peaks corresponding to HCl, HF, H2O, CO2, CO, and SiF4 were obtained, allowing qualitative and quantitative determination of these combustion products. The flame temperature was estimated, and the spectral radiation distribution was reported. Rapid IR spectrophotometric determination of total nitrogen in silicon dioxide/silicon nitride (SiO2/Si3N4) mixtures is presented by Wisniewski and co-workers (D81). Samples of 1-1.5 mg were ground with 200 mg of KBr and pressed into a tablet. The IR spectra were recorded from 400 to 2000 cm-1. The N content was quantitated from the transmittance at 575 and 1450 cm-1 (background) using a formula described. The calibrations were linear over the 20-35% nitrogen levels. The RSD of 1 σ were 7.5-44.5% relative. These results were compared to those obtained using a volumetric method where samples are dissolved in acid and the NH3 is released using distillation. The analysis time was 0.5 h. The method was applicable to mixtures containing lithium and magnesium salts and is suitable for evaluating the yield of Si3N4 synthesis from SiO2 and NH3. In situ temperatureprogrammed diffuse-reflectance infrared Fourier transform spectroscopy (TPDRIFTS) is used to monitor chemical changes in the surface structure of vanadium oxide/titanium oxide catalysts during reactions by Centeno et al. (D82). A general article is

presented by Brimmer describing the development and applications of near-infrared spectroscopy in the chemical and pharmaceutical industries (D83). Fiber-optic-based FT-IR was used by Salim et al. for in situ measurements to monitor the concentration of various reactants in a vertical OMVPE reactor (D84). In situ ATR infrared spectroscopy was used to evaluate and study the various aspects of laminar flow in a channel-type electrochemical cell by Tolmachev et al. (D85). In situ IR reflectance spectroscopy was used by Lamy et al. to study and monitor the electrooxidation of methanol at Pt-Ru electrodes (D86). In situ DRIFTS was used by Benitez et al. (D87) for the study of the reversibility of CdGeON sensors toward oxygen. In situ ATR FT-IR spectroscopy was described in a doctoral dissertation by Dunuwila for the measurement of crystallization phenomena for research and development of batch crystallization processes (D88). In situ infrared spectroscopy was used by Epple et al. to study the solid-state formation reaction of polyglycolide (D89). In situ near-infrared spectroscopic investigation of the kinetics and mechanisms of reactions between phenyl glycidyl ether (PGE) and multifunctional aromatic amines is described by Xu et al. (D90). Remote mid-infrared spectroscopy is used by Mijovic and Andjelic to monitor reactive processing for polymers (D91). Developments of in situ molecular spectroscopy in catalysis studies is given by Xin and Qin (D92). Crossdispersion infrared spectrometry (CDIRS) for remote chemical sensing is described by Stevens et al. (D93). In situ infrared spectroscopy is used to by Eder-Mirth to study the mechanisms of selective alkylation of aromatics over zeolites (D94). Rapid infrared sampling methods are described by Scamehorn for the organic chemistry laboratory using a DRIFTS accessory (D95). In situ FT-IR and FT-Raman studies of reactions in the hydrothermal environment are described by Schoppelrei (D96). The use of near-infrared monitoring of dimethyl terephthalate in a lowpressure methanolysis process is presented by Agyare (D97). In situ infrared studies on dehydroxylation of MCM-41 are described by Li et al. (D98). Infrared absorption spectroscopy is used by Zhang et al. as a method for rapid fault detection in borophosphosilicate glass thin films (D99). In situ FT-IR studies of the acidity and catalytic properties of Pt/zeolites is described by Chien et al. (D100). In situ FT-IR spectroscopy is related to kinetic studies of methanol synthesis from CO/H2 over ZnAl2O4 and CuZnAl2O4 catalysts by Le Peltier et al. (D101). FT-NIR is used by Binette and Buijs for process monitoring of multiple inorganic ions in aqueous solution. A scheme is presented for a near-infrared multipoint monitoring system for a three-stream plant producing soda using the Solvay process (D102). Near-infrared spectroscopy is used by Yeboah et al. for quantitative analysis of resorcinol in aqueous solution for chemical production in the photographic industry (D103). Food and Beverage Applications. Food and beveragerelated industries have traditionally used near-infrared and infrared measurements for quality control, blending, and process control. A presentation of the uses of ATR FT-IR spectroscopy for process control in the sugar industry is given by Veronique and Gilles (D104). A patent is presented for real-time on-line analysis of organic and nonorganic compounds for food, fertilizers, and pharmaceutical products (D105). Dispersive near-infrared spectroscopy was used by van den Berg et al. to analyze 30 alcoholic Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


beverages samples for ethanol (D106). Samples were contained using a 2-mm-thick quartz cuvette, with spectral measurements made over the wavelength range of 1100-2498 nm. The absorbance spectra were preprocessed using a second derivative to remove offset and sloping baseline. The reported results indicated the suitability of the method for ethanol determination in alcoholcontaining beverages. Increasing the variability in the training set by increasing the number of samples improved model performance. Neural networks combined with near-infrared spectroscopy are described by Borggaard and Thorup for on-line quality measurement of unhomogenized meat products (D107). FT-IR combined with a liquid ATR cell heated to 50 °C was used by Mossoba et al. to determine the total trans content of neat hydrogenated edible oils. The sloping background of the infrared trans band at 966 cm-1 was eliminated by taking to ratio of the single-beam spectrum of the sample to the corresponding spectrum of the unhydrogenated material; the trans band could then be integrated between 990 and 945 cm-1. The limits of detection and determination were 0.2 and 1%, respectively (D108). Near-infrared diffuse reflectance spectroscopy was used by Lee et al. for the rapid analysis of curds during a cheddar cheesemaking process (D109). Blended curd samples were packed into rotating 5-g sample cells. The samples were measured using a tilting filter NIR scanner equipped with three tilting filters for scanning from 1900 to 2320 nm. Data analysis was performed using multivariate computer software. Estimated values for moisture, fat, and lactose were calculated. MLR equations were used to compare the results by NIR spectroscopy to classical chemical analysis. NIR spectrometry was reported to be suitable for rapid monitoring of chemical changes occurring in curds during the making of cheddar cheese. An FT-IR method is reported by Hewavitharana and van Brakel for the direct determination of casein in milk. FT-IR spectra were recorded for homogenized milk samples while a sample temperature of 40 °C was maintained (D110). Data acquisition parameters, such as resolution and the types of background spectra, were studied to achieve optimum conditions. Two chemometric methods, viz., PLS and PCR, were used for data processing. The best results were obtained using a resolution of 4 cm-1 with PLS calibration over the spectral regions 3000-2800, 1600-1500, and 1300-1000 cm-1. For comparison, the results obtained agreed closely with those obtained by the International Dairy Federation reference method. Near-infrared spectroscopy is described by Berding and Brotherton as a means to measure heterogeneous sugar cane samples (D111). The typical labor-intensive subsampling necessary to overcome the within-sample heterogeneity inherent in the batch sampling of agricultural products, specifically sugar cane, has prompted the development of an at-line device for the presentation of large (1.5-3-kg) samples to the measuring instrument. An NIR spectrophotometer fitted with a remote reflectance probe was used to demonstrate the analysis of 30 samples/h. Near-infrared transmission (NIT) spectroscopy was applied by Lovasz et al. for the determination of several quality parameters in apples (D112). Apples were analyzed for firmness, RI, pH, titratable acid, dry matter, and alcohol-insoluble solids using NIR transmission spectrometry. Short-wave near-infrared spectra were acquired over the wavelength range of 800-1100 nm. Data were processed using a chemometrics software package 132R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

which included partial least-squares regression (PLSR), multiplicative scatter correction (MSC), automatic outlier diagnostics, and standard cross-validation. The NIT results for each parameter (except titratable acid) were similar to those obtained using the standard reference methods. NIT spectroscopy was used by Wold et al. to determine the average fat content of Atlantic salmon (D113). Salmon were collected, killed, bled, and kept on ice; samples were shaped into cylinders of intact tissue (23 mm thick, 20 mm diameter). NIT spectra were over the wavelength range of 850-1048 nm with the bandwidth set at 6 nm. Samples with skin intact were orientated with the skin side facing the incident light; samples were kept at 4-6 °C throughout the measurement cycle. A good correlation was obtained between average fat content of salmon and transmission spectra measured on intact fish muscle with skin and scales. The method is rapid and nondestructive. NIR spectroscopy was performed by Downey on fresh salmon through the skin to determine its moisture and fat content (D114). Six sample sites were analyzed on the ventral and six on the dorsal side of each salmon. Spectra were recorded on a scanning spectrophotometer with a surface interactance fiber-optic probe and a transmittance detector module at wavelengths from 400 to 1100 nm at 2-nm intervals. The best calibration was produced by a partial least-squares treatment of the first-derivative spectra. The method should be useful for optimizing the oil and moisture content of farmed salmon. A combined visible and short-wave nearinfrared transportable spectrophotometer system was developed by Chen et al. for on-line classification of poultry carcass quality (D115). The spectrophotometer is a bifurcated optical-fiber system for measuring the reflectance of carcasses in the wavelength region of 471-963.7 nm. A diagram of the probe that collects light energy from the skin and muscle tissue of the carcass is presented, and the remaining components comprise a tungsten halogen lamp, a grating spectrograph with a photodiode-array detector, and a PC. A neural network for classification into normal, septicemic, and cadaver carcasses was developed. When this device was used to classify normal and abnormal (septicemic or cadaver) carcasses, an overall accuracy of 97.4% was obtained. A review is presented by Lipp containing 125 references describing the developments of methods for the determination and quantification of triglycerides in milk (D116). The referenced papers appear in the literature during the period 1990-1994. In particular, gas (GC) and high-pressure liquid chromatographic (HPLC) methods are highlighted. Analytical methods such as IR spectroscopy, MS, 12C/13C isotopic analysis, differential scanning calorimetry (DSC), gravimetry, and titrimetry are also discussed. An evaluation of the different approaches to the detection of adulteration of milk fat was also described. A rapid, automated Fourier transform infrared spectroscopic method was developed by van de Voort et al. for the determination of cis and trans content of edible fats and oils (D117). A sample-handling system for use in routine quality control measurement of fats and oils is described. Prior to analysis, samples were warmed to within 5 °C of the operating temperature of the instrument using a microwave oven. The heated samples were aspirated into the IR measurement liquid cell. The spectra were measured at a resolution of 4 cm-1 for 128 scans/spectrum. The liquid cell was emptied and reloaded with the next sequential sample. Spectra were calibrated by PLS

analysis for the cis and trans content with results given. Two infrared spectroscopic methods for cheese analysis are compared by McQueen et al. The methods were NIR filter (i.e., named optothermal spectroscopy; cf., Int. Lab. 1990, 20, 16) and ATR FT-IR spectroscopy (D118). In NIR filter spectroscopy, spectra were recorded using optical transmission filters of 1740, 1935, and 2180 nm. The ATR spectra were recorded using a conventional FT-IR spectrometer at a resolution of 8 cm-1. PLS regression analysis was used to correlate the spectral data to the composition of fat, protein, and moisture determined in 24 cheeses using reference analytical methods. With the NIR filter spectroscopy, correlation coefficients (r) were 0.95-0.97 with the corresponding values obtained by using ATR FT-IR 0.97-0.99. The NIR technique was more rapid and somewhat easier to use than FT-IR. Carbon dioxide in water samples was analyzed using an Analytical Development nondispersive IR CO2 analyzer as reported by Abdullah and Eek. Coupled to a ChemLab segmented-flow Auto-Analyzer 2 system, the sample stream was introduced to the analyzer using a flow rate of 0.4 mL/min. Once introduced, the sample was mixed with deionized H2O (0.8 mL/min) and 0.5 M HCl (0.1 mL/min). The stream was sent into a gas stripper with N2 as carrier gas flowing at 800 mL/min. The gas phase was passed over H2SO4 and then through a magnesium perchlorate column before IR analysis; calibrations were linear from 0 to 3 mM total CO2. This technique was used to analyze seawater, estuarine water, and freshwaters and was reported superior to alternative methods based on measurements of total alkalinity (D119). FT-IR spectroscopy was investigated by Bellon-Maurel et al. for on-line monitoring of the concentrations of glucose, maltose, maltotriose, and maltodextrin in starch hydrolysis mixtures. Measurements were made using a conventional FT-IR spectrometer with a zinc selenide ATR flat crystal accessory. Calibrations were accomplished using a PLS regression approach with a calibration set of 30 mixtures. Acceptable results were obtained for the determination of glucose and maltose: the standard errors of prediction for glucose, maltose, maltotriose, and maltodextrin were 4.1, 3.4, 4.9, and 4.4 g/kg, respectively, compared to the required precision of 8, 10, 5, and 5 g/kg, respectively. Both repeatability and reproducibility were satisfactory (D120). A near-infrared method for rapid yeast trehalose measurement has been described by Moonsamy et al. The yeast slurry samples were prepared for analysis by dilution in beer, while dried yeast samples were prepared by a method described previously (cf. J. Am. Soc. Brew. Chem. 1994, 52, 145). Dried yeast samples were ground to a powder before spectral measurement using a dry sample fiber-optic probe. The slurry sample probe was introduced directly to the diluted yeast slurries. Both trehalose and glycogen spectra were preprocessed using a secondderivative data analysis over a range of 1000-2222 nm (45009996 cm-1) using 459 data points. The method was reported to produce acceptable performance and may be suitable for rapid in-line measurements of yeast intracellular components (D121). NIR is reported by King-Brink et al. for use in the rapid analysis of meat composition over the typical raw material ranges for fat (6.8-58.8%), water (31.4-72.2%), and protein (8.9-21.0%) (D122). NIR has been used for many years in the routine analysis and discrimination of wheat products for quality control or assessment of the end products as reported by Bertrand et al. (D123). NIR is

presented by Li et al. as a means for rapid assessment of the potential malting quality of barley and malt. In this work, 10 different components are monitored in the barley and malt for direct measurements of malting quality (D124). NIR has been proposed as a rapid method for determination of free fatty acid in fish and its application in fish quality assessment by Zhang and Lee (D125). FT-IR is described by Hewavitharana and van Brake as a method for the rapid determination of casein in raw milk (D126). NIR and a customized large cassette presentation module are used by Berding and Brotherton for analysis of heterogeneous samples from sugar cane evaluation trials (D127). A summary of the use of NIR for rapid nondestructive raw material identification in the cosmetic industry is described by Grunewald (D128). A general description of the uses for NIR in the food industry is given by Hoyer (D129). Song and Otto describe rapid determination of constituents in sausage products by near-infrared transmission spectroscopy (D130). Near-infrared transmission spectroscopy over the silicon detection region is presented by Tenhunen et al. for use in the malting industry (D131). NIR assessment of curds in cheese making is described by Lee et al. (D132). NIR has also been demonstrated by Pascual et al. to be a useful measurement technique for automatic determination of protein fractions in Manchega ewe’s milk (D133). Rapid detection of insect contamination in cereal grains is reported by Chambers and Ridgway (D134). In-line NIR measurements are compared to density/sound velocity measurements for the determination of alcohol and extract in brewed alcoholic beverages by Stokes and Blazier (D135). A variety of molecular spectroscopic techniques such as UV, IR, and Raman spectroscopy are compared by Chmielarz et al. as measurement techniques for studies of the double bond positional isomerization process in linseed oil (D136). NIR technology applied for on-line assessment and monitoring during the manufacture of sugar syrups is described by Jones et al. (D137). A general discussion and review of NIR applications in the milling industry is described by Brotherton and Berding (D138). Rapid fruit juice analysis using infrared spectroscopy is described by Meurens et al. (D139). In-line NIR measurement of important beer components is described by Petersen et al. (D140). NIT spectroscopy is described as a “powerful tool for fast process and product control in breweries and malthouses” by Zahn (D141). Medicine and Clinical Chemistry Applications. Cigarette smoke samples were collected on a glass fiber filter. The filters were measured by Di Luzio et al. using the NIR region from 1445 to 2345 nm, with one measurement per filter. Calibration results were linear from 0.08 to 1.49, 0.02 to 2.46, and 1.13 to 17.15 mg/ cigarette for nicotine, moisture, and tar, respectively. This technique is fast, simple, nondestructive, and safe. It can be used for prescreening samples (D142). Near-infrared Fourier transform Raman spectroscopy has been used with PLS regression to analyze cholesterol/cholesterol, linoleate/cholesterol oleate, and cholesterol palmitate/cholesterol stearate mixtures and to determine single components within the ternary mixtures. The method is being developed by Le Cacheux et al. for eventual application to in situ measurements of arterial walls (D143). Fiber-optic evanescent-wave spectroscopy was combined with FT-IR by Simhi et al. to measure the total protein, cholesterol, urea, and uric acid in 2 mL of human serum. Multivariate calibration models derived Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


from a PLS algorithm were compared to a neural network calibration. The method was reported as applicable for analysis of glucose and triglycerides in serum (D144). Lipids were determined directly in fecal homogenate by FT-IR-ATR and partial least-squares regression modeling by Franck et al. A zinc selenide ATR accessory was used for optical sampling of the slurries. The infrared spectral regions 1763-1730, 1680-1612, 1595-1568, 1564-1524, 1491-1448, 1429-1400, and 1390-1367 cm-1 were used for quantitation. The infrared spectral results were within 2.5% of those obtained using the reference gravimetry method (D145). An overview is presented by Stevens and Vadgama of the use of IR spectroscopy in clinical chemistry applications. The use of near-IR spectroscopy to determine hemoglobin, glucose, urea, and bilirubin in blood is described. Additional uses for NIR for monitoring oxygen levels and measuring body fat are presented. A general discussion for the development of near-IR analyzers for use in noninvasive near patient testing is discussed (D146). Nearinfrared Raman spectrometer systems are proposed by Brennan et al. for in vivo studies of diseased human tissue. Safe and rapid data acquisition are benefits of such technology. Two designs of such systems are described: (1) a high-quality-spectra instrument for laboratory use and (2) a trolley-mounted optical-fiber instrument for clinical use (D147). A portable FT-IR gas analyzer was evaluated by Ahonen et al. for industrial hygiene measurements made in different factories using organic solvent mixtures. The detection limit was ∼1 mg/m3. The results obtained were compared with those obtained by the generally used adsorption tube method. Continuous monitoring with this system allows exposure peaks to be identified and allows preventative measures to be taken (D148). Fourier transform infrared spectrometry was used by Cole and Martin to determine gas-phase sidestream cigarette smoke components (i.e., the smoke generated during smolder from the lit end of a cigarette). Five gas-phase components, viz., NH3, CO2, CO, HCN, and NO, were determined in the sidestream cigarette smoke from five different cigarette brands. Gas concentrations were calculated from absorbance measurements made every 30 s. A 20-fold decrease in sample analysis time was achieved with the described method (D149). Mid-IR DRIFTS and NIR spectrometry were used by Lewis et al. to analyze fibrous (asbestos) and nonfibrous forms of serpentine and amphibale minerals. Raman spectra were obtained with excitation at 785, 632.8, and 1064 nm. Results indicated that NIR spectrometry was the most useful method due to the following: (1) relatively easy to interpret spectra in the 7400-6900-cm-1 range, (2) a higher single-to-noise ratio; (3) the use of silica opticalfiber cables; and (4) the short analysis time in comparison to Raman and FT-IR. The paper reports that the method compares well with other conventional methods of asbestos analysis such as optical microscopy and transmission electron microscopy (D150). NIR spectroscopy or “near-infrared hemoglobinometry”, is described by Kuenstner and Norris for human blood samples. NIR spectra were measured from 400 to 2500 nm at 2-nm intervals in an open-faced cell with a vertical light path. The combination of (A1638 - A1628)/(A1638 - A1626) gave highly accurate results, with R2 ) 0.99 and standard errors of 0.32 g/dL. It was noted that no diodes are available in this region, making it difficult to produce a dedicated low-cost instrument at this time (D151). The applica134R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

tion of noninvasive near-infrared hemoglobin spectroscopy for in vivo monitoring of tumor oxygenation and response to oxygen modifiers is described by Hull and Foster (D152). Near-infrared analysis of dried human serum for determining human serum urea concentration is described by Hall and Pollard. Sera was allowed to reach room temperature and a 400-uL portion was spotted onto filter paper. The filter was then dried at 80 °C for 3 min. After cooling to room temperature for 30 s, the sample measured using a spectral range from 400 to 2500 nm. Urea-related spectral bands were observed at 1450, 1470, and 1980 nm. Unmodified serum samples were also analyzed using a standard method for comparison. Results are presented and discussed (D153). Near-infrared spectroscopy was used for whole blood and blood serum sample analysis by Domjan et al. Examples for qualitative detection of protein and lipid in human sera, as well as distinction of albumin and globulin dissolved in physiological saline solution, are discussed. Other constituents measured were protein, lipoprotein, oxygen saturation, and carbon dioxide pressure in whole blood. Results indicate that NIR spectrometry is a rapid, accurate, and cost-effective method for determining quality parameters in either whole blood and serum (D154). Kromoscopic analysis is suggested by Sodickson and Block as a possible alternative to standard spectroscopic analysis for noninvasive measurement of analytes in vivo. Kromoscopy utilizes broad-band spectrally overlapping NIR detectors. The analyte bands are integrated using two or more detectors with different relative weightings. The response of the detectors to glucose was tested by measurement of the transmission of a series of glucose solutions in a 4-cm-path cell with a range up to 0.6 M (D155). The use of NIR spectroscopy for process control of antithrombin III production in biotechnology manufacturing is discussed by Harthun et al. (D156). Specific details of wavelength selection and correlation data are given by Domjan et al. for the rapid analysis of β-lipoprotein in human blood serum using near-infrared spectroscopy (D157). A spectrograph for noninvasive glucose sensing is described in a patent (D158). A general discussion of the uses of noninvasive infrared spectroscopic analysis in clinical chemistry is described by Stevens and Vadgama (D159). In situ characterization of β-amyloid in Alzheimer’s diseased tissue by synchrotron Fourier transform infrared microspectroscopy is described by Choo et al. (D160). Noninvasive blood glucose monitoring by means of near-infrared spectroscopy is described, and methods for improving the reliability of the calibration models is presented by Mueller et al. (D161). A paper discussing the means of enhancing calibration models for noninvasive nearinfrared “spectroscopical” blood glucose determination is presented by Fischbacher et al. (D162). Passive remote Fourier transform infrared spectroscopy is used for detection of emission sources by Demirgian et al. (D163). Noninvasive near-infrared spectroscopy is described by Wolf et al. for monitoring regional cerebral blood oxygenation changes during physiological experiments in the rat (D164). A dissertation by Pastrana-Rios describes the in situ infrared spectroscopic test of the pulmonary surfactant system (D165). Laser diodes and a photoacoustic sensor head are presented by Spanner and Niessner for the noninvasive determination of blood constituents (D166). Near-infrared spectroscopy is used by Hamaoka et al. for noninvasive measurement of oxidative metabolism on working human muscles (D167). A

general discussion by Heise on the current state of the art using near-infrared spectroscopy for noninvasive monitoring of metabolites is given (D168). Fourier transform infrared microspectroscopy is used by LeVine and Wetzel for in situ chemical analyses of frozen tissue sections of the white matter in the rat brain (D169, D170). A novel approach to using NIR spectroscopy for noninvasive, reflectance spectroscopy is described by Schlager and Ruchti (D171). The use of near-infrared spectroscopy and neural networks for noninvasive determination of blood/tissue glucose is described by Jagemann et al. (D172). Noninvasive detection of hemoglobin oxygenation changes during cortical spreading depression in the rat brain is described by Wolf et al. (D173). Second-derivative near-infrared spectroscopy is described by Cooper et al. as a method for the noninvasive measurement of absolute cerebral deoxyhemoglobin concentration and mean optical path length in the neonatal brain (D174). Near-infrared spectroscopy is presented for the noninvasive measurement of blood glucose by Noda et al. (D175). Petroleum, Natural Gas, and Fuel Applications. On-line information obtained from a photodiode-array NIR absorption spectrometer and a linear CCD fluorescence spectrometer were combined by Litani-Barzilai et al. to predict 10 petroleum properties including research and motor octane numbers, vapor pressure, API gravity, and aromatic content. The proposed combined spectral technique produced a standard error of prediction (SEP) for the octane number of 0.2. When NIR was used alone, the SEP for the octane number was 0.4 (D176). Near-IR reflectance spectroscopy was proposed by Stallard et al. for the determination of motor oil contamination in sandy loam. The sieved and dried sandy loam containing virtually no organic matter was weighed, treated with a few drops of Pennzoil 10 W 30, weighed again, mixed with a spatula, and mounted in a commercial FT-NIR spectrometer fitted with a diffuse reflectance accessory. The angle of incidence was 45°, and both the specularly and diffusely reflected beams were collected. The spectrum was recorded over the range 1600-1900 nm. For 0.13-0.26% (w/w) contamination (95% confidence), the 1 σ precision was 0.17% (w/w). The design of a portable instrument for field use having seven spectral channels each having a resolution of 10 nm is suggested (D177). Automation of FT-IR spectrometry for liquid samples is described, including hardware and software, by Stanek. Turnkey applications have been demonstrated for lubricating oil analysis, fuel, and edible oils and fats (D178). Qualitative analysis of oil sand slurries using on-line NIR spectroscopy and an optical-fiber-bundle probe was demonstrated by Friesen. Diffuse reflectance NIR spectra were measured from 1100 to 2300 nm. The condition of the stream was indicated by using principal component scores plots. Changes in the steady state of the sand were indicated by directional changes in the scores plots (D179). FT-IR spectrometry and multivariate regression were reported by Andrarde et al. to be useful in the quality control of kerosene production. Results are presented of the progress from sequential univariate tests to multivariate approaches, leading to improved laboratory performance and reduced delay times for critical production information (D180). The details for on-line advanced control of a gasoline blender using NIR spectroscopy are presented by Lambert et al. The NIR analyzer monitors the product stream from a gasoline blender.

This information is passed to the control software. The software adjusts the flow rate of the components to meet the blending target specifications. The final product was also analyzed off-line for quality control using NIR spectrometry (D181). On-line NIR analysis has also been demonstrated by Lambert et al. as a means to optimization of steam-cracking operations. NIR analytical technology has been used since 1991 for in-house on-line process control for a steam cracker. The NIR analyzer consists of a sampling conditioner for sample filtration, water removal, temperature control, and the NIR spectrometer. A multiplexer is utilized with two measurement cells linked by fiber optics. The measurement cells can be used to analyze up to 13 naphtha properties and 16 raw gasoline properties, respectively, in less than 1 min (D182). Hydrocarbon and CO2 measurements were performed using GC, free induction decay (FID), FT-IR spectrometry and the results compared by Stephens et al. (D183). The results obtained using a combination of FT-NIR and GC are used by Kania to optimize gasoline blending. FT-NIR spectrometry is used for on-line analysis of gasoline for octane number, while GC analysis is used to monitor distillation products, aromatics, total olefins, and oxygenates (D184). AOTF transmission spectrophotometry has been demonstrated by Pruefer and Mamma as a measurement method for the on-line analysis of motor octane number in gasoline (D185). Applications of on-line NIR, including steam cracker feedstock optimization and gasoline blending, are described by Lambert et al. (D186). An in situ infrared spectroscopic investigation of selective alkylation of toluene over basic zeolites by Palomares et al. is found (D187). A review with 52 references of the use of NIR in refinery and petroleum process monitoring is given by Workman (D188). In situ FT-IR monitoring of a solar flux-induced chemical process is presented by Markham et al. (D189). A general description of the use of NIR analysis for refineries is given by Buttner (D190). The method for control of a steam-cracking process by near-infrared spectroscopy is described in a patent (D191). A general description of the use of spectroscopic techniques used to study combustion chamber chemistry is given by Masi (D192). A patent using NIR to determine cracking properties is described by Descales et al. (D193). A method using NIR to determine lubricant properties is described in the patent literature (D194). An in situ infrared absorption method for determination of high-pressure phase diagrams of methane-heavy hydrocarbon mixtures is given (D195). Gas chromatographic and IR spectroscopic data combined with PLS regression analysis is presented for quality control of jet fuels by von Seinsche et al. (D196). An on-line application of an acoustooptic NIR analyzer for product monitoring and plant control in refineries is described by Buettner et al. (D197). The use of on-line near-infrared optimization for refining and petrochemical processes is given by Lambert et al. (D198). The on-line determination of total olefins in gasoline by process gas chromatography and Fourier transform infrared spectroscopy is found in a paper by Bade et al. (D199). In situ examination of coal macerals oxidation by infrared microspectroscopy is given in a paper by Landais (D200). A method and apparatus for nonintrusive in situ chemical analysis of a lubricant film in running reciprocating machinery using near-infrared spectroscopy is described in a patent (D201). A NIR method for in situ measurement and control of a low-inventory alkylation unit Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


is described in a patent (D202). A method for FT-IR process monitoring of metal powder temperature and size distribution is detailed by Rosenthal et al. (D203). Pharmaceutical Applications. Quantitation of the active compound and major excipients in a pharmaceutical formulation by near-infrared diffuse-reflectance spectroscopy with a fiber optical probe is presented by Blanco et al. Spasmoctyl samples, containing otilonium bromide as active agent and microcrystalline cellulose, maize starch, sodium starch glycolate, and glyceryl palmitostearate as excipients, were analyzed by NIR spectrometry (D204). A novel approach to the transfer of multivariate calibrations based on finite impulse response (FIR) filtering of a set of spectra is described by Blank et al. The method uses a spectrum (frequently the mean of a calibration set) on the target (secondary) instrument as a means to calculate the transfer filter. The method NIR was compared with direct transfer and piecewise direct transfer (PDT) on NIR reflectance spectra of different sample types (D205). A clean method using flow injection Fourier transform infrared spectrometry for the simultaneous determination of propyphenazone and caffeine in powdered pharmaceutical tablets is given by Bouhsain et al. (D206). On-line near-infrared spectroscopy is used to determine the end point for polymorph conversions in pharmaceutical processes as described by Norris et al. (D207). The incorporation of HPLC/FT-IR spectrometry into pharmaceutical research programs in such a way that it can compete with, for example, NMR and MS, and can be automated is discussed by Pivonka and Kirkland (D208). A near-infrared pattern recognition method is described by Aldridge et al. for the identification and determination of the quality of the solid drug substance, polymorph A. The method can discriminate between the desired polymorphic form of the drug and another polymorph, detect samples containing trace levels of the undesired form, and discriminate between the desired polymorph and other crystalline forms. The Mahalanobis distance pattern recognition method was more suitable than the soft independent modeling of Class analogy (SIMCA) residual variance test. It was not sensitive to variation in response between NIR instruments, so multivariate standardization was not required (D209). NIR spectroscopy was evaluated by Sekulic et al. as an on-line technique for monitoring the homogeneity of pharmaceutical mixtures during the blending process. A model mixture of 10% sodium benzoate, 39% microcrystalline cellulose, 50% lactose, and 1% magnesium stearate was used to evaluate the measurement technique. A V-blender was modified to accept a fiber-optic probe located directly at the axis of rotation. Sample components were charged to the blender in the same order throughout the experiments. Transmission spectra were collected from 1100 to 2500 nm (25 spectra/run) over a 25-min period. The method was able to determine blend homogeneity long before the typical blending period was complete (D210). An NIR instrument was interfaced to a blending vessel using a fiber-optic cable. Data were fed to a PC computer via a parallel digital input-output interface and processed by use of a customized software package. The system was tested by preparing a three-component mixture of 10% active ingredient, 45% lactose, and 45% maize starch. The system was acceptable for controlling a fully automated blending system (D211). NIR spectroscopy was coupled with a polar qualification system (PQS) by Plugge and 136R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

Van der Vlies for detecting slight differences in the physical and chemical characteristics of pharmaceutical materials. Measurements were performed using a spectrometer equipped with a rotating sample holder for powders. Each spectrum was transformed into the second derivative and then to a polar spectrum in which the absorbance valves were the radii and the angle of the function was designated as the wavelength. The center of gravity of the polar spectrum was taken as a quality parameter indicator. The software to run the PQS was written in PASCAL 7.0. The method was used to study the differences in particle size distributions between various lactose samples and also the effectiveness of a blending process for producing homogeneous mixtures (D212). Quantitative analysis of headache tablet components was described by Schmidt using NIR reflectance spectroscopy. Powdered samples were introduced to the FT-NIR spectrometer by an automatic system. The measurement technique utilized the diffuse reflectance of the NIR radiation guided onto the sample using optical fibers; the reflected light was also directed onto the detector by optical fibers. A PLS method was used for calibration. Spectra were acquired from 6500 to 5500 cm-1 at 4-cm-1 resolution and a 20-s measurement time. Samples were also analyzed through the plastic packaging material at 2-cm-1 resolution. Results agreed with those obtained by HPLC and gravimetry (D213). Near-infrared reflectance spectroscopy was used by Han and Faulkner to quantitatively analyze tablets during production. Spectra were acquired over the spectral range of 1100-2500 nm and converted to second-derivative spectra. For moisture determination, the scan range was 1400-1450 nm. Partial least-squares regression was used for calibration. For coating thickness measurements, the primary analysis wavelength was 2162 nm. Identification of tablets inside blister packaging was also demonstrated. Results are compared to those obtained using HPLC (D214). Qualifying pharmaceutical substances using nearIR spectroscopy and a PQS is described by van der Viles et al. The PQS is a discriminant method used in detecting small differences in chemical and physical properties for qualifying pharmaceutical substances by fingerprinting. Full details of the investigation have been published previously (cf. Pharm. Technol. Eur. 1995, 7, 43) (D215). Tablets (pink, pentagonal) or film-coated tablets (white, oblong) were analyzed by NIR (1100-2350 nm) reflectance spectrometry through the blister pack plastic using a fiber-optic bundle as described by Dempster et al. (D216). Wavelength distance, Mahalanobis wavelength distance, and SIMCA residual variance distance were investigated as discriminant techniques. Library validation and test set data are presented, and the advantages and disadvantages of each discriminant method is presented. The use of FT-NIR spectrometry for the quality control of solid pharmaceutical formulations was investigated by Dreassi et al. The quantitative performance of the method was demonstrated by testing for three solid drug formulations, viz., powders containing benzydamine hydrochloride and tricetol, pills containing ibuprofen, and tablets containing paracetamol (D217). The coupling of chiral methods such as optical rotatory dispersion or circular dichroism with more conventional methods such as UVvisible or IR spectrometry was performed by Erskine et al. to obtain optical purity and concentration information. Mathematical approaches such as partial least-squares and principal component

regression methods were used to predict the enantiomeric purities of test mixtures (D218). In situ Fourier transform infrared and calorimetric techniques were used by Landau et al. to study the preparation of a pharmaceutical intermediate. A reaction effluent passed through an ATR flow cell and was monitored by FT-IR spectrometry. Periodic samples of the reaction effluent and the off-gas were analyzed by HPLC. The data obtained were used to give a mechanistic insight into the process, and statistical tests on the data showed when the process deviated from the desired operating conditions (D219). The use of NIR reflectance spectrometry for the quality control of ranitidine hydrochloride tablet production is described by Dreassi et al. Discriminant analysis was applied to the IR spectral data to construct calibration maps to identify ranitidine hydrochloride and mixtures of ranitidine hydrochloride for tablets, cores, and coated tablets. Multiple linear regression was used to construct calibration systems to determine the compound and its H2O content during the various production stages up to the finished product. The method gave satisfactory results in all instances, thereby allowing qualitative and quantitative checks at all stages of the production cycle (D220). NIR spectrometry was used by Gonzalez et al. to analyze a pharmaceutical mixture of a polyalcohol, a cellulose and a thickener. Qualification of the spectra was achieved using pattern recognition techniques. A computer program was developed for comparison of the sample spectra with a standard spectrum. The program enabled samples with a “fail” qualification to be classified into four groups according to the shape of the plot and the wavelength that presented the greatest distance from the mean (D221). NIR (1100-2500 nm) reflectance spectroscopy was used by Dreassi et al. to characterize several pharmaceuticals on the basis of their physical properties. NIR spectra were recorded and linear discriminant analysis was used to process the raw absorbance data. The method was used to distinguish substances with different crystalline states (nitrofurantoin, gemfibrozil, chenodeoxycholic acid, levamisole-tetramisole) and densities (paracetamol, ibuprofen) (D222). FT-IR spectrometry was used by Severdia to detect traces of lubricating grease on the product contact surfaces of pharmaceutical tablet presses. Press surfaces were swabbed with Whatman No. 1 filter paper soaked in n-hexane (HPLC solvent grade). The swab was allowed to dry in air and the lubricant residue was extracted by sonication of the swab in n-hexane. After removal of the filter paper, the extract was gently evaporated to dryness, the residue was dissolved in a known volume of CDCl3, and the transmission IR spectrum of the solution was recorded (32 scans) at a spectral resolution of 4 cm-1 against a solvent reference (D223). An NIR spectrometric method is described by Blanco et al. for the quality control of the manufacturing process for a solid pharmaceutical preparation. The parameters measured include the identification and qualification of the product and the quantitation of the active component. The NIR spectrum was recorded from 1100 to 2500 nm using a spinning cuvet sample holder or a fiberoptic probe. The second derivative of the spectrum was compared to the reference spectra contained in a library compiled for 163 pharmaceutical materials. The sample was identified by determining the match that produced the highest correlation. Once identified, a comparison of the sample and reference spectra yielded information concerning all the parameters affecting the

NIR spectra, which allowed qualification of the sample. The concentration of the active component was determined by partial least-squares calibration using two strategies to select samples for inclusion into the calibration set, namely, flat calibration or sample selection (D224). A near-infrared reflectance analysis method for the noninvasive identification of film-coated and non-film-coated, blister-packed tablets is described by Dempster et al. (D225). Quantitative Fourier transform near-infrared spectroscopy is used in the quality control of solid pharmaceutical formulations as presented by Dreassi et al. (D226). In situ Fourier transform infrared and calorimetric studies of the preparation of a pharmaceutical intermediate is detailed by Landau et al. (D227). NIR spectroscopy is used as a tool for in-process control in pharmaceutical production as presented by Steffens and List (D228). Nondestructive quality control of pharmaceutical tablets by near-infrared reflectance spectroscopy is described by Weiler and Sarinas (D229). Internet sites for infrared and near-infrared spectrometry are given by Kraemer and Lodder, including on-line instruction and direct communication (D230). Process control and end point determination of a fluid bed granulation by application of near-infrared spectroscopy is described by Frake et al. (D231). NIR spectroscopy is described by Wilson et al. for quality control in the pharmaceutical industry (D232). Determination of moisture in hard gelatin capsules using near-infrared spectroscopy for at-line process control of pharmaceutics with applications to at-line process control of pharmaceutics is given by Berntsson et al. (D233). Near-infrared spectroscopy as a tool to improve quality is described by Plugge and van der Vlies (D234). Polymer Applications. A new correction algorithm for FTIR on-line film thickness measurement is presented by Wang et al. (D235). Process stream interfacing of in-line NIR analyzers is described by Mindel (D236). Near-IR FT Raman spectroscopy is used by Ozpozan et al. for on-line monitoring of the polymerization of vinyl acetate at 54 °C in H2O containing sodium lauryl sulfate, NaH2PO4, and K2S2O8 in a 150-mL reaction vessel. The 1064-nm line from a Nd:YAG laser passed through the sidearm of the reaction vessel using backscattering geometry. Spectra were recorded at 8-min intervals with scanning from 20 to 3500 cm-1 and a resolution of 4 cm-1. The method is applicable to the monitoring of industrial production of the polymer (D237). An NIR instrument using an InGaAs detector (i.e., 900-1700 nm) was used by Feldhoff et al. in conjunction with a 30-cm-wide conveyor belt moving at 1 m/s carrying waste polymer vessels for identification and sorting. Use of the FuzzyARTMAP neural network classifier as per Wienke et al. (in Chemom. Intell. Lab. Syst. 1996, 32, 165). The device gave an accuracy of >97% in the classification of objects made of poly(ethylene terephthalate), polystyrene, PVC, and cardboard, but was slightly less reliable at distinguishing between polyethylene and polypropylene (D238). An array detector was used by van den Broek et al. over several incremental wavelength regions of 1546-1578, 1545-1655, 16551745, 1700-2150, 2207-2321, and 2115-2550 nm defined by a set of interference filters, over a wavelength range of 1.2-4.6 µm. The system was used to discriminate a variety of plastic materials automatically (D239). The structure of the conducting polymer poly(paraphenylene), synthesized under different electrochemical conditions, was Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


studied in situ by Damlin et al. using external reflection FT-IR spectrometry (D240). An FT-IR microspectrometric system is described for the real-time on-polymer-bead monitoring of multistep combinatorial syntheses carried out in a flow cell, a diagram of which is presented by Pivonka et al. (D241). A description of the use of NIR absorption spectroscopy to study inter- and intramolecular hydrogen-bonding and other interactions of polymers is given by Lachenal and Stevenson. This information is useful for the curing of epoxy resins and for process control in the production of polymers and their blends (D242). A FT-IR spectrophotometer was used by Kirsch et al. to analyze the monomer and comonomer in the reactor feed system of a polyolefin process. The system was used in closed-loop control of the comonomer feed flow to attempt to control product density and type changes (D243). Formation of toxic gases during pyrolysis of polyacrylonitrile and nylons is analyzed by Nielsen et al. using an on-line optoacoustic FT-IR gas analyzer with a fixedbed pyrolysis reactor that consisted of a vertical quartz tube (1 cm i.d.) heated by a Lindberg furnace. The major toxic products of polyacrylonitrile were HCN and NH3; nylons produced NH3 but only minor amounts of HCN. Both types of polymer produced CO (D244). A review by Siesler containing 22 references is presented on applications of optical-fiber remote near-infrared spectroscopy for polymer reaction and process control including synthesis, processing, and recycling (D245). A NIR spectrometer is connected by optical fibers by Brush to a stainless steel immersion-sampling absorption cell (path length 6 mm). The manufacturing conditions of polyoxyethylene and polyoxypropylene glycols necessitate remote in situ analysis at temperatures of >260 °C, with turbulence. Calibration equations for hydroxyl number and acid value using MLR gave standard errors of calibration (SECs) of 0.41 and 0.28, respectively (D246). NIR spectroscopy is described by Scott for the automatic sorting of postconsumer plastic waste using two-color fixed filters transmitting at 1660 and 1716 nm. The ratio of the 1716-nm signal to the 1660-nm signal allows the identification and separation of plastics made from poly(ethylene terephthalate) and from PVC (D247). Near-infrared external reflection spectroscopy was reported for on-line remote monitoring of thermoset resin cure in a polymer manufacturing application by Xu et al. (D248). A general article by Dallin describing the uses of near-infrared spectroscopy in monitoring polymer reactions is found (D249). The challenges associated with using near-infrared spectroscopy for on-line monitoring in polyol production is described by Stromberg (D250). Infrared microspectroscopy is combined with a microtome for assessing potential toxins in polymer packaging by Jickells. This instrumental method allows much more rapid analysis than is possible with dissolution techniques and can be used to study the penetration and migration of food and toxin materials within plastics (D251). Specular-reflectance infrared spectroscopy was used by Gaarenstroom et al. for rapid identification of automotive plastics in dismantling operations (D252). External beam mid-infrared spectroscopy is presented as a method for nondestructive rapid identification of plastics by Mucci (D253). Mid-infrared reflectance spectroscopy is proposed by Zachmann as a method for rapid characterization of black polymeric material (D254). In situ Fourier transform infrared analysis of poly(ester-urethanes) at low temperature is described 138R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

by Rehman et al. using a newly constructed liquid nitrogen-cooled sample stage (D255). Evanescent-wave FT-IR spectroscopy is used by Linossier et al. for in situ study of the degradation of polymersubstrate systems (D256). NIR spectroscopy is used by Fischer et al. for in-line process monitoring on polymer melts (D257). NIR spectroscopy is also proposed by Hansen and Vedula as a means for in-line measurement of copolymer composition and melt index (D258). Step-scan photoacoustic FT-IR spectroscopic studies were used by Niu and Urban, to study surfactant exudation and film formation in 50%/59% Sty/n-BA latex films (D259). Infrared and resonance Raman spectroscopic studies on the photopolymerization process of the Langmuir-Blodgett films of a diacetylenemonocarboxylic acid are described by Saito et al. (D260). A patent describing near-infrared measurement and control of polyamide manufacture is given (D261). Time-dependent polarization modulation infrared spectroscopy was used for an in situ study of photoinduced orientation in azo polymers by Buffeteau and Pezolet (D262). An ultrafast near-infrared sensor is described by Chilling and Ritzmann for rapid on-line identification of plastics (D263). On-line Fourier transform infrared spectroscopy is described by Kirsch et al. as a means of improved process control in polyolefin polymerization (D264). Near-IR spectroscopy is applied by Long et al. as a process monitor for the synthesis of polyester graft copolymers via the macromonomer technique (D265). Spectroscopic attenuated total reflectance and photoacoustic Fourier transform IR approaches are described by Niu et al. for use in monitoring latex film formation at surfaces and interfaces (D266). Infrared spectrometric classification techniques are used by Van Every and Elder for validation in polyolefin formulation (D267). In situ external reflection Fourier transform infrared spectroscopy is presented by Damlin et al. as a method to study the structure of the conducting polymer poly(p-phenylene) (D268). Molecular spectroscopy is described by Zachmann and Turner as a means for identification of black plastics (D269). Near-IR spectroscopy is described in general terms as a method for monitoring the quality and process conditions during polyol production by DeThomas and Brush (D270). A patent describing a process for measuring and controlling the neutralization of inorganic acids in an aromatic polyamide solution based on nearinfrared spectroscopy is given by Moessner (D271). A patent using near-infrared spectroscopy for chemical property determination in polymerization and organic reactions is presented (D272). The use of near-IR spectroscopy for fast on-line identification of plastics for use in recycling processes is given by Eisenreich et al. (D273). In situ, quantitative FT-IR studies of polymers irradiated with synchrotron radiation are described by Wollersheim and Hormes (D274). Remote, in-line monitoring of emulsion polymerization of styrene by short-wavelength near-infrared spectroscopy is described by Wu et al. for aspects relating to performance in the face of normal runs (D275) and process upsets (D276). Remote in situ real-time fiber-optic near-infrared spectroscopy is described by Mijovic and Andjelic in the study of the mechanism and rate of bismaleimide curing process (D277). Near-infrared spectroscopy is described by Sekulic et al. as a method for on-line monitoring of powder blend homogeneity for the mixing of pharmaceuticals (D278). In situ infrared spectroscopy is used by

Yu and Fina for studying the electric field-induced dipole reorientation in oriented nylon 11 (D279). A dissertation by Khettry describing the use of infrared fiber optics for the in-line monitoring of molten polymeric processes is presented (D280). In situ realtime monitoring of epoxy/amine kinetics by remote near-infrared spectroscopy is described by Mijovic et al. (D281). Near-infrared spectroscopy is described for the monitoring of the film-coating process in pharmaceuticals by Kirsch and Drennen (D282). In situ FT-IR spectroscopy, synchrotron SAXS, and rheology are used by Elwell et al. to study the structure development during the reactive processing of model flexible polyurethane foam systems (D283). Near-IR is described by Hansen and Khettry as a method for in-line composition monitoring of molten poly(ethylene-vinyl acetate) processes (D284). A patent presenting a process for measurement of the degree of cure and percent resin of glass fiber-reinforced epoxy resin prepreg is given (D285). Fourier transform infrared spectroscopy is described by Ge et al. as a means for quantitative monitoring of an epoxidation process (D286). Near-IR spectroscopy is used for real-time process monitoring of solution polymerization kinetics by Long et al. (D287). The general use of on-line NIR monitoring of polyols is presented by Miller and Curtin (D288). The application of FT-IR spectrum method in the photocuring process for poly(ester acrylate) is described by Cao et al. (D289). Remote fiber-optic near-infrared spectroscopy is described for in situ real-time monitoring of reactive polymer systems by Mijovic and Andjelic (D290). Molecular spectroscopy is used by Graham et al. to provide rapid identification of plastics components recovered from scrap automobiles (D291). NIR spectroscopy is described by Hansen and Khettry for in-line monitoring of molten polymers using robust fiber-optic probes and rapid data analysis (D292). Recent developments in on-line analytical techniques applicable to the polymer industry are described by Patel et al. (D293). The use of near-infrared spectroscopy to monitor the production of polyurethanes is described by Halla and DeThomas (D294). A Ph.D. dissertation by Lew describing the use of size exclusion chromatography and in-line near-infrared spectroscopy for polyolefin extrusion process monitoring is presented (D295). The use of molecular spectroscopy for the in-line monitoring of titanium dioxide content in a poly(ethylene terephthalate) extrusion process is described by Batra and Hansen (D296). A world patent outlining the process for measurement of the degree of cure and percent resin of fiberglass-reinforced epoxy resin prepreg using molecular spectroscopy is given (D297). This is also reported in this review as a U.S. patent (D285). Near-infrared spectroscopy is presented as a means for an in-line monitoring of molten polymers by Hansen and Khettry (D298). FT-IR external reflection is used by Lutz et al. for a low-cost unit for rapid analysis of carbon-filled rubbers (D299). Process FT-IR spectroscopy is used for real-time analysis of PC/ABS blend composition by Fidler (D300). IR-RAS is used by Tamada et al. for real-time in situ observation of vapor deposition polymerization of N-methylolacrylamide (D301). Nearinfrared spectroscopy is described as a means for at-line (D302) and in-line (D303) process analysis of a variety of properties for poly(ethylene terephthalate) chips. DSC and infrared spectroscopy are used for in situ investigations of thermal processes in polymers by Ziegler et al. (D304). A description of the value offered by

process molecular spectroscopy of polymers is given by Siesler (D305). PROCESS RAMAN SPECTROSCOPY Advances in laser diode and fiber probe technologies have enabled rapid developments in the application of Raman spectroscopy to process relevant samples and systems. Also, concurrent advances in compact spectrometer design and low-noise, highsensitivity array detection systems coupled with these new diode lasers have enabled the application of Raman spectroscopy in areas that were once thought to be forbidden. This review section covers process relevant advances in Raman spectroscopy from 1994 to 1998. Angel et al. (E1) broadly reviewed the state of the art in the utilization of diode lasers for Raman spectroscopy and compared the performance of diode lasers to conventional Raman laser technology. Denton et al. (E2) reviewed advances in detector technology in terms of their impact on the application of Raman spectroscopy for process analysis. Milnes et al. (E3) reported on the uses of fiber-optic sensing technology and included a brief overview of Raman spectroscopy for chemical analysis in processes. Adar et al. (E4, E5) extensively reviewed Raman spectroscopy for process analysis and control. Lewis and Griffiths (E6) detailed the advances in fiber-optic sampling probe technology for process Raman spectroscopy. Al-Khanbashi et al. (E7) reviewed the application of in-line Raman spectroscopy for emulsion polymerization reaction monitoring. Fiber optics have played a significant role in the development of field-portable and process-deployable Raman instrumentation. Therefore, it is useful here to provide a review of the advances in fiber-optic-based Raman probe and hardware design. Several Raman probe designs and methods have been patented for use in various environmental and industrial environments (E8-E10). Tsai et al. (E11) have designed a tapered fiber-optic Raman probe with a subwavelength aperture and demonstrated its usefulness for high-spatial-resolution analytical spectroscopy. Ma and Li (E12) described a fiber-optic coupled graded index lens Raman probe that was spatially optimized for low background interference. They also reported on a reduced background dual-fiber probe with titled end faces (E13). The performance of an on-line turnkey fiberoptic dispersive Raman system was evaluated by Everall et al. (E14). The system was tested with known densities of poly(ethylene terephthalate) and, using multivariate calibration, generated predictions to a precision of 0.002 g/cm3 which compared favorably with previous calibrations using a FT-Raman laboratory system. Nave et al. (E15, E16) described a rugged fiber-optic probe having in-line dichroic filters, a six-around-one excitationcollection fiber geometry, and a sapphire window separating the optics from the process for remote chemical sensing in nuclear environments. Cooney and co-workers (E17, E18) developed numerical models for comparing Raman probes of differing fiber size, number of bundled fibers, and tip geometry. They compared the effects of fiber end taper and lensing on collection efficiency and signal-to-noise ratio. Greek et al. (E19, E20) demonstrated an approach to resonance Raman probe design using numerical modeling and have verified their designs by developing an in situ resonance Raman probe for collecting resonance Raman spectra Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


of tryptophan and DNA. In related work, they reported on the use of resonance Raman probes for obtaining spectra for a variety of biological samples (E21-E24). Lin and co-workers (E25) characterized a two-fiber probe for in situ fluorescence, Raman, and reflectance measurements. Gilmore et al. (E26) described the quantitative detection of several environmentally relevant dyes using diode laser excitation and fiber-optic probe having a single source fiber surrounded by 18 collection fibers. Aust and co-workers (E27) developed an apparatus for measuring polymer curing where the distal end of a fiber probe is inserted into a Teflon tube containing the epoxy sample. The sample tube functioned as a waveguide and yielded a 15 times improvement in Raman signal intensity, and the authors were able to extract cure and sample temperature information from the resulting Raman spectra. Al-Khanbashi et al. (E28) demonstrated a fiber-optic Raman probe based on epi-illumination of the sample. This illumination scheme reduced the background signal from opaque emulsion polymerization reaction samples and allowed for collection of Raman spectra with what the authors called “acceptable performance” for most in-line applications. Dai et al. demonstrated several high-temperature fiber probes for application in molten salts (E29, E30) and an all-silica fiber probe end-coated with an evaporated diamond thin film acting as an internal standard (E31). They also utilized the measurement of anti-Stokes Raman spectra to reduce the silica fiber background in high-temperature applications (E32). Several researchers (E33, E34) reported the development of distributed Raman fiber thermometry systems for high-temperature measurements. Feced et al. (E35) described a similar distributed temperature measurement system with 0.1-m spatial resolution. Sharma and Sheppard (E36) investigated the application of holmium-doped fiber as an in-line notch filter for rejecting the excitation line in fiber Raman systems. Munro and co-workers (E37) demonstrated the use of dielectric stack filters for use in the design and application of a fiber-optic UV Raman probe. Heaton (E38) presented a conceptual design for a compact fiberoptic interferometer-based FT-Raman system and developed theoretical expressions for the expected resolution, bandwidth, signal-to-noise ratio, and sensitivity. Skinner and co-workers (E39) described a remote Raman microimaging system using a coherent microfiber bundle, acoustooptic tunable filter, and CCD imaging camera for analysis of Raman scattering pellet samples with spatial resolution of 4 µm. A multichannel FT-Raman spectrometer using a CCD to record the resulting interferogram was designed and evaluated by Zhao and McCreery (E40). A six-around-one fiber probe was used by Sprunt and Jayasoriya (E41) to couple a FT-Raman system to a differential scanning calorimeter in order to record simultaneous Raman spectra and DSC data as ammonium nitrate changes phase. A lowloss amorphous fluouropolymer capillary liquid core waveguide for Raman signal enhancement in aqueous systems has been described by Altkorn and co-workers (E42). Kamogawa and coworkers (E43) designed a dual-prism multipass Raman cell for analyzing dilute aqueous solutions. The system exhibited a 7-10fold sensitivity improvement over conventional single-pass cells. Greek et al. (E44) described a mathematical processing technique based on a two-point entropy regularization methods for Raman signal recovery and signal-to-noise enhancement. The 140R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

method demonstrated signal-to-noise enhancement over derivative smoothing methods for simulated and experimentally noisy data and required no a priori knowledge of the spectral features. Daniel and co-workers (E45) reported the application of a suite of statistical models for analysis of explosives which demonstrated the classification, novelty detection, and clustering of Raman spectra. Lewis et al. (E46) compared Raman with infrared spectroscopy for the analysis of asbestos and nonfibrous forms of serpentine and amphibole minerals. Lombardi and co-workers (E47) characterized the Raman response to a reference mixture of complex inorganic anions analogous to the environment found in the nuclear waste tanks at the Hanford site, Richland, WA, and, in addition, demonstrated the ability of fiber-coupled Raman spectroscopy to differentiate anionic constituents in an untreated core sample. Bell et al. (E48) designed a high-pressure cell and Raman probe (rated to 15 000 psi), monitored the conversion of acetic anhydride to acetic acid in hydrothermal/supercritical water using the probe, and made use of second-order multivariate curve resolution techniques to obtain the resulting concentration and spectral profiles. Lipp and Grosse (E49) monitored on-line the composition of methylchlorosilane streams during a distillation process with 1000 parts per million detection limit and, in some cases, 100 parts per million sensitivity to changes in composition of the stream. Ewing et al. (E50, E51) and co-workers used fiberoptic Raman coupled with solid-phase extraction to detect trace phosphonate vapors and benzene in water with parts per million sensitivity and have utilized fiber-optic Raman for various vaporphase organic analyses (E52-E54). de Bakker and Fredericks (E55) demonstrated the feasibility of applying Fourier transform fiber optic Raman with partial least-squares calibration in on-line petroleum analysis. Flecher and co-workers (E56, E57) analyzed mock petroleum fuel mixtures of benzene, toluene, and ethylbenzene and determined octane numbers and Reid vapor pressure for a series of commercial gasolines using a dispersive system with a fiber-optic probe and multivariate calibration techniques. Deshpande and co-workers (E58) monitored the vinyl acetate concentration and melt index of molten poly(ethylene-vinyl acetate) copolymer during extrusion. Stellman et al. (E59) performed the first in situ Raman spectroscopic study of microwave polymerization reactions. In similar work (E60), the group used a Raman probe to determine the percent cure of epoxide polymers and composites over a range of 50-99% cure with a reliability of 0.54%. Cooper and co-workers (E61) demonstrated a modulated FT-Raman probe approach for monitoring high temperature which reduced the thermal background for samples at temperatures greater than 370 °C. This modified system generated Raman spectra with enhanced signal-to-noise ratios relative to the unmodified instrument for samples at both ambient and high temperature. PROCESS ELECTRONIC SPECTROSCOPY (UV-VISIBLE AND FLUORESCENCE) Advances in the application of electronic spectroscopy in process analysis can be linked to the development of fiber-opticbased sensors and probes. The installation of remote probes, coupled with fiber optics to an instrument house, has allowed for process spectroscopy to be carried out in real time much like temperature, pressure, and humidity measurements.

It important to note here that not every application described in this review contains a direct process relevance delineated by the authors. Rather, works are included in this discussion that have potential for application in process environments. The types of devices and applications not explicitly designed for process applications but having definite relevance include compact or ruggedized instrumentation, novel optical designs, and remote environmental probes. The following section will review the literature reports of process relevant advances in UV-visible and fluorescence spectroscopy from 1994 to 1998. Curiously, no significant reviews devoted to process UV-Vis spectroscopy were published during the time period covered by this review. However, advances were made in both instrumentation/probe design and applications of fiber-optic-coupled spectrophotometric techniques. Bratz and co-workers (F1) described the design and application of a “raster spectral device” sensor that utilizes several light sources which are individually rastered in the flow cell compartment. The sensor was applied to measurements of nitrite, microbial growth and degradation, and pH. A tunable laser source that operates in the UV based on a fiber Raman laser was demonstrated by Ilev et al. (F2) and its potential as a new source for UV spectroscopy discussed. The group applied the source to refractive index dispersion sensing for characterization of bulk optical materials (F3). Anderson et al. (F4, F5) described the theoretical and experimental development of grating light reflection spectroscopy and demonstrated its utility for simultaneous refractive index and absorbance measurements of optically dense liquids and slurries (F6). Landis and Seliskar (F7) compared the theoretical and experimental optical performance of fiber coupling spheres and gradient index (GRIN) lenses for remote spectroscopy and found that GRIN lenses were the preferred optic to use for in- and out-coupling radiation from chemical processes. Morgan and Wilson (F8) reported the design of an imaging analyzer utilizing a coherent fiber bundle, two CCD detectors, and an array of wavelength-selective filters which was applied to the analysis of plasma in the JET pumped divertor. Kubo and co-workers (F9) described a high-resolution (5.3 pm) spectrometer used to detect Doppler broadening of spectral lines in a Tokamak plasma. Klinkhammer (F10) reported the development of a fibercoupled ultraviolet-visible spectrometer, called the zero-angle photon spectrometer (ZAPS) probe, for use in oceanographic environments which was used as a fluorometer for analysis of organic matter and as a chemical sensor for analysis of Mn to 0.1 nM in ocean water. Dierking and Karim (F11) described a lowcost, rugged, solid block, stationary visible Fourier transform spectrometer and compared its performance to a standard laboratory radiometer using a low-pressure mercury lamp and a incoherent red light-emitting diode. Morgan and co-workers (F12) detailed the construction and performance of a compact frequency domain fluorometer featuring a modulated deuterium light source with potential as a compact, fiber-coupled fluorescence lifetime sensor. An all-fiber-optic infrared Mach-Zender-based Fourier transform spectrometer was designed and described by Stelzle et al. (F13) and the extension of the instrument to ultraviolet absorbance applications was discussed. Vaarala and group (F14) demonstrated a fiber-coupled prism-grating-prism imaging spec-

trograph/charge-coupled device (CCD) system for on-line inspection of oil film thickness on rolled steel by UV fluorescence. Cho et al. (F15) described a wavelength calibration method for charge-coupled devices and the optical probes (F16) used in multichannel fiber-optic spectroscopy of pharmaceutical tablet dissolution. The group also reported the use of the system coupled with chemometric data analysis for multicomponent drug dissolution determinations (F17). The design and application of a complete robotic drug dissolution-monitoring system was reported by Aldridge et al. (F18). The system utilized on-line fiber-opticbased UV detection of drug species, was composed entirely of off-the-shelf items, and eliminated the need for off-line HPLC and UV spectroscopic analyses. Chen and Brown (F19) described a drug dissolution monitor based on a mechanically scanned multiplexed fiber probe design and a conventional spectrophotometer which made use of multivariate calibration schemes to determine release rates for various drug formulations. A method for the transfer of a multivariate calibration model from a sample cell to a fiber-optic probe based on piecewise direct standardization and Fourier preprocessing was described by Chen et al. (F20) and shown to be effective in transferring calibrations between probes connected to the same instrument. Foret and co-workers (F21) described a compact, rugged, dualdetection capillary electrophoresis/mass spectrometer with UV detection close to the tip of the electrospray interface. Lindberg et al. (F22) described a capillary electrophoresis detection cell based on a fiber-optic coupling with focusing optics and aperture to maximize throughput and reduce stray light via matching the spot size to the inner diameter of the capillary. Altkorn et al. (F23, F24) described the fabrication and characterization of a capillary cell coated with low index of refraction fluoropolymers which allows for waveguiding of scattered fluorescence and Raman scattered light in aqueous systems. Spectroscopic sensitivity was increased using this arrangement because a larger portion of the isotropically emitted light was guided down the liquid core to the detector. Rhodes and Fox (F25) developed a corrosion-resistant fiber-optic-coupled high-pressure cell for supercritical fluid studies using time-resolved fluorescence and absorbance measurements. Prodromidis et al. (F26) described the determination of trace copper(II) (4 ng/mL) based on its catalytic effect on a phenol oxidation reaction using a fiber-optic dip probe coupled to a filter spectrophotometer. Wruck and co-workers (F27) developed a technique for plutonium determination by thermal lens spectroscopy using a fiber-coupled diode laser source and found the method compared favorably to dye laser-based photothermal or photoacoustic methods. Bublitz and co-workers (F28-F30) reported the application of fiber-optic time-resolved, laser-induced fluorescence spectroscopy for the detection and analysis of environmental pollutants such as engine oil in water and polyaromatic hydrocarbons in natural soils. A pH probe based on the visible spectoelectrochemical monitoring of a poly(1-naphththylamine) film immobilized on a quartz crystal microbalance in solution was demonstrated by Xu et al. (F31). Egami and co-workers (F32) developed a fiber-optic pH probe based on evanescent absorption of methyl red immobilized in the poly(methyl methacrylate) fiber cladding. Kuhn and Dyke (F33) demonstrated a renewable reagent visible fiberAnalytical Chemistry, Vol. 71, No. 12, June 15, 1999


optic sensor for high-acidity measurement based on a Nafion membrane and a flowing reagent stream. Zangaro and co-workers (F34) developed a fiber-optic rapid multiexcitation florescence spectrometer using a nitrogen pump laser and a rotating bank of laser dye cuvettes. The high-sensitivity spectrometer was applied to biological tissue analysis, but clearly has implications for rapid process analysis for scattering samples as it provided multiwavelength laser excitation over a 200-nm bandpass in 600 ms. Horvath and Glazier (F35) described the analysis of a model fermentation system containing high cell mass using a fiber-optic fluorescence probe and at various excitation/emission ranges. A short communication by Tena and Valcarcel (F36) detailed the fiber-optic interface they designed for on-line analysis of solid samples, in this case ground coffee, by supercritical extraction. Huston and co-workers (F37) described a fiber-opticcoupled thermoluminescent dosimeter for radiation detection in remote areas. Litani et al. (F38) developed an on-line gasoline properties monitor that employed a combination of fiber-coupled NIR and laser-induced fluorescence (LIF) spectroscopies. They found that the addition of LIF information into the multivariate data analysis improved octane number predictions 2-fold over standard NIR techniques. Remillard et al. (F39) reported the development of a gas-phase optical sensor based on the fluorescence response of a copper zeolite to reactive gas mixtures. The absorption of sulfur dioxide in dilute aqueous acids has been investigated spectroscopically by Krissmann et. al. (F40, F41) using a fiber-optic probe system that can interrogate both the gas and liquid phases. Gao and co-workers (F42) detailed the use of an optically transparent ion-exchangeable poly(acrylic acid)/poly(vinyl alcohol) blend for spectroscopic sensing of copper ion to 10-6 M where the polymer was immobilized as the cladding on a bare fiber core. Potyrailo et al. (F43) described a fiber-optic evanescent wave sensor operating in the near-ultraviolet and demonstrated its capability in the measurement of ozone over the range of 0.020.35 vol %. Schwotzer and co-workers (F44) applied fiber-optic evanescent-wave UV absorption spectroscopy as well to monitoring hydrocarbons in air. Dress and co-workers (F45, F46) described a cylindrical water core waveguide for environmental chlorine measurements in the deep ultraviolet. Belz and group (F47) discussed the development of a miniature “smart sensor” approach for residual chlorine analysis in water based on a fiber-optic probe and compact spectrometer. Anna et al. (F48) reported the theoretical and experimental development of a bitapered multimode fiber evanescent-wave absorbance sensor and showed an improvement in absorbance sensitivity over uniform radius probes of the same length. Nunes and Tong (F49) described the application of an in situ fiber-based degenerate four-wave-mixing laser absorbance sensor for trace detection of liquid-phase analytes and achieved a 10-6 absorbance limit of detection for Eosin B. Carrot and Wai (F50) demonstrated the design and application of a fiber-optic microscale ultravioletvisible sample cell for spectroscopic measurement of uranium oxide complexes in supercritical CO2. Merschman and Tilotta (F51) reported the application of ultraviolet fiber-optic evanescent spectroscopy combined with solid-phase microextraction in determining benzene, ethylbenzene, toluene, and xylene (BETX) compounds in water. A number of researchers (F52-F54) have 142R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

applied fiber-coupled UV reflection spectroscopy for monitoring of cure parameters of industrial polymers and composites. Chen et al. (F55) described an on-line poultry carcass classifier based on a fiber-coupled visible-NIR reflectance spectrometer and neural network data analysis. Several groups (F56, F57) have applied various fiber-optic spectroscopy schemes for in situ analysis of thin film deposition processes. MASS SPECTROMETRY Mass spectrometry is particularly appealing for the analysis of process streams, due to its inherent speed, sensitivity, and molecular selectivity. As a technique for process and field-portable measurement, mass spectrometry is increasingly used as a detector for hyphenated methods (e.g., GC/MS and LC/MS) for the analysis of complex samples. Another reason for the appeal of mass spectrometry is that virtually any sample is amenable to analysis in a mass spectrometer, due to the vast array of ionization techniques that are available. The historical limitations in the implementation of mass spectrometry were typically seen to be size, cost, and complexity. These impediments are in various states of being removed, with extensive ongoing research in miniaturization of every kind of mass spectrometer (ion trap, ion cyclotron resonance, time of flight, magnetic sector, quadrupole), as well as continuing developments in computer technology and software. This review will be divided into two major sections. The former focuses on hardware (instruments, inlets, ionization sources), and the latter major section describes applications. The application section is further broken down into process control, environmental (air, water, soil, and chemical/biological weapons analysis), and the characterization of aerosols and airborne particulates. This review is for publications that appeared during the period 19951999, which skews the subject matter toward developmental efforts. For an in-depth review of “traditional” process mass spectrometry, the reader is referred to the previous articles in this series (G1, G2), or the citations in the general reviews section. Where possible, journal citations are made; in some cases, relevant work is noted from conference proceedings, though these are generally more difficult for the reader to obtain. For the purposes of this review, the applications of mass spectrometry cited do not need to be demonstrated in a production plant or in the field. Many citations are to journal articles that demonstrate an approach that might be readily implemented for process analysis or show trends in the development of the technology. It is interesting to note that environmental applications still show up in the majority of portable- and process-type citations, even though mass spectrometry has been used for such applications for quite some time. There has recently been a significant increase in interest in the analysis of air-borne particulates and aerosols, due to the amendments to the clean air act, as well as potential application to the analysis of chemical and biological warfare agents. The lines between “traditional” laboratory-based mass spectrometry and on-line analysis are blurring, with numerous pharmaceutical laboratories offering open-access analysis (G3) and “intelligent automated LC/MS/MS” systems (G4). These systems are different from conventional laboratory analysis, and closer to process mass spectrometry, in that the mass spectrometer system is being configured for minimal operator intervention: the system does analysis, data reduction, and some rudi-

mentary self-diagnostics without being acted upon. There are also early couplings of combinatorial microtiter arrays and highthroughput analysis (G5, G6), in some cases automated with robotic introduction (G7). Other approaches to facilitating highthroughput mass spectrometry include microchip (G8) and multichannel electrospray (G9) ionization sources. In other cases, efforts directed at miniaturization of instrumentation are enabling new applications areas, such as point of care diagnostics based on respired air (G10). Reviews. There have been a number of general review articles that might be of interest to people practicing in this field. The topic of field analytical chemistry, with a detailed section on portable mass spectrometer instrumentation, was summarized in the last Analytical Chemistry Applications review article (G11). (Consequently, the focus of the section on compact and fieldportable instruments will be on research from 1997 to 1999.) The same journal had sections on related areas that included analysis with mass spectrometry, particularly the sections on environmental analysis (G12) and air pollution (G13). Kotiaho (G14) has authored a very informative primer on process analysis with a mass spectrometer, including detailed descriptions of numerous common sample interfaces. Other broad reviews of process mass spectrometry, with applications in diverse areas, were authored by Bohatka (G15), Walsh and LaPack (G16), DesJardin et al. (G17), and Walsh et al. (G18). Compact Instrumentation. Within the last several years, there has been a significant effort directed at the development of ever smaller mass spectrometers. Time of flight (TOF) has undergone a renaissance in the last several years and has received considerable interest as a candidate for miniaturization. A group at the Johns Hopkins University Applied Physics Laboratory has described efforts (G19, G20) directed at the fabrication of a highresolution TOF for biological applications (G21) or coupled with matrix-assisted laser desorption/ionization (MALDI) for the analysis of trace drugs and explosives (G22, G23). A group at Argonne National Laboratories has also investigated the potential of compact TOF with a photoionization source for nonproliferation applications (G24, G25). A TOF mass spectrometer for cometary analysis is presented (G26), and the inherent limitations on analyzer volume for an interplanetary vehicle require a unique design approach. A membrane-inlet TOF system the size of a suitcase has been developed for environmental applications (G27). Several TOF instruments of various dimensions (“benchtop” to “lunch box” size) that serve as rapid-scanning detector for fastGC has been described by commercial vendors (G28-G30). Although the highest volume of published work on miniature instrumentation involved the TOF analyzer, almost every mass spectromet technology has been investigated as a candidate for miniaturization. A portable magnetic sector instrument with a volume of ∼1.25 ft3 and a weight of 20 kg has been developed (G31). A Wein filter/magnetic sector double-focusing instrument with a main path radius of 56 mm has been described (G32). Microfabrication techniques will no doubt appear frequently in the future for the construction of mass spectrometer systems, and this approach has been used to fabricate a microscale quadrupole (G33). Several publications have described a miniature quadrupole array, which, due to its small size, has the unique ability to operate efficiently at pressures in the range of 10-2 Torr (G34-G36).

Numerous investigators have evaluated the potential applicability of small, low-field Fourier transform ion cyclotron resonance (FTICR) mass spectrometers (G37-G40), and one article evaluated the potential use of ion traps for ICR-type experiments without magnets (G41). Cylindrical quadrupole ion traps have also been constructed and characterized in miniature form (G42, G43). Although not described as a process instrument, a unique triplequadrupole instrument has been described that has dimensions similar to conventional benchtop instruments (G44). As such, this device could readily be packaged as a field-deployable or process instrument. Fieldable Instrumentation. As instrumentation decreases in size, mass spectrometry is being used more frequently outside the laboratory environment. An indication of the growth of this market is the increase in the number of commercial vendors providing turnkey portable systems for field mass spectrometry and GC/MS. Nevertheless, the majority of the published work on field-portable mass spectrometers was on systems that were developed in-house, on modified commercial benchtop systems. Wise and Guerin et al. have authored two reviews on direct sampling mass spectrometry, with a focus on the use of an ion trap mass spectrometer for environmental analysis (G45, G46). Another such system is an automatic, trailer-deployed integrated laboratory for remotely measuring VOCs in the field, described by Daughtrey et al. (G47), The process of developing and evaluating a portable GC/MS-based field test method is explained (G48), and this approach is contrasted with numerous iterations of conventional method 301 validation tests. A similar approach is described in the development of performance-based field methods for the analysis of substituted phenols using GC/MS, as compared to standard U.S. EPA methods (G49). A feasibility study of field-deployed GC/MS for industrial hygiene applications was conducted by NIOSH (G50). Field evaluation of fieldable GC/ MS instruments illustrated the capabilities and limitations of the technology and provided examples of the application of this technology (G51-G53). The evaluation of direct sampling ion trap MS has also been performed (G54). The modular design of the inlet systems for the commercially available integrated instruments has been described, with a description of performance against the design criteria (G55-G58). A description of the design criteria themselves, and their rationale for portable GC/MS, is detailed (G59). A summary of recent work at Los Alamos National Laboratories with the ion trap as a tool for complex environmental analysis was presented by Cisper and Hemberger (G60, G61). Method development on portable instrumentation for emergency responders (fires and chemical accidents) was derived as a first step toward development of standard operating procedures (G62, G63). The development of an ion trap/TOF instrument is detailed, with potential applications to monitoring air quality (G64, G65). The description of ion traps as chemical sensors is presented by Hart (G66). Special configurations are described for the target analyses of actinides (G67, G68), surface contaminants (G69), and drugs of abuse (G70). Inlet Technology. Since mass spectrometers require vacuum to operate, one of the major areas of focus has been sample introduction. For a great number of applications, sample can be introduced to the mass spectrometer source through simple methods, such as direct capillary introduction or membrane inlets. Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


Investigation of membrane configurations and materials comprises the most significant area of published research in this field. The following citations are those addressing membrane fundamentals, as opposed to those articles with an application focus (found in the subsequent section). Several authors have recently reviewed developments in membrane inlet mass spectrometry (MIMS) (G71-G74). Membrane inlets are included in a discussion in recent developments using ion trap mass spectrometry (G75). Three different membrane probes were compared for dissolved gases (G76). The construction and evaluation of a membrane inlet on a quadrupole system is detailed (G77), as is a miniaturized membrane inlet system for biological applications (G78). Temperature-programmed desorption from a membrane substrate is investigated as a way to sharpen analyte temporal profiles (G79, G80). This work followed on earlier investigations, where the liquid sample flow could be periodically stopped, with a resulting heating of the membrane by the electron beam in the ion source (G81). The use of the membrane material to trap-and-release analytes is suggested by Lauritsen and Ketola (G82). Modulating sample flow across the membrane surface, and using the temporal nature of different response functions for different analytes, have been suggested and investigated (G83-G85). Sample matrixes can be purged with a gas, and the analyte-enriched vapor passed across a membrane in a technique called “purge-and-membrane” mass spectrometry (G86). A comparison of MIMS to purge-and-trap GC/MS and static headspace GC is detailed (G87). A parametric evaluation of experimental variables has been performed (G88), and the resultant effect on signal is described. A dual-membrane configuration, with concentric silicone and Nafion tubes, is described (G89), and the performance of a unique zeolite membrane is detailed (G90). Another novel membrane composition is a plasma-deposited polydimethylsilicone layer on top of a hollow microporous polypropylene fiber (G91). Membranes from copolymers (acrylonitrile/butadiene, styrene/butadiene) were evaluated in parallel with a silicone device (G92). Maden and Hayward (G93, G94) have delineated a number of other prospective materials for use as membranes. An evaluation and characterization of a number of different materials as sheet pervaporation membranes for flow injection analysis and MIMS has been performed (G95). Liquid-based membranes have also been evaluated, with the liquid matrix supported on a microporous substrate (G96-G98). The solution dynamics of a flow injection system for MIMS have been evaluated and optimized (G99). Different modes of operation for scanning an ion trap and the resultant effect on spectral quality are described (G100), and these authors also looked at principal component analysis and partial least-squares regression to extract information from membraneinlet MS data (G101). Other configurations of MIMS include a plug-in sterilizable MIMS device for fermentors (G102). Chemically modified membranes are described that enhance selectivity, through enhanced membrane interaction to specific functional groups (G103). The combination of a MIMS inlet and a jet separator for enhanced sensitivity is presented (G104). Another approach to enhanced sensitivity is obtained by coupling a MIMS system with a cold trap (G105). The construction of a membrane probe, and the resultant limitations for its use in a fermentation process, are described (G106). The modifications that are required 144R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

to a convert a common benchtop mass spectrometer to allow direct insertion of a membrane probe have been presented (G107). A miniprobe device, designed for mapping analyte concentrations with a spatial resolution on the order of 1 mm is illustrated (G108, G109). A caveat about time- and concentration-dependent changes in relative peak intensities is delivered (G110). Ionization Modes. Another area of research focus has been in novel modes of ionization. Although electron impact (EI) has been used in the vast majority of applications, in large part due to the ability to perform searches on libraries of EI spectra, unique applications take advantage of heightened sensitivity and selectivity afforded by other ionization modes. A brief description of several ionization techniques has appeared, listing the respective benefits and drawbacks for each ionization technique in the direct analysis of organics in air (G111). A single high-resolution quadrupole mass spectrometer, operated under different EI source conditions, can be used for conventional analysis (up to 128 amu) or for high-resolution low-mass analysis, to resolve helium (mass of 4.0026 amu) and deuterium (mass of 4.0282 amu) (G112). Proton-transfer ionization mass spectrometry (PTR-MS), has been an active area for research, due to sensitivities in the range of parts per trillion by volume (pptv). This ionization source has been used for trace level impurities in semiconductor manufacture (G113) and has recently been modified to detect trace level nitrogen in argon with the addition of purified hydrogen into the ion source (G114). Lindinger and various coauthors have numerous publications describing this ionization source (G115-G119), which, in their applications, uses gas-phase reactions of H3O+ ions and the analytes (commonly volatile organic compounds) that possess higher proton affinities. Another source used for direct ionization of trace level organic materials in air is resonanceenhanced multiphoton ionization (REMPI) (G120-G123). Two reviews of laser ionization techniques (including REMPI) have also been written (G124, G125). Several permutations for alignment of the photon and molecular beams were investigated to determine optimum conditions for ion production in a laser source (G126). Hyperthermal surface ionization (HSI) has recently been described (G127-G129), and it involves the collision of a molecular beam at supersonic velocities with a heated rhenium foil. The molecular beam, which obtains the supersonic velocities as it expands into vacuum, can also be subjected to normal electron ionization, providing complementary information. The development of a thermal ionization cavity source has been described (G130-G132), culminating in the design of a miniature source (G133), which could be readily coupled with a benchtop instrument for portable analysis. A microwave plasma ionization source has been coupled with an ion trap mass spectrometer, resulting in an instrument that has the potential for isotopic analysis in the field (G134). The instrument performance was evaluated on xenon and krypton isotope measurements from air. Microwave plasma sources in general have been the topic of a recent review (G135). Nanoelectrospray has been used to provide ionization in a study that monitored protein chemical reactions in real time (G136). A low-pressure electrospray source has been described (G137) that would significantly reduce gas load on the vacuum system, facilitating the manufacture of smaller systems. Several sources (other then EI) have been investigated in conjunction with membrane inlets. Examples of such references

include the combination of membranes and REMPI for the analysis of polyaromatic hydrocarbons in water (G138), the design of a desorption chemical ionization source for MIMS (G139), the coupling of MIMS and charge-exchange ionization (using copermeated oxygen as the charge-exchange donor) (G140), and the construction of a MIMS system using a Penning ion source (G141). Process Monitoring and Control. The following example citations will hopefully demonstrate the broad scope of potential applications for process mass spectrometry. As will be demonstrated, with the appropriate inlets and ionization, mass spectrometry performs quite credibly in the analysis of diverse groups of molecules, with an impressive combination of specificity and selectivity. Semiconductor fabrication has been a fertile application base for process mass spectrometry. A review article discusses the use of time-of-flight-secondary ion mass spectrometry (TOFSIMS) to control various processes in the fabrication of semiconductor devices (G142). Mass spectrometry is compared to optical emission spectroscopy and an rf probe for feedback control in reactive ion etching (G143). A mass spectrometer has been used to monitor selectivity of the etch rate in a plasma fabrication system (G144), and a system is described for monitoring end point uniformity in silicon dioxide plasma etching (G145). As a diagnostic device for the etching of silicon with a fluorine beam, an in situ mass spectrometer has provided insight into the surface chemistry (G146). In a similar manner, mass spectral data were used in conjunction with data from an optical spectroscopy to study a chlorine plasma etching process of silicon (G147). Mass spectrometric analysis has been used to measure vapor composition in real time for the control of molecular beam epitaxy (G148, G149). The kinetics of hydrogen abstraction and etching reactions are derived through the use of a real-time analysis via mass spectrometry (G150). A plasma emission monitor (PEM) system has been constructed, using both mass spectrometry and an optical emission spectrometer, for the closed-loop control of a reactive sputter deposition unit that deposits indium tin oxide onto a polymer substrate (G151). The fundamental reactions in silicon chemical vapor deposition (CVD) were investigated with a quadrupole mass spectrometer (G152, G153), and a real-time mass spectrometer was used to infer the dynamic deposition rate for process control in a rapid thermal CVD process (G154). A discussion of potential applications of in situ mass spectrometry to process sensing and metrology is derived from examples in plasma-enhanced CVD of silicon (G155). A discussion is presented on real-time feedback control of a plasma-enhanced CVD process, based on the analysis of data provided by a quadrupole mass spectrometer (G156-G158). The role of mass spectrometry in controlling plasma-enhanced CVD and the resultant benefits to reproducible manufacturing have been addressed (G159). A novel particle beam mass spectrometer has been developed that can be used for sizing and characterizing ultrafine particles in lowpressure environments, such as in semiconductor processing equipment (G160). As mentioned previously, numerous researchers have evaluated membrane inlets as an efficient way to sample process streams into the vacuum that a mass spectrometer requires. There are at least as many applications for membrane inlets as there are designs for inlet systems. There have been a number of

reviews on the use of membrane inlets for mass spectrometers (G161-G165). Few authors describe true feedback control based on MIMS, but examples of such systems have been described for the control of penicillin fermentation (G166) and for glucose fermentation with a yeast strain that has been genetically engineered to provide high tolerance to ethanol (G167, G168). A MIMS system was demonstrated in a pilot plant-scale (9000 L) fermentation reactor and demonstrated both comparable performance and acceptable agreement with analyses performed offline using HPLC (G169). The monitoring of fermentation and other bioconversion processes by MIMS has been extensively described, including articles studying the metabolism of the fungus Bjerkandera adusta in the production of chlorinated methoxybenzaldehydes (G170, G171) and other halogenated organic compounds (G172), the description of an on-line GC/ MS system that is amenable to process control for potentially toxic organic gases (G173), a system for the control of an anaerobic bioreactor for wastewater treatment (G174), a fermentation study that illustrated detection of nonvolatile species by MIMS (G175), a kinetics investigation of the bacterial oxidation of cis-1,2dichloroethylene in a biofilm reactor (G176), and a comparison of ion trap and hybrid mass spectrometers in a study on the degradation of a herbicide in biofilm (G177). A special case of membrane usage is on-line microdialysis, frequently coupled with microelectrospray or atmospheric pressure ionization, to monitor bioavailability and drug metabolism. Such studies can be performed either in vitro (G178) or in vivo (G179-G181). Frequently MS/MS techniques are required, due to the complex chemistry in living systems (G182-G184). Microdialysis has also been used simply as a cleanup step for more conventional analysis on biological systems (G185-G187), with investigation of novel approaches, including microfabrication techniques (G188) and dual-microdialysis systems (G189). Membrane introduction has broad applicability outside of biological systems. Aqueous systems with organic impurities are particularly well-suited to membrane inlets, such as the analysis of terpenes in water streams (G190), the analysis of volatile organic compounds in seawater (G191), and the quantification of phenolics in water (G192). Dynamic studies of the photolysis of benzyl acetate in a water matrix (G193), copper-catalyzed reduction of nitric oxide by ammonia in aqueous solutions (G194), and the acid-catalyzed ring opening of epichlorohydrin in water (G195, G196) have all been accomplished using MIMS. Other diverse examples of MIMS for process monitoring include studies on the diurnal shifts in gas emissions from peat bogs (G197, G198), the analysis of oxidation products arising from the pyrolysis of several peptides (G199), denitrification studies on estuarine sediments (G200-G202), and temporal studies on components in automobile exhaust in a test stand, a dynamometer, and real traffic conditions (G203). Applications of process mass spectrometry without MIMS are similarly diverse. A number of “traditional” applications of process mass spectrometry have been described recently, including control of ethylene furnaces (G204), the measurement and control of a vacuum-dryer (G205), monitoring solvents, monomers, and fixed gases in polyethylene manufacture (G206, G207), controlling the manufacturing of steel (G208), as an early detection system for runaway reactions (G209), and as continuous emission Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


monitors (CEMs) for thermal oxidizers and incinerators (G210G213). A process mass spectrometer configuration that has had components configured for use as a flexible diagnostic device for understanding process chemistries has been described (G214). In many cases, special data treatment is required to extract the maximum information from the mass spectrometer signal. Such applications range from the investigation of least-squares methods for speciation of isomers and analogues (G215, G216), to the development of PLS tools for evaluation of predictive models (G217), to the deconvolution of composite mass spectra using artificial neural networks (G218). An application based on PLS has also been described for the analysis of chemical or physical properties of complex mixtures, rather than single-component analysis. Properties that could be predicted included octane numbers, cloud point, density, and viscosity (G219). To analyze and control thermal combustion processes, REMPI has been used to measure both simple aromatics (benzene, toluene, xylenes) (G220, G221) and polychlorinated dibenzodioxins and polychlorinated dibenzofurans (PCDD/PCDFs) (G222, G223). In a similar approach, laser ionization mass spectrometry has been used on vehicle exhaust systems to monitor combustion efficiency and combustion products (G224, G225), as well as the determination of catalytic converter efficiency in real time (G226). Combustion products arising from burning a mixture of coal and various biomass materials in a quartz reactor were monitored with an on-line molecular beam mass spectrometer (G227). The same system was used to investigate the production of hydrogen from pyrolysis of biomass, followed by catalytic reforming of the pyrolysis oils (G228, G229). That system has been modified into a transportable system, suitable for field trials (G230). Other recent articles that involve the use of on-line GC/MS to analyze reactor processes real time include a description of the design and construction of an apparatus for monitoring high-temperature and high-pressure conversion reactions (G231) with example applications, studies on the catalytic degradation of high-density polyethylene (G232, G233), and an investigation of the thermal decomposition of wood and cellulose in the presence of water and methanol vapors (G234). Mass spectrometry has always been useful for kinetics studies, since data acquisition rates can be as high as several points per second. Examples of such experiments include a study of the kinetics of corrosion of metals in water, based on the measurement of oxygen consumption and hydrogen generation (G235), fundamental studies on reactions of elementary importance to atmospheric chemistry (G236), decomposition kinetics in a novel reactor design that uses counterflow jets, rather than a heated vessel (G237), and evaluation of a carbothermic reduction reaction in a high-vacuum chamber with a dedicated mass spectrometer (G238). Enzyme kinetics have also been monitored in real time with mass spectrometry but required the use of HPLC (G239). Mass spectrometry is discussed extensively in a detailed review on transient methods in heterogeneous catalysis (G240). A liquid-phase organic synthesis following the stepwise esterification of Boc-Gly-OH with poly(ethylene glycol)s has been used as a model system to demonstrate the use of electrospray mass spectrometry for reaction monitoring (G241). MALDI mass spectrometry has been used to follow organic reactions in real time, directly from the polymeric supports used in solid-phase 146R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

synthesis (G242). FIA coupled with detection based on mass spectrometry operated with an APCI source has shown promise for following reaction completion on pharmaceutical chemistries (G243, G244). As mentioned in the introduction, mass spectrometry can provide detailed component analysis on respired air for point-ofcare diagnostics. A number of research groups are taking different tacks toward this problem, including GC/MS with conventional EI (G245, G246) and MS/MS with low-pressure chemical ionization (CI) (G247) and with a glow discharge ionization source (G248, G249). Different classes of compounds in respired air have been investigated with PTR-MS, including acetonitrile and benzene as model compounds for passive smoking (G250, G251), sulfides and acetone following the consumption of garlic (G252), and propanol (G253). Environmental Monitoring. Mass spectrometry has been used successfully for some time for the analysis of volatile organic compounds in air. A multimembrane inlet ion trap system for isoprene and other transient alkenes in the atmosphere using selective CI has been described (G254). On-line membrane sampling has been compared to analysis of samples collected on a Tenax trap (G255). Membrane inlets were also employed in a study that used glow discharge ionization and MS/MS for VOCs (G256). An evaluation of MIMS for volatile organic sulfur compounds in air gave detection limits in the ppb/ppt range (G257). Trace level detection of both chlorofluorocarbons (CFCs) (G258, G259), regulated by the Montreal protocol, and their replacements, hydrofluorocarbons (HFCs) and hydrochlorofluorocarbons (HCFCs) (G260), have been performed with fieldportable instrumentation. A roving GC/MS system was described that obtained VOC gradients with respect to both space and time (G261), while another system used an quadrupole ion storage-TOF system for the same purpose (G262). Monitoring of VOCs (G263) and air quality (G264) has also been done on the MIR space station with mass spectrometry. Sensitivities on the order of 50 parts per trillion by volume (pptv) for common VOCs were obtained with a PTRMS system (G265). Volatile aromatic compounds could be sampled from ambient air with sensitivities in the low part per billion range (ppb) with REMPI (G266), and the system was fieldtested on a small cart (G267). Simple aromatics were also analyzed on a direct sampling ion trap system, with sensitivities in the hundreds of ppb (G268). Simple aromatics and polycyclic aromatic compounds have been measured in automobile exhaust, using a membrane inlet (G269), and automobile exhaust was the matrix for the measurement of oxygenates, using laser ionization (G270). The analysis of semivolatiles in air was facilitated with the use of a MIMS inlet (G271, G272). Membrane inlets also play a crucial role in the analysis of drinking water, surface water, and groundwater samples with mass spectrometry. The U.S. Environmental Protection Agency (EPA) has been moving toward the adoption of performance-based methods, which would allow for the introduction of a number of new methods, spurring the investigation of MIMS applications. Residual levels of disinfection byproducts (DBPs) were investigated by MIMS, and the results compared to the validated analysis by EPA method 524.2 (G273). A portable system has been described that could rapidly measure total trihalomethane in

drinking water (G274). In-field analysis of VOCs (G275) and SVOCs (G276) with membrane inlets showed that ppb detection levels can be obtained. Membrane inlets have also been demonstrated to facilitate the measurement of low-level volatile organic sulfur compounds in water samples (G277) and the continuous monitoring of dissolved gas concentrations in groundwater (G278). Water quality can be rapidly assayed with MIMS in industrial wastewater samples (G279). Migration of hazardous waste from hazardous waste sites has been monitored using an in situ sparging system (G280), to assay the distribution of various compound classes in the water table. Near real-time determination of the spatial distribution could also aid in remediation activities, saving valuable time and money (G281). A system that can collect data in real time at depths of up to 200 ft below the water table has been described to provide depth profiles of VOCs (G282, G283). A compact cold trap-membrane inlet system has been developed to obtain analysis of a broad range of VOCs in water at the low-ppt level (G284). Direct field analysis of soils for assessment of contamination has also been described (G285). Analysis has been based on direct insertion probes, for PAH screening (G286), and to assess contamination from fuel (G287). Purge-and-membrane (G288) or headspace-MIMS (G289) approaches have also been investigated for enhanced detection limits. Reviews on the analysis of explosives (G290, G291) and chemical warfare agents (G292) have dealt extensively with detection using portable mass spectrometry. It should be obvious what drives the development of technology for this application, and the molecular specificity and sensitivity of mass spectrometry proves to be a good match with the design criteria. Systems based on quadrupole (G293) and ion traps (G294, G295) have been described in the literature. A field portable version has been developed as a “generic detector” for the Chemical Weapons Convention Treaty compounds list (G296, G297). The capabilities of MS/MS are a claimed to afford a significant advantage for the acute sensitivity required in highly variable matrixes (G298). A different approach to selectivity is the use of ultrashort laser pulses using REMPI, to overcome difficulties in spectral interpretation arising from competitive pathways by conventional REMPI (G299). Preliminary work on field detection of bacterial pathogens, based on MALDI analysis in a laboratory, shows promise, and future work will be directed at a fieldable version (G300). Aerosol and Particulate Monitoring. Another application area that has been extensively reviewed is the analysis of aerosols and particulates (G301-G304). In large part spurred by recent EPA amendments regulating air-borne particulate matter, as well as potential applications to assess air-borne spores and bacteria, numerous researchers have made this an active area of research, with the majority of the publications in this area arising from conference proceedings. Potential applications of this technology range from emission monitoring of automobile exhaust particles (G305) to tracking uranium in airborne particles at nuclear production facilities (G306). Almost all the instruments cited herein have a drift region through which the particles migrate. Following characterization of the particle count and size, usually based on a light-scattering technique, the particles, or at least some portion thereof, are ionized, most commonly with a laser pulse.

At the present time, the majority of systems described in the literature are based on mass analysis by TOF (G307), but the use of ion trap makes possible MS/MS experimentation, which can help address matrix interferences (G308), particularly in the analysis of organic materials (G309, G310). Transportable instrumentation has been developed for tropospheric aerosol measurements aboard a research aircraft (G311), as well as for the tracking of spatial distributions of air-borne pollutants and emissions (G312, G313). Example applications include the characterization of a particulate source, based on unique spectral features (G314, G315), tracking of diurnal patterns in the chemical composition of particles (G316), the radiocarbon “dating” of individual trace organic PAHs in atmospheric aerosol (G317), a comparison of the chemical nature of aqueous aerosol to dry aerosol (G318), and a real-time temporal study of pyrotechnically derived particles (G319). Particle analysis can also be used for fundamental reaction studies. Well-characterized particles can be directed through a reaction chamber, for subsequent analysis that probes the interactions between reagent gases and the particle surface (G320, G321). There have also been studies on individual plasma particles in a semiconductor manufacturing operation (G322). Another area of recent interest has been the speciation of bioaerosols. In one study, prominent peaks were observed for all the examined microorganisms, suggesting facile identification of biological agents (G323). Computer algorithms (G324) and pattern recognition software (G325) are currently being investigated for speciating these materials. PROCESS CHEMOMETRICS Chemometrics is often defined as the use of multivariate data analysis and mathematical tools to extract information from chemical data. Historically, chemometrics has found application in the areas of pattern recognition, classification, signal resolution, and instrument calibration. Modern chemical processes are capable of generating and processing huge amounts of process data, especially with the increased use of chemical analyzers and instrumentation. The application, or development, of chemometric tools to this wealth of process data is termed “process chemometrics” and seeks to provide additional insights into the chemical process through monitoring, modeling, and control. In this review, we tried to limit references to actual process applications that highlight a particular chemometrics technique or new application area. Applications where the chemometrics aspect is secondary to the main thrust or the work (e.g., routine calibration in spectroscopy) are presumably covered in their respective measurement sections. Regrettably, the limited space in this review also forces omission of some fields, such as food products and drug design, that are on the periphery of what is considered chemical processing yet contain many significant applications of chemometrics. Likewise, several topics may not receive the depth of coverage they deserve in order to allow more breadth. In assembling this review, it became very clear that chemometrics has matured and gained acceptance within the process modeling and control communities. This is probably due to the combined influence of the increasingly data intensive and multivariate nature of their needs and the increased depth of understanding and sophistication of the chemometric tools that are available (H1). Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


Evidence of these trends can be seen in the synergistic use of “soft” modeling methods combined with fundamental approaches in process control and the attention focused on development of more robust chemometric tools. These are the themes that drive the selection of works included in this review. General Information. Since the last review there have been several review or overview papers in process chemometrics. Wise and Gallagher presented a review of the use of chemometrics in chemical process monitoring and fault detection. Topics discussed include a definition of chemometric techniques, theory and applications of PCA, multivariate statistical process control (MSPC), multiway PCA, evolving factor analysis, multivariate curve resolution, multivariate regression modeling, inverse least squares, PLS, PCR, ridge regression (RR), PLS for regression-adjusted variables, multiblock PLS, and multiway PLS (H2). Wise and Kowalski presented a tutorial and review on commonly used chemometric methods illustrated with some example problems in chemical process monitoring, process instrument calibration, and dynamic process modeling (H3). Schonkopf presented a review of the fundamentals, types, and performance of multivariate modeling methods for industrial process applications (H4). Davis et al. provide a general discussion of the current state of the art in process monitoring, data analysis, and data interpretation and a road map for future development of process-monitoring systems. By taking a pattern recognition viewpoint where data analysis and interpretation are considered as feature extraction and label assignment steps, respectively, they provide a unifying perspective that reduces the large number of apparently disparate methods into a small set of categories defined by key performance characteristics (H5). Miller described seven discrete procedures needed to successfully implement a process spectroscopy method: (i) gross outlier removal; (ii) data filtering/prelinearization; (iii) sample selection; (iv) variable selection; (v) calibration; (vi) on-line method fault detection; and (vii) calibration updating/ maintenance. Each of these steps was discussed in terms of available chemometric tools and examples from a process spectrometer installed on a batch operated polymer production unit (H6). Gurden and others described how chemometric techniques have been used in a pilot plant environment with the objective of increasing the general understanding of the process. MSPC techniques are used to follow the operation of the plant and for detection and diagnosis of process disturbances. The effect of process conditions on product quality is analyzed using crosscorrelation with latent variables, and significant process variables and time delay structures are identified. They conclude that the experience and process understanding gained by the pilot plant staff has enabled them to propose the installation of new sensors and analyzers in the main process based upon sound business benefits (H7). For reviews of chemometrics applied to spectroscopy, see Duckworth (H8), Geladi (H9), the three-part series by Workman et al. (H10-H12), and other sections of this review. Workman and Brown present a review related to an ASTM practice for quantitative IR (near- and mid-IR) analyses based on the use of chemometrics (H13). Ciurczak presented a review on validation of spectroscopic methods in pharmaceutical analysis and pointed out that validation of spectroscopic methods, especially for inprocess analyses, is problematic due to the difficulty of obtaining 148R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

applicable samples. The FDA has offered some instrumental guidelines for near-IR spectrometers used in the diffuse reflectance mode (H14). Workman et al. presented some requirements for chemometrics in process spectrometers (H15). Lombard presented some statistical methods for assessing the precision and bias of an analyzer (H16) with discussion presented by Calvin (H17) and Lyman (H18). Brimmer discusses some important issues related to NIR method development in the chemical/ polymer industry that are ubiquitous to the entire process analytical field. He points out that process analysts who maintain the instrumentation and calibrations are generally not experts in spectroscopy, or chemometrics, and therefore the methods developed must also be easy to maintain. Considerations for developing methods that are insensitive to the short- and longterm process variations are discussed. Different calibration methods are compared for systems that have changed over time to determine which methods are more stable under changing process conditions (H19). Much of the recent development in chemometrics has been driven by the industrial need for improved robustness of on-line chemometrics applications. Issues of calibration model transferability, uncertainty estimates of predicted results, sensitivity to process upsets or disturbances, sensor failure modes, and chemometric model updating have all been investigated. While most of these developments have few demonstrated applications using process data, they clearly will impact future applications. The reader should consult the two fundamental reviews of chemometrics that have appeared during this reporting period for more details (H20, H21). Two areas that have found applications (i.e., calibration transfer, or standardization, and variable selection to improve chemometric model robustness) are briefly reviewed next. Campbell discusses when to use chemometric NIR transfer steps and use of forward vs backward standardization. He recommends use of a backward transfer and suggests that it is good practice to use a transfer step even if one does not initially seem beneficial to help avoid future problems (H22). Dreassi et al. describe results of an investigation of the transfer of calibration in NIR spectrometry. Four approaches were tested: direct standardization, direct standardization combined with PCA, ordinary least squares, and slope/intercept correction (H23). Fearn et al. describe wavelength standardization in filter instruments to correct for differences between filters (H24). Herrero and Ortiz applied a piecewise direct standardization (PDS) method to transfer from one day to another the PLS models built in the polarographic determination of Cu, Pb, Cd, and Zn. Once the PLS models are built, the calibration transfer is carried out to overcome the instrumental change over time (H25). Bouveresse et al. presented an application of instrument standardization, where two different modules (internal measurement cell and external fiberoptic module) of a NIR spectrometer must be standardized for the quantitative determination of an active compound in pharmaceutical tablets (H26). Ozdemir et al. discuss using hybrid calibration models, in which spectra from each instrument were used simultaneously in the calibration, as an alternative to standardization. The application of PLS and genetic regression (GR) to the problem of generating these hybrid calibrations is presented. Calibration models were found that perform well on both instruments, even

when only a single spectrum from the second instrument was used during the calibration process (H27). Puigdomenech et al. present a novel approach to standardization using the hierarchical, or twoblock, PLS to allow the standardization between the spectra of forage measured in a network of different (tilting filter and scanning monochromator) NIR instruments. PLS models of the individual instruments (lower block) are combined in the upper block to build a PLS calibration equation for the network of instruments. The possibility is afforded for instruments of lower resolution to participate in the same network as higher resolution instruments (H28). Rutan et al. described the combination of experimental design and PCA to identify the main sources of variation in the spectra and to estimate their influence on the quantitative predictions by comparing variations in a set of measured, replicate spectra to spectra with simulated variations. The approach was applied to the hydroxyl number determination of polyols by NIR spectroscopy and PLS (H29). It is well-known that more parsimonious and robust models can result through a judicious selection of variables in multivariate calibration and that variable selection can overcome many of the shortcomings of MLR and make it competitive with PCR or PLS. Many different approaches to variable selection including iterative methods (H30, H31), genetic algorithms (H32-H35), permutation (H36, H37), Bayesian selection (H38), bootstrapping (H39), and expert opinion (H40) have been reported. Despagne and Massart discuss variable selection for neural networks (NN) (H41), McShane et al. discuss variable selection in the calibration of a NIR glucose sensor for cell culture medium monitoring (H42), and Swierenga et al. discuss robust wavelength selection as an alternative to standardization in PLS calibration. Variable selection by simulated annealing enhanced the model’s robustness with respect to model transfer and improved its predictive ability in a NIR application determining water in pharmaceutical tables (H43). Spiegelman and co-workers identified the mathematical basis of improved calibration through selection of informative variables for PLS. A theoretical investigation of calibration slopes indicates that including uninformative wavelengths negatively affects calibrations by producing both large relative bias toward zero and small additive bias away from the origin (H44). Process Modeling. In this review, process modeling is distinguished from process monitoring in that the first is mainly concerned with integrating multiple process data sources to assess the status, or “health”, of the process while the latter is mainly directed toward elucidating some particular process-relevant parameter through measurements. Within process modeling, most of the activity has been driven by the combination of multivariate techniques and the principals of statistical process control (SPC) to provide MSPC visualizations, typically as control charts, of process variation. A good overview of this methodology can be found in the tutorial presented by Kourti and MacGregor on multivariate projection methods (PCA, PLS, multiblock PLS, and multiway principal component analysis (MPCA)) to develop MSPC of continuous and batch processes. Applications to a catalytic cracking section of a large petrochemical refinery, monitoring and diagnosis of a continuous low-density polyethylene (LDPE) polymerization process, and monitoring of an industrial batch process are presented (H45). In a later paper, they extend the

MPCA approach to batch monitoring using multiway PLS (MPLS) by using the final quality variables of the product to extract more relevant information from the process measurement variable trajectories. Problems with on-line implementation (i.e., incomplete X block data until the end of the batch run) are handled by forward projection estimates of the missing data of a new evolving batch. Approximate confidence intervals are developed for the predictions from PLS models. The approach is illustrated using a simulation study of a styrene-butadiene batch reactor (H46). Finally, a review by the same group was presented on industrial applications of projection methods for MSPC (H47). Wikstrom et al. presented a two-paper series describing the application of MSPC and multivariate time series analysis to understand and model the dynamics of an electrolytic process manufacturing extremely pure Cu. They discussed how the dynamics of the process could be incorporated into the latent variables (PCs) using multivariate time series analysis (i.e., PLS modeling of the lagged latent variables). Multivariate Shewhart, cumulative sum (CUSUM) and exponentially weighted moving average (EWMA) control charts on the PCA or PLS results are used and compared (H48, H49). Hartnett and others discussed the techniques of PCA and PLS as multivariate methods for modeling process plants. Two case studies with data from a continuous stirred tank reactor (CSTR) simulation and a LDPE reactor simulation exemplify these concepts (H50). Zullo presented an application of MSPC using multivariate CUMSUM plots of PCs to a multicomponent distillation unit. He found that the method allowed the engineer to identify whether the plant was operating at the steady state and, possibly, at which steady state among the known ones. Deviations from steady state were easily detected as well as dynamic transients between different steady states (H51). Thomas et al. describe several PCA models developed for the depropanizer of a heavy gasoline-FCC unit (H52). Morud describes an application of multivariate statistical methods with the aim to improve the production of titanium dioxide (H53). Costa and co-workers describe how the analysis and statistical treatment of process data resulted in a better understanding of a pulp and paper plant and identification of major variables influencing its behavior (H54). Gallagher and others presented an extension of MSPC techniques for batch processes to monitoring a semibatch process by focusing on periodic process set point changes and continuous processes that have repeated upsets or perturbations. The MSPC technique was demonstrated for a nuclear waste storage tank that undergoes periodic agitation from a mixing pump (H55). Neogi and Schlags applied MPCA and MPLS to an industrial emulsion polymerization batch process for batch analysis, process monitoring, fault diagnosis, product quality prediction, and improved process insight. One key feature of their work is that reaction extent was used as the common reference scale to align batches with varying time duration. They found that MPCA/MPLS technologies (i) detected potential process abnormalities, (ii) determined the time an abnormal event occurred, and (iii) indicated the likely variable or variables that caused the abnormality (H56). Albert and co-workers discussed MSPC applications for monitoring batch behavior in a methyl methacrylate polymerization reactor and an industrial fed-batch fermentation process. Multiway PCR was used for predicting key quality variables that Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


are not measured on-line or monitored at a much slower frequency than the process variables (H57, H58). Ku et al. discussed including dynamic behavior in the PCA model, using the “time lag shift” method, and tested the approach with the Tennessee Eastman process simulation (H59). Bharati and MacGregor describe using MPCA for the extraction of information from online imaging sensors to detect and locate features for statistical process control or feedback control of spatial processes (H60). Wold and co-workers presented a review of PLS and PC modeling, with emphasis on the selection of variables. An alternative to reduction of variables is to divide the variables into conceptually meaningful blocks and to apply hierarchical multiblock PLS, or PC, models (H61). The approach was illustrated for data taken from a petroleum residue catalytic cracking unit. Rannar and co-workers used this method to monitor industrial batch polymerization processes, overcoming the need for estimating the unknown part of the process variable trajectory deviations from the current time until the end of the batch. The hierarchical multiblock approach is compared to multiway PCA/PLS approaches and shown to have significant benefits when multistage batch processes are monitored, where the latent variable structure can change at several points during the batch (H62). Westerhuis recently compared variants of multiblock and hierarchical PCA and PLS methods from a theoretical and algorithmic viewpoint and illustrate their differences with case studies (H63). Zhang et al. point out that, for highly nonlinear processes, conventional MSPC monitoring may not be efficient and suggest the use of nonlinear PCA with the use of accumulated scores plots. An application to the condition monitoring of a polymerization reactor demonstrates the effectiveness of the nonlinear monitoring approach (H64). Bakshi and Utojo provide a unified framework for combining empirical modeling methods that combine inputs by linear projection (including linear methods such as ordinary least squares (OLS), PLS, and PCR and nonlinear methods such as back-propagation networks with a single hidden layer, projection pursuit regression, nonlinear PLS, and nonlinear PCR) into a single method called nonlinear continuum regression (NLCR). The modeling ability of NLCR and its performance are illustrated with simulations and industrial data (H65). Wold et al. describe an alternative approach to batch modeling based on a different unfolding of the three-way data matrix obtained from multiple batch runs than that used by Nomikos and MacGregor. This approach is based on an initial PLS analysis of the unfolded matrix ((batch × time) × variables) with “local time” used as a single y-variable. This is followed by a simple statistical analysis of the resulting scores and results in multivariate control charts suitable for monitoring the kinetics of new experiments or batches. The techniques are illustrated by an industrial example of yeast production (H66). Finally, Rius et al. considered analytical systems as process themselves and applied process modeling techniques (i.e., control charts, time series models, and recurrent neural networks) to study their reliability over time (H67). Nijhuis et al. followed similar reasoning in applying the principals of MSPC to the modeling of variations in chromatographic systems (H68). Negiz and Cinar describe approaches for sensor auditing (assessing the correctness of information generated by a sensor) using additional relevant process information (i.e., functional redundancy) to detect 150R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

multiple sensor abnormalities. Where the successive measurements are not autocorrelated, a multipass PLS algorithm can be used. For process measurements that are strongly correlated in time, the multipass PLS algorithm is modified by replacing the PLS models with canonical variate (CV) state space models to generate the functional redundancy in terms of model residuals. Results from a milk pasteurization pilot plant are used to illustrate the applicability of both methods (H69). Process Monitoring. This section is devoted to illustrations of interesting applications of chemometrics to process measurement data. Realizing the explosive growth in this area, this reviewer is relying on the measurement sections of this review, as well as other reviews in this issue, for “routine” applications. One area that uniquely relies on chemometrics is the development of inferential, or soft, sensors (a modeling approach to estimate hard-to-measure process variables from other easy-to-measure, online sensors). Since many sensors are used as input variables to estimate the output, the probability that one of the sensors fails increases significantly. A self-validating soft-sensor approach, based on PCA for fault identification and reconstruction, is described by Qin et al. and applied to air emission monitoring (H70). Several workers described a soft sensor to estimate dynamic particle size distribution in the hydrocyclone overflow of a grinding circuit. Du and co-workers used a simplified NN with PCA for data reduction to simplify the neural model structure (H71), while Casali and co-workers used an ARMAX model determined using stepwise regression (H72, H73). The petrochemical industry was an early adapter of chemometrics methods tied to monitoring applications. Andrade et al. discuss and review several advantages and drawbacks encountered in using FT-IR spectroscopy and PLS regression in industrial facilities (exemplified with mainly petrochemical applications). Typical drawbacks cited are selection of data pretreatment, errors in reference methods, selection of calibration and validation sets, and model aging (H74). Rest and others describe a cryostat cell constructed to enable the in situ liquefaction of methane, ethane, and natural gas from the mains. This cell was used to study the NIR spectra of methane, ethane, propane, n-butane, n-hexane, n-heptane, liquefied mains gas, and bulk storage liquid natural gas (LNG). Band positions and their extinction coefficients corresponding to overtone and combination bands were obtained under standard conditions. They concluded that the concentrations of the longer chain alkanes are so small, and their spectra so similar, that it would be impossible to determine their individual concentrations in the LNG samples without the aid of chemometrics (H75). Helland described the use of NIR spectroscopy for on-line determination of hydrocarbon gases at a petrochemical plant in a mixture of ethane, ethene, propane, and propene. The aim of the work was partly to find the optimal pretreatment and correction of the spectroscopic data and updating of the multivariate regression models to improve and correct the models over time (H76). The application of on-line measurements and chemometrics to predict properties of gasoline and other fuels continues to receive attention. Flecher and co-workers describe the use of dispersive fiber-optic Raman spectroscopy to remotely analyze fuels of varying composition for pump octane number, motor octane number (MON), research octane number (RON), and Reid

vapor pressure (RVP) using PLS and several preprocessing steps (H77). Cooper et al. describe a low-cost dispersive Raman instrument used to analyze mock petroleum samples (especially gasoline) that contain high BTEX concentrations. PLS was used to correlate the individual xylene isomer concentrations to the Raman spectral signal without the use of an internal standard, and leverage plots were used to identify Raman spectra that involve diode laser mode hops or significant fiber backscatter (H78). Garrigues et al. investigated predicting seven aircraft fuel quality properties from FT-IR spectra, using MLR (stepwise and full), PCR (full spectrum and stepwise), and PLS. In comparing the chemometric methods, the SEP, repeatability, and reproducibility were considered (H79). Litani-Barzilai et al. describe an approach to the on-line prediction of 10 gasoline properties by combining on-line information from a SW-NIR photodiode-array spectrometer with LIF spectra yielding improved octane number prediction. These two setups (NIR and LIF) can be combined into one integrated system, based on common optical fibers and detectors (H80). Muller describes the application of AOTF NIR spectroscopy to predict multiple gasoline properties. He discusses which probes and sampling systems give good results and chemometric systems providing robust transferable calibrations (H81). Workman investigated multiple chemometric techniques to optimize octane prediction using third overtone NIR spectral data. Algorithms tested include linear PCR and PLS and nonlinear forms of PCR and PLS, as well as locally weighted regression with PCR (LWR2) and locally weighted regression with PLS (LWR3). He concluded that using nonlinear techniques for modeling the octane numbers from SW-NIR spectra could improve the analysis performance and resistance of the model to noise in the instrument response and reference data (H82). Sikora and Salacki applied NIR spectroscopy and chemometrics for simultaneous estimation of the density, viscosity, cetane number index, and freezing point of oil fractions and diesel fuels and suggest its use for on-line monitoring and control of fuel-blending processes (H83). Meusinger investigated the use of MLR and high-resolution proton NMR spectroscopy to quantitatively determine the amounts of methanol, MTBE, benzene, and aromatics using an internal standard for ∼140 gasolines. These results were compared with data obtained by fluorescent indicator adsorption and GC analyses. The structural groups that produce the NMR signals were determined by two-dimensional measurements and their influences on octane number were evaluated by cluster and factor analyses. Nine of these structural groups were used to describe quantitatively the influence of structure on RON and MON, confirming the importance of chemical substructures in determining the octane number (H84). Kapur and others used stepwise MLR and 13C NMR spectroscopy for the precise estimation of heavy alkylated benzenes (HAB) in industrial oils. Two separate models were developed: one applicable for the estimation of HAB in paraffinic mineral oils and the other for the estimation in naphthenic basestocks (H85). Yalvac et al. studied the effects of elevated temperature and pressure on the FT-NIR spectra of light alkenes and found that the effects of elevated pressures on spectra were insignificant but the temperature effects were critical, especially for the lighter molecules such as ethylene and propylene (H86).

The pharmaceutical industry has found an increased use of on-line analyzers and chemometrics, mainly in the monitoring of fermentation processes. Lindberg and Lundstedt presented a review on application of chemometrics to the characterization of macromolecules in pharmaceutical development. The chemometric methods were limited to multivariate analysis, PCA, and PLS. A classification technique, SIMCA, was presented as a product control application of multivariate analysis (H87). Erskine et al. describe a chemometric approach for determining enantiomeric purity from chirally sensitive spectral measurements in combination with conventional absorbance spectra to monitor process streams to ensure enantiomeric purity of chiral products without the need for chromatographic separation. By coupling these methods, one can obtain information about the concentrations of the desired optically active components in a mixture as well as their enantiomeric purity. PLS was shown to yield the most accurate prediction of the concentrations of even a complex mixture containing similar compounds with nearly indistinguishable spectral features (H88). Jones et al. reviewed quantification of microbial productivity via multiangle light scattering and supervised learning and described the use of chemometric methods for prediction of biological parameters of cell suspensions on the basis of their lightscattering profiles. Use of such models opens up the possibility of estimating the biological properties of fermentation broths extremely rapidly. They demonstrated the usefulness of this approach for estimation of yeast cell counts over a wide range of values, although it was less successful in predicting cell viability in such suspensions (H89). Warnes et al. discussed data-based modeling techniques for fermentation process where six different modeling techniques (MLR, PCR, PLS, ARMAX, NARMAX, ANN) were considered for the recombinant Escherichia coli fermentation process. The models use industrial on-line data from the process as input variables in order to forecast the concentrations of biomass and recombinant protein normally only a available from off-line laboratory analysis (H90). Chandwani and co-workers described the use of PCA to characterize chromatographic separations from a size exclusion chromatographic system. Process chromatograms were generated, and the PCA model was used for classification of process performance. The study is of special interest to downstream processing schemes for the purification of recombinant proteins prepared by fermentation (H91). A later work describes the use of gross and disjoint PCA models that provided easily interpretable two-dimensional diagnostic plots revealing clusters of chromatograms obtained under similar operating conditions that may be useful for the diagnosis of subtle deviations from process specification not readily distinguishable to the operator (H92). Polymer processes are another area that has benefited from the growing acceptance and availability of chemometrics, particularly in the application of NIR spectroscopic monitoring. Lachenal recently published two reviews devoted to NIR spectroscopy of polymeric materials. Despite that author’s bias toward a more classic approach to spectral interpretation, he concludes that, in the presence of highly overlapping and low-resolution spectra, chemometric methods are the only way to obtain accurate results (H93, H94). Dallin discusses some of the issues raised and the solutions offered by process near-IR analysis (H95). Chalmers and Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


Everall highlight the potential of FT-Raman and FT-IR capabilities coupled with multivariate data analysis through novel applications evaluated for the compositional and physical structure characterization of polymers (H96). Li et al. discussed monitoring vinyl ester-styrene network formation by FT-IR under conditions typically used for “on-line” composite processing (H97). Aust et al. developed a probe design and multivariate analysis method for real-time detection of the percent of cure of epoxides using in in situ fiber-optic Raman spectroscopy (H98). Biesman and co-workers describe a method for measuring the dynamic surface tension behavior of silicone surfactants in model liquids representative of the polyol component in a polyurethane foam-forming mixture. PCA analysis of commercially available silicone surfactants allowed the classification of the various surfactants into several subsets, each characterized by a specific dynamic surface tension behavior. The combined method of dynamic surface tension measurements and principal component analyses allows fast screening of surfactants used in the polyurethane industry (H99). Tabe and Simons discuss the use of PCA and NN to convert multivariate data from tomographic images into useful information suitable for the control and optimization of multiphase chemical processes (H100). Process Control. In process control, there has been significant acceptance and integration of chemometrics techniques (primarily because of their ability to deal with highly collinear data) in model-based control strategies. This has led to the development of new chemometric algorithms to accommodate the special needs of control algorithms. Dayal and MacGregor describe a fast recursive, exponentially weighted PLS algorithm that provides greatly improved parameter estimates in most process situations. The potential of this algorithm is illustrated with two process examples: (i) adaptive control of a two-by-two simulated multivariable continuous stirred tank reactor and (ii) updating of a prediction model for an industrial flotation circuit. The main advantage of the recursive PLS (RPLS) algorithm is that it does not suffer from the problems associated with correlated variables and short data windows (H101). Gay and Ray discussed the basic properties of the singular value decomposition (SVD) for integral equation models of distributed parameter systems (DPS) in the context of process identification and model-based control (H102). Lee and co-workers compared iterative identification methods based on SVD decomposition, QR decomposition, and LU decomposition that implement the input and output transformations in the context of integrating into control strategies (H103). Qin compares several RPLS algorithms for (i) on-line process modeling to adapt process changes and (ii) off-line to reduce computation time and computer memory usage in PLS regression and cross-validation. A block-wise RPLS algorithm is proposed with a moving window and forgetting factor adaptation schemes and is extended to dynamic modeling and nonlinear modeling. An application of the block recursive PLS algorithm to a catalytic reformer is presented (H104). Finally, the kernel algorithm shows dramatic speed improvements, especially in cross-validation and recursion, that allows for rapid on-line model adaptation (H105-H109). Dayal and MacGregor presented a paper describing the theoretical justifications for using multioutput identification for a multivariate process as an alternative to the common practice of 152R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

identifying a multiple-input single-output (MISO) model for each output separately and then combining the individual models into a final multiple-input, multiple-output (MIMO) model. The potential benefits of performing identification on all outputs simultaneously were investigated via simulations on two process examples: a quality control example and an extractive distillation column. The identification of both the parsimonious transfer function models using multivariate prediction error methods, and of nonparsimonious finite impulse response (FIR) models using partial least squares (PLS2), canonical correlation regression (CCR), and reduced rank regression (RRR) are considered. The major benefits are in the stability and performance robustness of controllers based on the identified MIMO models, features the authors termed more “control relevant” (H110). Patwardhan et al. discussed methods of implementing an input-constrained, nonlinear, model-predictive controller in latent spaces using PLS-based Hammerstein and Wiener models for MIMO systems and highlighted the constraint mappings needed. The PLS-based Wiener models are for constrained control of nonlinear systems, and the use of Hammerstein models for control involves solution of a nonlinear program. The proposed approach is demonstrated on a simulated pH-level control of an acid-base neutralization process (H111). Chen and McAvoy presented a process control approach using steady-state multivariate statistical models to improve product quality when the quality measurements are not available on-line, or they have long time delays. A statistical controller is designed to control the equivalent score space representation of the process. The issue of how to account for the correlation structure of input variables when closing a feedback loop around the PCA model is specifically addressed (H112). In a later paper, Chen et al. describe a PCA control approach that incorporates time-lagged variables in a model-predictive control (MPC) framework to control the equivalent score space representation of the process. The score predictive model for the MPC algorithm is built using PLS. The proposed controller can be developed from and implemented on top of existing PID control systems (H113). Both papers use two case studies involving a binary distillation column and the Tennessee Eastman process to illustrate the techniques. Pottmann and Seborg present a control strategy for nonlinear processes based on radial basis function models used to train a nonlinear predictive controller implemented as a radial basis function network. When applied to an experimental pH neutralization process, it provided excellent set point tracking and disturbance rejection when compared to conventional PI control (H114). Arkun and Kayihan present a reduced-order internal mode control (IMC) design for control of cross-directional profiles in sheetforming processes using adaptive PCA to resolve the potential problems associated with large dimensional and ill-conditioned systems. Simulations illustrate that the resulting controller is easy to design and able to isolate and reject the significant disturbance modes (H115). Lakshminarayanan et al. discuss modeling and control of multivariable chemical process systems using dynamic PLS where discrete input/output data are utilized to construct a model that captures the dominant features of the process under study. The structure of the resulting model enables the synthesis of a multiloop control system. The design of feed-forward control for

multivariable systems using the dynamic PLS framework is also presented. Three case studies are used to illustrate the modeling and control of multivariable linear and nonlinear systems using the suggested approach (H116). Sans et al. described an application of PCA in the determination of stoichiometric models from on-line spectroscopy for mechanism elucidation, process optimization, and on-line quality control of semibatch processes. On-line spectroscopy and kinetic multidetection are used to optimize kinetic parameters. The mathematically extracted concentrationtime profiles were verified with traditional methods of quantitative analysis (H117). Maria and Rippin review development of the modified integral transformation procedure (MIP) estimation method (H118) that combines elements of similarity analysis and prior information about similar model structures to the classical integral transformation procedure (IP) for kinetic parameter estimation. MIP allows rapid adaptation of a kinetic model, describing an already studied process, to a similar process under study with only the product distribution known (H119). The MIP was integrated in an expert system for kinetic identification and coupled with statistical analysis (H120). The interaction with the prior information allows on-line adaptations of the model structure and parameters, comparable with extended Kalman filter (EKF)based recursive estimators (H121). Holcomb et al. discuss tools from nonlinear time series analysis, classical statistics, and chemometrics for building nonlinear models based on input/output data. The false nearestneighbor method is used to determine the order of nonlinear ARMAX models, and PLS is used to determine the local linear structure of the input/output map. By putting PLS in the general context of “significance regression”, it is possible to deal effectively with collinear data involving errors on the inputs and outliers (H122). This work was extended by Bomberger and Seborg to dynamic, discrete-time nonlinear autoregressive models with exogenous inputs (NARX) models. The two methods are applied to several dynamic systems, including realistic process simulations and experimental data from the UCSB pH neutralization process. The usefulness of model order detection methods for radial basis function network (RBFN) identification is also examined (H123). Peng and Jang discuss the use of fractal analysis to reduce the size of a time-series data set for high-quality nonlinear model predictive control. Simulation examples, including the dual composition control of a high-purity distillation column demonstrate that the nonlinear model predictive scheme is quite useful for those cases in which linear model predictive controller has failed (H124). Ikonomopoulos and Endou describe a nonlinear MISO empirical model for monitoring vital system parameters in a nuclear reactor environment using a scheme of nonparametric smoothing that models the local dynamics of each fitting point individually using independent estimators (H125). Heaven et al. discuss applications of model-based tuning and analysis tools to paper machine control. These tools use chemometrics to identify a process model based on process input/output data and use these models to determine controller tuning and decoupling relationships. Case studies of these tools applied to cross-directional (CD) profile control to determine input/output relationships, response characteristics, alignment, controller parameters, and multivariable models are examined. A benchmark using minimum variance control theory against which the per-

formance of a traditional CD control system can be judged is illustrated (H126). Neural Networks. Pikington et al. provide a review of neural networks as process classifiers and nonlinear regression models for applications in process control, process modeling, condition monitoring, and related manufacturing contexts. A description of neural networks was followed by a description of the development cycle that has been used successfully to deliver robust models for real engineering problems. An industrial drying process, which is typical of the type of complex problem encountered in process engineering, demonstrated the methods (H127). Braake et al. recently completed a review and discussion of predictive control in biotechnology using fuzzy and neural models and model-based predictive control (MBPC). Both fuzzy and neural models were developed and identified for a nonlinear pressure control problem in a fermentor and shown to outperform the classical PI controller (H128). Significant theoretical efforts have been made toward improving the robustness and understanding of neural nets. Derks and Buydens presented two theoretical papers on aspects of network training and validation on noisy data (H129, H130). Jiang et al. describe a recursive algorithm for optimizing the architecture of feed-forward neural networks (FF-NN) by the stepwise addition of a reasonable number of hidden nodes and the use of genetic algorithms, modified for network training, to circumvent the local optimum problem (H131). Shao et al. present a method for the calculation of confidence bounds for FF-NN models. The new technique is then applied to the modeling of specific quality variables in a batch polymerization process (H132). Burden et al. discussed cross-validatory selection of test and validation sets in multivariate calibration (PCR, PLS) and ANN and proposed a method for the production of a true cross-validated neural network regression model (H133). Wang et al. proposed a robust backpropagation algorithm by introducing transforms on the residual term to circumvent the problem of overfitting to outlier points (H134). Gernoth and Clark describe a modified back-propagation algorithm, based on a maximum likelihood criterion, for training neural networks on data with known error estimates (H135). Nord and Jacobsson investigated two algorithmic approaches to interpreting each variable’s contribution to ANN models and compared results with the corresponding variable’s contribution in PLS regression models (H136). Kurtanjek describes the modeling of baker’s yeast production by the principal component-based artificial neural networks (PCANN) in adaptive control of fermentation by the IMC method. PCA of process variables result in projection of patterns to a space of low dimension, which facilitates determination of ANN structure, removes data collinearity and random components of measurement signals, and reduces the risk of model degradation by overtraining. In view of IMC application, the models for prediction of the controlled variable (ethanol partial pressure) and the inverse model for manipulative variable (molasses feed rate) are determined and tested (H137). Cubillos and Lima describe an adaptive hybrid neural model (HNM), based on fundamental conservation laws associated with a PC-ANN used to model the uncertain parameters. Since a neural net within HNM has fewer parameters than a pure black box NN model, an on-line training method may be used. The adaptive HNM approach was applied Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


to two simulated processes: a four-stage flotation unit (H138) and a highly nonlinear CSTR (H139). Aguiar et al. show that a simple hybrid network, which uses deterministic model results to reinforce neural net mappings, is able to improve model prediction as well as training time for pulping processes and pulp properties (H140). Tung et al. describe a hybrid network architecture combining the (FF-NN) and the recurrent network (RNN). Similar to the adjustment of the Kalman filter gain, the recursion gain is changed according to an approximated noise covariance to the error covariance ratio. A variable back-propagation through time (BPTT) algorithm is proposed for the training of this variable-structure network and demonstrated on an experimental nonlinear pH neutralization system. The hybrid network is then extended to a multiple-feedback-path system for dynamic data rectification (i.e., the task of removing errors from measured process data). Three neural network architectures, the hybrid network, the Elman net and the FFN, are tested on a literature CSTR example (H141). Karjala and Himmelblau describe a similar approach to the dynamic rectification of data involving the use of recurrent neural networks (RNNs) and the EKF. By interpreting RNNs within a nonlinear state-space context, a state-augmented EKF can be used to optimally estimate both the states of the RNNs and noise and bias models (H142). Aldrich and van Deventer presented a comparison of different artificial neural nets for the detection and location of gross errors in process systems (H143). Qin and McAvoy describe a dynamic modeling method using nonlinear finite impulse response (NFIR) models and a neural net-PLS (NNPLS) approach. The NNPLS method is an integration of PLS and neural networks that solve the collinearity problem typically associated with far-IR models. For comparison, an alternative approach using NARX is discussed. The NNPLS/NFIR approach is demonstrated on data from two industrial MIMO processes (H144). Another NNPLS application was described by Andersson et al. for the transformation of nonlinear response data via a neural network and subsequent standard linear PLS regression. The applicability of this approach was demonstrated using three real-life industrial data sets (H145). Jiang and Wang discuss nonlinear discriminant feature extraction using a modified back-propagation (BP) network (meansquare criterion replaced with a discriminant criterion) that extends linear feature extraction techniques to a wide variety of nonlinear pattern recognition applications (H146). In a separate paper, Jiang et al. incorporated nonlinear mapping (NLM) into a BP algorithm to visualize multidimensional clusters by mapping the multidimensional data to a perceivable low-dimensional space (H147). Wienke and Buydens discuss artificial neural networks based on adaptive resonance theory (ART) that form a collection of distinct pattern-recognition methods for classification of sensor signals, process data analysis, spectral interpretation, and image analysis. The advantages of ART are considered, with applications, including its use as a built-in detector for outliers, rapid training speed, self-organizational behavior, full chemical interpretability, and real-time and on-line applicability (H148). Wienke et al. described the FuzzyARTMAP (H149) (adaptive resonance theorybased neural network) pattern recognition algorithm applied to automated identification of postconsumer plastics by NIR spectroscopy (H150, H151). 154R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

Welsh et al. evaluated several computer-based classifiers as potential tools for pharmaceutical fingerprinting by utilizing normalized data obtained from HPLC trace organic impurity of drug lots from six different manufacturers. The performance of several ANN architectures was compared with that of two standard chemometric methods, K-nearest neighbors (KNN) and SIMCA (H152). Rademan et al. described efforts to analyze and model an ill-defined and poorly understood leaching process from historical data by a learning vector quantization (LVQ) neural network and a BP neural network. The LVQ neural net models were used to perform sensitivity analyses to determine the relative importance of the different process variables (H153). Moolman and others extended previous work based on neural network classification of industrial flotation froths (H154) to include neural network image compression and morphological edge detection for advanced process control systems. A systematic approach for the application of neural network data compression and the correct choice of parameters was proposed and illustrated with examples (H155). Chen and Bruns extended the WaveARX neural network to include net adaptation and on-line implementation in real time. Two demonstrations are implemented on-line using process simulations of a chaotic map and a pH CSTR model and provide illustrations of several capabilities of the network, along with comparisons to other identification techniques (H156). Krishnapura and Jutan described a new autoregressive moving average (ARMA) neuron network designed for modeling nonlinear dynamics often encountered in chemical engineering processes. The ARMA neurons are used to model both stimulated and experimental nonlinear dynamic processes, including an industrial fluidized bed reactor (H157). Ryman-Tubb showed that the recognition ability of an odor sensor array would be significantly improved using a neural computing approach to discriminate between similar odors (H158). Xing and He described an array of nine piezoelectric quartz crystal sensors, each coated with a different crown ether derivative, constructed for multicomponent analysis of organic vapors and comparison of three chemometric techniques (ANN, PLS, nonlinear PLS) used in the data analysis (H159). Savkovic-Stevanovic described a control system using a neural net, based on process inverse dynamic modeling, applied for product composition control of a distillation plant (H160). MacMurray and Himmelblau discussed the modeling of a packed distillation column as an interesting example of complex modeling because the column exhibits a change in the sign of the gain under various operating conditions. It is shown that artificial neural networks can model the column and is as good or better than a simplified first-principles model when used for model predictive control (H161). Clark and co-workers describe an effort to control a cupola melting process using hybrid neural networks as an inverse process model in a cascade, feed-forward controller in conjunction with a finite difference model of the cupola. A separate implementation of the neural net serves as a stand-alone visualization tool (H162). Kelkar and Mahajan described a hybrid physical model/neural network approach for the modeling and optimization of a vertical MOCVD reactor. Optimum process conditions to obtain a uniform thickness of the deposited film were determined and tested using the ANN model (H163).

Syu and Hou describe the neural network study on the dynamic identification of a fermentation system provided by a dynamic learning and prediction process that moved along the time sequence batchwise. The generalization ability of the network to another batch of fermentations was tested. The prediction results from the time-delayed neural networks were also studied (H164, H165). Lednicky and Meszaros investigated self-recurrent neural networks in dynamic modeling of continuous fermentation. Maximization of cellular productivity of the baker’s yeast continuous fermentation was used as the goal of the proposed optimizing control problem (H166). Ronen et al. described a modular neural network where the data space is partitioned into several overlapping domains, defined by an unsupervised fuzzy clustering procedure, and a neural network in each domain map input/output relations. The effectiveness of several cluster validity measures was compared and tested with data obtained from fermentation model simulations and a fermentation of yeast-like fungus (H167). Wavelets. As in other fields of data analysis, wavelets and wavelet transforms have found applications in process chemometrics. Walczak and Massart published an introductory paper on the fundamentals of wavelets and wavelet transforms and the applications to signal compression and denoising, image processing, data set compression, and modeling of multivariate data sets (H168). A treatment of the wavelet packet transform (WPT) applied to real and simulated signals for signal compression and denoising are described in a tutorial (H169). Different approaches to the WPT coefficient selection are discussed. In a later paper, they present a new approach to best-basis selection for a set of signals, enabling significant and uniform data compression in the time-frequency domain. The high degree of data compression can be used as the first step of fast approximate PCA of data sets with high rank. The performance of the proposed approach is tested for two data sets and compared with previous approaches to fast approximate PCA (H170). Lu and Mo describe the general expressions for wavelet analysis and review its application in chemometrics (H171). A paper by Depczynski et al. describes the mathematical background of the wavelet transform on compact intervals with pseudocode and a brief spectroscopic example (H172). Leung et al. produced a review, covering the 1989-1997 period, of wavelet transform techniques in chemical analysis applications (H173). A very readable discussion of wavelets, and their relation to chemometrics, can be found in the tutorial by Alsberg et al. (H174). The tutorial is presented in a time-frequency framework with comparisons to the more familiar Fourier transform. Examples and visual interpretation of the wavelet transform steps help the new reader in the field grasp the important concepts. Application areas of wavelet transforms in chemistry and analytical biotechnology are discussed. References to available software tools (commercial and freely available) should help the interested reader get started using wavelets. Nikolov et al. describe application of a wavelet shrinkage algorithm for denoising of two-dimensional element distribution images generated by scanning secondary ion mass spectrometry (SIMS). In reconstructions of SIMS images resulting from this algorithm, the noise is significantly suppressed without great loss of lateral resolution (H175). Wolkenstein et al. compared a wavelet denoising algorithm to several well-known digital smoothing

operators for enhancing the visual impression of images from electron probe microanalysis (EPMA). The wavelet transform denoising yielded results comparable to the best filters in use (H176). Mittermayr et al. compared the wavelet soft universal thresholding algorithm to Fourier filters and to polynomial smoothers on the basis of mean squared error, signal-to-noise ratio, and change in the peak area. Four common wavelets (Haar, Daubechies, Symmlets, and Coiflets) were evaluated for denoising of simulated signals (i.e., narrow Gaussian peaks with added white noise). Their results showed that wavelet denoising was superior to classical filter techniques in most cases (H177). Lu and Mo presented B-spline wavelet multiresolution analysis for denoising UV spectra (H178). Alsberg and co-workers investigated six wavelet denoising methods applied to infrared spectra and compared their performance with Fourier and moving mean filtering in terms of RMSE between the pure and denoised spectra and visual quality of the denoised spectrum (H179). Trygg and Wold discuss using the fast wavelet transform on NIR-visible spectra as a preprocessing method in PLS calibration. Compression of the data set to 3% of its original size was achieved while the predictive ability and the diagnostics are basically the same as for the original uncompressed regression model (H180). Jouan-Rimbaud and co-workers described variable selection in the wavelet domain for multivariate calibration and compared performance with variable elimination and standard PLS for the modeling of near-IR data. This approach can also be used to remove noise and irrelevant information from spectra for multivariate calibration (H181). Mallet et al. discuss several wavelet approaches in linear and nonlinear discriminant analysis of full spectrum and feature-extracted spectral data. The discrete wavelet transform is introduced as a method for extracting features and compared to the Fourier transform, principal component analysis, stepwise strategies, and other variable selection methods (H182). Woodward et al. point out that the structure of noise in a data set and, in particular, whether it is homoscedastic or heteroscedastic, can significantly affect the properties of multivariate models. Optimal denoising schemes based on wavelet shrinkage take into account the noise structure of a data set (H183). Fang and Chen studied the WPT as a tool of multiresolution analysis for electroanalytical chemistry responses and described an adaptive wavelet filter that can automatically select the optimum filter for various signals with lower frequency band, such as voltammograms, chromatograms, and FIA responses (H184). Shen et al. discussed multiresolution analysis based on orthogonal wavelet bases to remove the influence of complex backgrounds and to correctly determine the number of components in twoway chromatographic data without assumptions of the profiles of the baseline or the background spectra. Thus, most kinds of background can be eliminated regardless of their behaviors (H185). Shao and Cai investigated using wavelet transform and window factor analysis (WFA) for the resolution and quantitative determination of multicomponent chromatograms with noise (H186). Young and co-workers describe an iterative method for differentiating between known resonances and uncharacterized baseline contributions in magnetic resonance spectra using wavelet shrinkage and denoising (H187). Walczak et al. described a standardization method based on transferring NIR spectra in the wavelet domain and compared it Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


with the PDS method, the biggest improvement occurring when very few standardization samples were used to estimate the standardization parameters (H188). Stork et al. presented an automated method integrating wavelet processing and techniques from MSPC for the simultaneous localization, detection, and identification of disturbances in spectral data (H189). Nesic and co-workers describe the analysis of paper machine process data using discrete wavelet transforms. Wavelets are an effective representation for the detection of basis weight and moisture process variations in noisy data and lead to improved estimation and visualization of the machine-direction and cross-direction variations. The use of wavelet compression to decrease data storage requirements is important in mill-wide process monitoring systems (H190). Safavi and others describe an application of wavelets and multiresolution analysis for the purpose of process monitoring and data analysis (H191). Nikolaou and Vuthandam describe a methodology for the direct identification of parsimonious far-IR models through wavelet-based signal compression and compare it with other far-IR identification methods. Certain industrial practices are shown to be special cases of the proposed formalism (H192). Bakshi et al. describe a new class of methods developed for the on-line rectification of stationary random errors in the absence of fundamental or empirical process models. A new multiscale wavelet PCA method is developed that provides better rectification than PCA, by simultaneously extracting the relationship among the variables and among the measurements. The performance of the multiscale univariate filtering and multiscale PCA are illustrated by several examples and are shown to provide better rectification than the widely used method of exponential smoothing (H193). Esbensen et al. discuss the angle measure technique (AMT) as another potentially useful transform in process monitoring and chemometric applications. The AMT has several features similar to wavelet transforms (i.e., localized scales, noise filtering, compression, etc.) (H194). FLOW AND SEQUENTIAL INJECTION ANALYSIS FIA (I1) and its more recent format sequential injection analysis (SIA) (I2-I5) can provide a great deal of versatility in process analytical applications. Many published FIA/SIA methods claim to be “on-line” methods, possibly since the intended chemical or physical transformations leading to detection take place in a flow line. Yet very few of these published FIA/SIA methods are actually on-line methods, providing real-time data for a process line. In FIA or SIA, all samples and reagents are manipulated fluidically by the pumps, valves, and reactors that comprise the flow system. The flow system’s sample line can be situated in a production reactor vessel, process line, or flask containing the same material. Once sampling occurs, the automated fluid handling coupled with flow-through detection is the same regardless of the origin of the sampled volume. This review covers the period 1995-1998. Newman (I6) summarized commercially available flow and sequential injection analysis equipment in 1996. Two excellent searchable databases covering the full range of FIA/SIA applications are available on the World Wide Web (I7, I8). For the purpose of this review, process analytical applications for FIA/SIA will be divided into two categories: (1) “traditional” 156R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

chemical process monitoring utilizing a number of different detection approaches and (2) bioprocess monitoring, primarily utilizing specific enzyme reactivity. Its scope is limited to FIA/ SIA applications involving real industrial process streams, intermediates, or products. However, not all applications described are truly on-line methods. Many FIA/SIA methods are focused on automation of routine manual methods; such automation can be beneficial in providing process parameter information if samples presented to the flow system have had no pretreatment steps or only a dissolution/dilution step is carried out. Simple pretreatment steps can readily be carried out by plant personnel outside of the laboratory environment, allowing the FIA/SIA method to provide data with a much higher throughput than can be obtained with the corresponding manual methods. Chemical Process Monitoring. While fully on-line methods of analysis can be useful for chemical process monitoring, such monitoring is in many cases impractical, since in many batch chemical processes, solid products are produced or solid intermediates are isolated for further processing. In such cases, automation of routine manual methods combined with the relative speed provided by the FIA/SIA system allows effective process control procedures to be implemented. Johanssen et al. (I9) describe a FIA method for determination of a primary aromatic amine in viscous, nonionic, iodinated X-ray contrast media. X-ray contrast media are orally or intravenously administered pharmaceutical substances used in diagnostic radiology procedures which enhance visual contrast in X-ray images of different tissues and organs. The concentration of the specific primary aromatic amine, a process intermediate, is controlled during production. The FIA method was developed to replace a time-intensive manual procedure used for quality control. The FIA and manual methods are based on the same chemistry, forming an azo dye which is detected spectrophotometrically; only dissolution of the samples is required prior to introduction to the FIA system. Two different commercial FIA systems were used in method development, but the particular system used to obtain result validation data is not specified. While this FIA method is not on-line, significant reduction in analytical costs in the production environment can be realized due to increased speed of analysis. This work provides a good example of method development steps needed to make FIA work on real industrial samples. Especially good attention is given to refractive index effects, which are often overlooked when spectrophotometric detection is used in FIA/SIA methods. van Staden and van der Merwe (I10) propose a SIA system for monitoring nitrite, an impurity in fertilizer process streams, based on the same chemistry and similar spectrophotometric detection used in the previous paper; only the target analyte has changed. The SIA system used is partly custom-built and partly commercial. This paper provides a good comparison of SIA and FIA implementations of the same chemistry, as in the Johanssen paper described above (I9). No on-line results were reported, and only limited results for real samples were reported, with only one fertilizer process stream sample analyzed. Cosano et al. (I11) describe methods for on-line monitoring of several parameters in industrial ammonium thiosulfate production. Two determinations were for high concentrations (∼60%) of feedstock ammonium sulfite and product ammonium thiosulfate. For these determinations, large dilutions were needed, and the

same chemistry and flow system were used, with only stream selection differing. Excellent reproducibility and repeatability were needed for these determinations; good signal precision was reported, but no results for real samples or on-line results were reported. Determination of impurity ammonium sulfite in the final product required different chemistry and a separate flow system. Good comparative off-line results for impurity ammonium sulfite in eight real samples (FIA vs gravimetric method) were reported. Customized hardware and software were used for construction of the two flow systems. On-line monitoring was claimed to have been checked for six months in a pilot plant; however, no data for this monitoring were reported. Nyman and Ivaska (I12) report spectrophotometric SIA determination of calcium in white liquor from paper production. A commercial SIA system comprising both hardware and software was used. Results from on-line monitoring at a paper mill are reported, as are results from off-line determinations in six liquor samples. Comparative results for both on-line and off-line determinations are provided by an unspecified dc plasma atomic emission spectrometry (DCP-AES) method. Agreement in determined calcium concentrations between SIA and DCP-AES methods is poor and is attributed to the difference between “free and weakly bound” (SIA) and “total” (DCP-AES) calcium concentrations which the respective methods determine. While this explanation may be partially true, it does not explain the relatively poor precision obtained by the SIA method for the off-line samples. Other effects, such as refractive index differences between the standards (in a simple matrix) and the white liquor samples (a far more complicated matrix) need to be investigated and eliminated before the proposed SIA method could be considered reliable. In another application related to paper production, Chai and Danielsson (I13) describe an on-line FIA method for spectrophotometric monitoring of sulfur species during electrochemical production of polysulfides. While this work was limited to a laboratory study, FIA was used to implement real-time process monitoring. Useful techniques for effecting very large dilution factors and using cross-flow filtration during sampling are presented. The flow system used was entirely custom-built. Albertu´s et al. (I14) determined total ammonia and carbon dioxide spectrophotometrically in nickel extraction liquors with a two-channel FIA system. These liquors are essentially carbonated ammonia solutions, with the ammonia/carbon dioxide ratio being the most important process parameter. These liquors are quite unstable with respect to compositional change unless great care is taken, and the titrations normally employed for these determinations are slow and labor-intensive. Using a commercial FIA system, acceptable ammonia/carbon dioxide ratios are found in real process liquors. As developed, the method requires dilution of the liquors prior to introduction to the FIA system, so the method could not be tested on-line. Andrew et al. (I15) reported a custom-built, automated, portable on-line FIA system for monitoring ammonia in effluent. This system uses a gas diffusion cell to transfer ammonia from an alkaline sample stream into an indicator stream with spectrophotometric detection of the reacted indicator. The portable system showed a positive bias relative to a manual method, but the system successfully monitored an effluent stream over several

days. The FIA results reflected all variations in stream ammonia levels. In all of the examples discussed above, reaction chemistry was used to create detectable species from the target analyte, and spectrophotometry was employed for detection. Other detection methods can be utilized in flow systems for process analysis. Two FIA methods used in the manufacturing process control of a drug substance were described by Chong et al. (I16). One method, for determination of activity of a lipase used for selective enzymatic hydrolysis, is based on conductivity. The other method, for determination of residual Triton X-100 surfactant, is based on online strong cation-exchange solid-phase extraction of all reaction components other than the surfactant, which is detected by UV absorption. The methods are not used on-line, since the lipase activity is determined prior to starting the synthesis, and residual Triton X-100 is determined on an intermediate isolated as a solid. Dissolution was the only sample preparation step required before sample introduction into the FIA system. The FIA systems used are both custom-built. Barnett et al. (I17) describe a SIA method for determination of morphine in water-immiscible process streams based on chemiluminescence in a water carrier stream. The custom-built system was not used on-line, but a series of process samples were analyzed, and results comparable to a standard liquid chromatographic method were obtained, with the SIA method providing less precision due to the heterogeneous reaction and detection conditions. Only dilution was required prior to sample introduction to the SIA system. This paper also discusses some practical differences between FIA and SIA when used for process analytical applications. Barnett et al. (I18) also discuss a FIA method for determination of codeine in process streams, based on chemiluminescence and using a heterogeneous lead oxide/silica reactor, but application of this method to real process samples was limited. Physical modulation of reaction chemistry is not the only use for FIA or SIA approaches to process analysis, however. Creasy (I19) and Capuano et al. (I20) show how FIA can be crucial to using infrared absorption for real-time monitoring of a chemical process, using the highly precise dilution capability of microfluidic manipulation. FIA was used to dilute samples that fouled surfaces of attenuated total reflectance infrared cells. The diluted samples could then be successfully analyzed by simple infrared absorption without fouling surfaces of the flow cell. On-line application to a laboratory polymerization reaction was shown, and it was claimed the system was installed at pilot plant and plant sites, although no data from these sites were provided. Dilution in a FIA system was also crucial to monitoring textile dye baths as reported by van Delden et al. (I21) and Wallace et al. (I22). These laboratory studies present fully on-line results in textile process monitoring applications, in which the dye bath sample has an extremely complicated matrix, and partial leastsquares regression is used for absorbance calibrations for solution dye concentration determinations (with several dyes present in the dye bath). HPLC was coupled with FIA in one of these studies (I22), allowing detailed information about reactive dye behavior during exhaustion to be obtained. The FIA systems described are custom-built, and these studies are excellent examples of working FIA process analysis systems for a challenging matrix. Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


Bioprocess Monitoring. Application of FIA and SIA approaches to on-line bioprocess monitoring is a very active area of investigation. A variety of review articles have been published by van der Merbel et al. (I23), van der Pol et al. (I24), Schu¨gerl et al. (I25), Christensen et al. (I26), Keay and Wang (I27), and Olsson et al. (I28), which attest to the increasing importance of FIA and SIA to bioprocess monitoring. The popularity of FIA and SIA is due to the limited range of target analytes (I23), the specific enzyme-based detection schemes (I28), similar reaction matrixes, and similar sampling requirements that occur with bioprocess monitoring. In contrast, chemical process monitoring applications of FIA and SIA, as discussed above, must adapt to a wide range of target analytes, varied chemistry to produce detectable species, extremely diverse reaction matrixes, and unique sampling requirements specific to a given process. However, essentially all published bioprocess monitoring FIA/SIA applications are limited to laboratory studies, while some of the published chemical process monitoring FIA/SIA applications are working in plants or on real process samples. For this review, only reports of actual on-line bioprocess monitoring FIA/SIA applications are discussed. Ju¨rgens et al. (I29) present enzyme cartridge FIA methods for monitoring glucose, ethanol, L-amino acids, and lactate in various bioprocesses. The cartridges were integrated into a commercial FIA system, and on-line data were given for glucose, ethanol, and lactate, with good agreement between the FIA online data and the corresponding appropriate off-line analytical method data. The enzyme cartridges showed reasonable stability, and they are claimed to be suitable for industrial process monitoring. This laboratory study used a 15-L reactor. van Putten et al. (I30) describe a six-channel FIA system for monitoring concentrations of starch, maltose, glucose, urea, ammonia, and phosphate as well as activity of alkaline serine protease. This laboratory study used a 20-L bioreactor, and the FIA system was custom-built, incorporating both enzyme reactors and standard reaction chemistry. Some on-line results show reasonable agreement with off-line data collected by standard methods, but on-line protease activity was different due to membrane fouling of the tubular sampling system. The complex FIA system was stable for an unspecified large number of runs. Schuhmann et al. (I31) developed a custom SIA system for on-line control of fermentation processes. Glucose was determined on-line over a two-day period by enzyme-based detection. Unfiltered fermentation broth containing yeast cells was introduced into the SIA system, degrading detector sensitivity over time, but frequent automated recalibration compensated for this degradation, allowing reasonable data to be collected during the entire monitoring period. ULTRASONIC ANALYSIS Ultrasonic measurements have been performed on solids, liquids, gases, and various types of mixtures for over 50 years (J1). The first reported observations for the use of ultrasound applied to a reacting system were made in 1946 to measure the extent of polymerization in a condensation or radical process (J2). Since that time, a diverse range of measurements have been made that consider ultrasonic interactions in terms of the effect of materials or processes on four ultrasound metrics: ultrasonic velocity; attenuation, absorption, and scattering as functions of 158R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

frequency and of composition; process reaction or phase (time/ rate); and temperature. However, the various ultrasound methods are probably the least understood of all those chemometric tools employed in process analytical chemistry. Recent years have seen a resurgence of interest in ultrasound technologies and its application for material characterization, process measurement, monitoring, and control. This has to a significant extent been facilitated by the formation of interdisciplinary teams to work on process problems and by major advances in the capabilities of advanced PCs, interface and digitization cards, and improved ultrasound transducers This review will consider ultrasonic technologies with particular consideration given to its application to the chemical and process plant. The applications of ultrasonic measurements in solid-state physics and much of more traditional physical acoustics and ultrasonic nondestructive testing (NDT) are outside the scope of this review. As this is the first review in this series to include ultrasound, for the sake of completeness, some references to the literature before the start of this current review period are included. Ultrasonic Systems and Measurements. Many of the fundamental aspects of the physics of ultrasonic measurements are found in classic texts, such as that by Truell et al. (J3). Further good references that consider ultrasonic transducers, velocity and attenuation measurements, viscosity measurement, and ultrasonic chemical relaxation spectroscopy are the contributed chapters in the book, Ultrasonics, in the series Methods of Experimental Physics, edited by Edmonds (J4). The relationships between ultrasound interaction with materials and molecular structure were of considerable interest in the 1970s (e.g, ref J5), and are again starting to receive some attention (e.g., ref J1). Much of the early literature that applies ultrasonic measurements for process measurement is reviewed and cited in the text by Lynnworth (J6). This book focuses on applications and considers ultrasonic methods for flowmetry, thermometry, density, porosity, and property measurement and interface sensing, together with a review of selected topics from more traditional NDT. This literature has recently been supplemented by two significant texts: Ultrasonic Techniques for Fluids Characterization, by Povey (J7), and Ultrasonic Sensors for Chemical and Process Plant, by Asher (J8). Both books should be considered essential reading by anyone seeking to review this technology, its potential, and the scope of prior work. These books include extensive bibliographies for the earlier literature, including numerous examples and data for applications of ultrasound technology to chemical measurements and processes. Velocity and Attenuation To Characterize Media and Monitor Processes. Numerous transmission measurements have been made by Povey that measure the velocity and attenuation of a fluid or mixture between a pair of transducers (J7). The theory for the velocity and attenuation of ultrasound in suspensions of particles in fluids was reviewed and extended by Harker and Temple (J9). Subsequent work by Harker et al. extended these approaches to the analysis of ultrasonic propagation in slurries, consisting mainly of silicon carbide in water and ethylene glycol (J10). Tsouris and Tavlarides (J11) used the speed of sound coupled with a linear model to estimate the volume fraction of water in oil.

Ultrasonic systems have been devised by Hale to give the density from ultrasonic data which can be used for process control (J12). This devise has been developed into an ultrasonic pulseecho reflectometer, for both pulse-echo, by McClements and Fairley (J13), and swept frequency (0.3-6 MHz) operation, by McClements and Fairley (J14). Such measurements can be used to investigate the properties of solutions, emulsions, and colloidal systems, including to monitor both melting and crystallization as shown by McClements et al. (J15). Ultrasound has been extensively used by Lavallee et al. to analyze the thermal and viscoelastic behavior of polyurethane/ poly(vinyl chloride) blends (J16) and the morphology of polymer blends by Piau and Verdier (J17) and Verdier and Piau (J18). These approaches have been extended and applied to the study of the temperature-dependent properties of epoxy prepolymers and the ultrasound data compared with that from Brillouin scattering as demonstrated by Matsukawa et al. (J19). The use of ultrasound velocity and attenuation measurements for on-line measurements has been integrated into extrusion systems by Gendron et al. for on-line real-time monitoring (J20). It is under investigation for the measurement of the properties of metal/ powder/viscous liquid suspensions, used in slurry casting synthesis by Schultz (J21). Monitoring Solidification (Interface Sensing). Ultrasonic techniques operating in pulse-echo have been used to monitor the position of interfaces during solidification. Early work by Bailey and Davila (J22) considered both low-melting-point metals and N-paraffins. Subsequent studies by Parker and Manning have monitored the solid/liquid interface during both solidification and melting (J23). Such measurements have been performed by Parker et al. in the 1-5-MHz frequency range and included investigation of the “mushy zone” and interfacial structures (J24). The same basic approaches have been applied more recently by McDonough and Faghri to follow the interface positions during solidification of an aqueous sodium carbonate (J25). The use of simple pulse-echo measurements can be extended to include the reconstruction of solid/liquid interfaces in solidifying bodies as was shown by Mauer et al. (J26). Acoustic Time Domain Reflectometry. This technique can be described as backscatter and interface sensing. The use of scattered ultrasound energy for structural analysis was initially developed for material characterization, including grain size analysis. It was extended to the analysis of two-phase and multiple scattering. A review of the fundamental aspects of structure analysis is provided by Goebbels (J27). The development of ultrasonic measurements for on-line real-time monitoring of processes in membrane science was started in 1994 by Bond et al. (J28). Pulse-echo ultrasonic measurements, were developed into acoustic time domain reflectometry (ATDR)sthe name was selected by analogy to optical TDR. ATDR has now been implemented as a tool to study fouling and compaction in membrane-based separation systems by Bond et al. (J28) and Peterson et al. (J29). This technique has been used to provide a tool for the investigation of the structural properties in thin films by Konagurthu et al. (J30) and for the measurement of thickness during evaporation casting of polymeric films by Kools et al. (J31). Other applications of this approach include monitoring of sedimentation, solidification, membrane treatments, and characteriza-

tion of inorganic, biological, and particulate fouling in a variety of processes. There is a common misconception that high-frequency ultrasonic measurements cannot be made in gases. The attenuation in gases does become high at room pressure as reported by Bond et al. (J32); however, at elevated pressures, process measurements and gas-coupled acoustic microscopy become possible at frequencies in the range 10-30 MHz as reported by Chiang et al. (J33) and Bond et al. (J34). Operation in gases has significant advantages in terms of the resolution that is achievable at a particular frequency; this is due to the lower wave velocity in the gas. Also attenuation and velocity measurements in gas and vapor provide the basis for composition analysis as demonstrated by Matheson (J5). Most recently, ultrasonic methods have been demonstrated for the on-line real-time characterization of state of mixing. This work includes the use transducers that operate in the lowmegahertz frequency range in an “A-scan” (pulse-echo or ATDR) mode and medical-style B-scan (2-D sector) imaging that provides real-time video images of mixing processes as shown by Bond et al. (J35). Three-Phase Reactors. Ultrasonic techniques have a long history of use to investigate properties of both saturated and partially saturated rock systems, those involving hydrocarbons, gases, and water. In some laboratory studies, measurements are made at very high temperatures and pressures, and in others, three-phase systems including glass beads are employed to simulate sandstone. Such studies provide many insights that are directly relevant to measurement of multiphase process beds. Preliminary ultrasonic velocity and attenuation have been reported by Soong et al. (J36) for slurries consisting of water, glass beads, and nitrogen bubbles. This work was then extended to characterization of slurries consisting of molten paraffin wax, glass beads, and nitrogen bubbles at 198 °C by Soong et al. (J37). These studies demonstrated that ultrasonic techniques do have potential for online real-time monitoring of two- and three-phase process reactors. Process Tomography Using Ultrasonic Methods. Recent years has seen a growing interest in the implementation of various measurement schemes that provide imaging for industrial flows and processes, and a family of methods now know as process tomography have been developed and implemented. An interesting review of the concept of image analysis providing information useful in chemistry was provided by Geladi and Esbensen (J38). A review of the various technologies for the noninvasive tomographic and velocimetric monitoring of multiphase flows was provided by Chaouki et al. (J39). This covers all measurement modalities that have been applied to give such data. The ultrasonic portion of the review is less comprehensive than other parts of the article. It does however provide much useful data and citations. Plaskowski et al. (J40) provided a book that presents the fundamentals for imaging industrial flows. It does include a short section on ultrasound, but it is only superficial in its coverage and rather negative in the view presented for the assessment of the capabilities of ultrasound. The more recent journal literature does include reports by Martin et al. (J41) on the application of ultrasonic imaging in pilot plant-scale process gas and liquidprocessing vessels. The use of ultrasonic tomography for monitoring gas/liquid flow is reported by Xu et al. (J42), and most recently, the use of air-coupled ultrasound transducers for imaging Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


temperature and flow fields in gases was demonstrated by Wright et al. (J43). Ultrasonic Holography. Brenden (J44) developed a real-time ultrasonic imaging system (RTUIS) that uses a 100 µs pulse that is transformed into an optical image by a pulse of coherent light of 6 µs duration. Images are formed at video frame rates. The system can be applied to visualize multiphase flows such as fluidized beds, bubbly flows, thermal gradients, and mixing phenomena in real time through vessel walls as shown by Shekarriz et al. (J45). Ultrasonic Transducers. Ultrasonic measurements can be implemented with a diverse range of types of transducers and include the use of designs and materials for operation in hazardous environments. Conventional ultrasonic transducers operate up to ∼50 °C. Clamp-on and waveguide transducers are available to operate to at least 300 °C as presented by Lynnworth et al. (J46). Custom and semicustom transducers are practical that will operate to 600 °C as shown by Dreacher-Krasicka et al. (J47) and Broomfield (J48). Alternate ultrasound generation and detection can employ lasers or electromagnetic acoustic transducers (EMATs) as shown by Silk (J49), both of which have particular advantages for specific types of application. Density Measurement. Several manufacturers produce instruments to measure density on-line and at-line as well as supplying laboratory instruments. Anton Paar (J50) has developed instruments that measure the speed of sound through a liquid as a function of temperature. Previous calibration measurements are used to determine the density of a given liquid for on-line measurements. Instruments using the frequency of oscillation of a U-shaped tube are also used to determine the density. Other manufacturers that produce instruments based upon measuring the speed of sound through a liquid are Cygnus Ltd. (J51), Canongate (J52), and Nusonics (J53), according to Povey (J54). Fuji Ultrasonics (J55) is another manufacturer to utilize the speed of sound technique for on-line measurements. Endress and Hauser (J56) produces meters that measure the density of a liquid or slurry based upon the Coriolis force, due to oscillations of the pipes containing the liquid. The next step is to consider the recent development of other ultrasonic instruments to measure the density of a liquid or slurry and to discuss their advantages. The on-line ultrasonic viscometer, designed by Sheen et al. (J57-J59), measures the density and viscosity of a liquid or a slurry flowing through a pipeline. It consists of two transducer wedges mounted on a pipe opposite one another and flush with the inner surface of the pipe. A longitudinal-wave transducer is mounted on one wedge and measures the reflection of ultrasound from the wedge/liquid interface; a shear-wave transducer on the second wedge measures a similar quantity. Each wedge has as offset surface so that a portion of the ultrasound is reflected from a wedge/air interface to provide a continuous reference signal for self-calibration. The density is determined by measuring the acoustic impedance of the liquid and the speed of sound through the liquid. The shear reflection coefficient is used to calculate the density-viscosity product. The typical error in the density measurement is 1% and in the viscosity measurement is 5%. The minimum value of the viscosity measurement is ∼50 cP. Different shear-wave wedge designs and materials are currently being investigated for lower viscosity measurement. One advantage of 160R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

this instrument is that it measures both the density and viscosity on-line. Another advantage is that no calibration measurements are needed to interpret the results of the measurements. Kline (J60) describes a method to measure the density of a liquid, such as aircraft fuel. The densitometer uses a transducer to transmit an ultrasonic pulse through the liquid to a block of reference material. The pulse is reflected from the near side and far side of the block, as well as an internally reflected portion of the pulse. The density is determined as a function of the amplitudes of the three return pulses. The velocity of sound in the liquid is determined from its measured time of flight. This system seems more complicated than the system of Sheen et al. that also measures the time of flight. McClements and Fairley (J61) have designed an ultrasonic pulse-echo reflectometer for use in the laboratory that measures the ultrasonic velocity, attenuation coefficient, and characteristic impedance of liquid materials. The density is obtained from the impedance and velocity measurements. This instrument is similar to that described by McHale (J62). The error in the density measurement was ∼0.5%. The instrument is immersed in a thermostatically controlled water bath to ensure constant temperature and results were averaged over ∼2000 signals. Greenwood (J63) used ultrasonic attenuation measurements to determine the concentration of a slurry. This method was used to determine the concentration of a slurry in experiments carried out in a 1/12-scale model of a double-shell tank on the Hanford reservation (J64). These were on-line measurements. However, calibration measurements of attenuation vs volume fraction of the slurry were carried out beforehand in order to interpret the on-line attenuation measurements. Greenwood et al. (J65, J66) developed an on-line ultrasonic density sensor in which six transducers (five longitudinal wave transducers and one shear wave transducer) are mounted upon a plastic wedge. The base of the wedge is in contact with the liquid or slurry. Longitudinal ultrasonic beams strike the base of the wedge at angles of 0°, 45°, and 60° with respect to the normal to the base. A shear wave at 0° also strikes the base of the wedge. The amount of reflection at the plastic/liquid interface depends on the incident angle, the density of the liquid or slurry, the speed of sound in the liquid or slurry, and the wedge parameters. The voltages on the receive transducers are compared to those when the base is immersed in water and the reflection coefficients are determined. By determining the reflection coefficient at two angles, the density of the liquid and the speed of sound can be determined. The error in the density measurement is ∼0.5% for liquids and about 1-2% for slurries. Currently, improvements are being implemented in the electronics and data analysis to increase the accuracy of the density measurements. There are several advantages to this instrument. The first is that the sensor consists of only one component. Since the measurement depends on the reflection at the plastic/liquid interface, and not the passage of ultrasound through a liquid, the density of a liquid that severely attenuates ultrasound can be determined by this method. The wedge base can be placed in a cutout section of a pipeline wall, and the compact design allows deployment in short pipe spool pieces. The sensor is not affected by electromagnetic noise and can be located in harsh environments and other areas crowded with machinery.

Ultrasonic Characterization of Multiphase Fluids and Flow. Ultrasonic signals have attributes that are well suited for characterization of multiphase fluids and flows. The signals have the ability to interrogate fluids and dense opaque suspensions, penetrate vessel and process walls, and not be degraded by noisy process conditions because the signal frequencies differ from that of machinery. Also measurement is not affected by flow rate. Ultrasonic sensors can be designed to provide real-time, in situ measurement or visualization of process characteristics; the sensors and sensing systems are compact, rugged, and inexpensive. Ultrasonic sensors can be designed to measure fluid density, viscosity, and velocity; slurry density, particle size, weight or volume percent solids concentration, stratification, and rheology; and colloid gelation and particle size and to quantify multiphase flow interfaces, state of mixing, homogeneity, and slurry transport (a) Slurry Particle Size and Concentration. Scott and Paul (J67) provided a summary of instruments available for on-line particle characterization. The sensing techniques reviewed include laser-light scattering or diffraction, laser-doppler and phasedoppler interferometry, electrical impedance or capacitance tomography, in-process video, focused beam, or dual-beam reflectance, electroacoustic spectroscopy, and ultrasonic attenuation or extinction. The majority of the instruments operate on dry (gassolid) mixtures. For measuring the particle size distribution or concentration of particulate laden slurries, ultrasonic attenuation has significant advantages. Sympatec Inc., System-Partikel-Technik (J68), markets an in-line system to measure particle size (0.13000 µm) at particle concentrations up to 70 vol %. Ultrasonic attenuation spectra from slurry signal penetration provide information about slurry particle size and concentration as reported by Bamberger et al. (J69). As the particle size and the acoustic frequency are changed, the relative importance of attenuation mechanisms changes, and the acoustic attenuation can be dominated by different effects. Three regimes are important for slurries consisting of solids in a liquid such as fine sand in water: viscous regime, inertial regime, and Rayleigh scattering regime as described by Kyto¨maa (J70). Researchers have used transitions between two regimes, the viscous and intertial regime, to quantify particle size and concentration in real time as per Boxman et al. (J71). Pendse et al. (J72-J74) have developed ultrasonic spectroscopy to measure the particle size of industrial slurries and colloids; an acoustic spectrometer for laboratory-scale measurements based on this technique is marketed by Dispersion Technology, Inc., Bedford Hills, NY. Pendse and Han (J75) have also extended this method to nonspherical particles. Pendse and Sharma (J73, J74) applied this method on a titanium dioxide slip stream; the unit operated at low flow rate (2 L/min) at frequencies up to 100 MHz. The configuration is limited to small transducer separation distances with ∼1 in. being the largest distance investigated (personal communication with H. Pendse, January 28, 1998). For diverse slurries, systems that capture transitions between three regimes are more robust as shown by Kyto¨maa (J70). In the viscous regime, Biot (J76), Urick (J77), Hampton (J78), Atkinson and Kyto¨maa (J79), and Allegra and Hawley (J80) have shown that the viscous boundary layer thickness is larger than the particle radius. As frequency is increased, the viscous boundary layer becomes thinner than the particle radius; losses

occur in the thin boundary layer surrounding the particles as presented by Biot (J76), Johnson et al. (J81), Sheng and Zhou (J82), Zhou and Sheng (J83), Atkinson and Kyto¨maa (J84), and Allegra and Hawley (J80). Atkinson and Kyto¨maa (J85) described the nonlinear behavior of attenuation at concentrations above 1020% for both regimes. Kyto¨maa and Corrington (J86) carried out the first documented experiments in the inertial regime to derive the nonlinear behavior of the added mass coefficient with concentration. This theory spans the two regimes, and the representation for the complex wavenumber was developed in detail by Atkinson and Kyto¨maa (J85). Derksen and Kytomaa (J87) used ultrasonics to determine the added mass coefficient. At sufficiently high frequencies, Rayleigh scattering initiates, where energy from coherent incident sound is scattered by the random distribution of particles attenuating the coherent as per Allegra and Hawley (J80). Experiments conducted by Salin and Scho¨n (J88), near maximum packing concentrations of glass spheres, showed that the transition can well be described by an additive combination of the Atkinson and Kyto¨maa theory to the Rayleigh scattering theory. This was later confirmed for slurries by tests conducted by Greenwood et al. (J89). Theoretical studies have also been performed by Spelt et al. (J90). For a fixed particle size, the transition between each of these regimes occurs at a frequency defined by the particle size and the mechanical properties of the solid and the fluid. The viscous regime exhibits a quadratic scaling with frequency; the inertial regime scales with second power of frequency, while the Rayleigh scattering regime scales with the fourth power of frequency. Each regime exhibits a constant slope on a log-log plot. The first transition from the viscous to the inertial regime is gradual and occurs over two decades or more, while Rayleigh scattering sets in over a narrower range in frequency. The frequency or range of frequencies over which transition occurs is strongly dependent upon particle size. For larger particles, the transitions will tend to occur at lower frequencies and, conversely, at higher frequencies for smaller particles. A real-time, in situ sensor to measure slurry mean particle size, distribution width, and concentration has been developed and tested with slurries composed of a range of particle types. The approach has been demonstrated for specific slurry types. Accuracy for mean particle diameter determination was within 1 µm, and accuracy for solids fraction was 1% as described by Bamberger et al. (J69). (b) Velocity Profiles and Rheology. Handa et al. (J91) obtained velocity profiles of rotating flow of a magnetic fluid using an ultrasound velocity profile monitor; in addition, the threedimensional flow structure was clarified. Takega et al. (J92) used ultrasonic doppler velocimetry to measure mercury flow in the target of a neutron source. The technique provides spatiotemporal information about the flow field and is used for flow mapping. Shekarriz et al. (J93) used ultrasonic doppler velocimetry and timeof-flight measurements. From these measurements, the local shear rate is determined from the local velocity in the pipe. (c) Multiphase Flow “Visualization”. Kyto¨maa and Corrington (J86) used clinical ultrasound backscatter to distinguish liquefied from settled states in transient events. The liquefied regions were identified by characteristic small-scale random fluctuations consistent with particle diffusion in suspensions. Good et al. (J94) used acoustics to monitor the curing of grout by Analytical Chemistry, Vol. 71, No. 12, June 15, 1999


measuring the elastic properties. Clark et al. (J95) investigated using ultrasonics for mapping radioactive waste stored in million gallon tanks using attenuation and speed of sound over the kiloto megahertz frequency ranges. Cavitation, Sonochemistry, and Sonoluminescence. The use of ultrasound to affect changes in chemical systems has increased greatly over the past decade. Likewise, the number of papers describing theory and practical applications has burgeoned. In 1994, this field was conferred its own journal, Ultrasonics Sonochemistry. Sonochemistry in a liquid generally arises when acoustic power is sufficiently intense to disrupt the intermolecular bonds, causing cavitation. Originating at seed nuclei, cavitation bubbles grow and collapse nonlinearly during the rarefication and compression phases of the acoustic waves. The rapid collapse of a bubble subjects the vapor contents and liquid/gas interface to temperatures and pressures on the order of 5000 K and 1000 bar, respectively. These intense conditions last for nanoseconds in an otherwise cool liquid. It is during this period that sonochemistry occurs, either from direct exposure to the extreme conditions or from reactions with chemical byproducts produced within the bubbles. Bubble collapse near heterogeneities involves mechanical effects. A jet of fluid shooting across the bubble’s interior will strike a solid with tremendous destructive force. For suspended particulates, the collapse can agglomerate smaller particles or break polymer bonds. Often, the violent collapse of cavitation generates picosecond flashes of light by a much debated mechanism. This light, known as sonoluminescence, has been used as a probe to monitor the interior conditions of cavitation. The history of cavitation and its destructive power dates to the 1870s when damage to ship propellers became problematic (J96). Its use for sonochemistry reaches back to the early part of this century. A review by Weissler (J97) covers this early period, citing over 150 papers from 1926 to 1950. Reviews by Henglein (J98) and Suslick (J99) cover the interim period to the late 1980s. The phenomenon remained relatively unknown and little used, due in part to poor equipment. As this situation improved, the field of sonochemistry began to grow. The first international conference on sonochemistry was held at Warwick University, England, in 1986. Shortly thereafter, several books became available. One of the most active early sonochemists, Mason wrote or edited several books (J100-J103) in the late 1980s and early 1990s. Mason and Cordemans (J104) provide a review of sonochemistry application in chemical and processing industries through 1995. Several key areas are discussed in these reviews and books, including medical applications, pharmaceuticals, biotechnology, chemical synthesis, polymer technology, minerals and powders, and metallurgy. Common uses such as homogenization and ultrasonic cleaning are also discussed. The concepts behind ultrasonic cavitation and sonochemistry have been well documented. Recent reviews can be found by Leighton (J105), Lauterborn et al. (J106), and Crum (J107). A book by Brennan (J108) covers these topics more completely. Barber et al. (J109) review the physics of a single bubble collapse and the unknowns of ensuing sonoluminescence. The mechanism of ultrasonic degassing is discussed by Eskin (J110). While this phenomenon has been studied since the 1930s, this paper discusses degassing of liquids and light alloy melts, including removal of hydrogen under industrial conditions. 162R

Analytical Chemistry, Vol. 71, No. 12, June 15, 1999

One of the most recent active areas of sonochemistry has been the decomposition of organic contaminants by ultrasound. This has application in the treatment of wastewater. Hirai et al. (J111) decomposed CFCs and HFCs in water. Chlorinated hydrocarbons and cyclohydrocarbons in water have been degraded by Kruus et al. (J112), Visscher and Langenhove (J113), and Shirgaonkar and Pandit (J114). Francony and Petrier (J115, J116) and Hoffmann et al. (J117) degraded chlorinated hydrocarbons, pesticides, phenols, esters, and explosives in water. Pure carbon disulfide was degraded by Entezari et al. (J118). Johnston received a patent (J119) for decomposing halogenated organic compounds with ultrasound. The acceleration of certain chemical reactions by the application of an ultrasonic field can occur by direct agitation and/or new reaction intermediates. This effect has been reported by Tuulmets et al. (J120) for Grignard reagent formation, Polackova et al. (J121) for the Cannizzaro reaction, and Enomoto et al. (J122) for iron powder processing. Photochemical reactions were accelerated by Gaplovsky et al. (J123), and Stephanis et al. (J124) accelerated formaldehyde reactions with ultrasound. During the sonication of diesel fuels, Price and McCollum (J125) found alkanes shorter than C20 “crack” to shorter alkanes and alkenes, while aromatic and nitrogen containing compounds polymerize. The high velocity and temperature collisions of suspended particles in a cavitation field, followed by rapid cooling, can accelerate reactions and create novel products. Luche (J126) discusses steric modification of cycloadditions under the influence of ultrasound. The use of ultrasound for electroless plating of ceramic materials is discussed by Zhao et al. (J127). Polymer synthesis has been enhanced and controlled by Price (J128), Portenlanger and Heusinger (J129), and Katoh et al. (J130). Stoffer and Sitton (J131) were issued a patent for a polymerization process using ultrasound on ethylenically unsaturated monomers. Amorphous formation of iron, cobalt, molybdenum, and tungsten has been studied by Suslick et al. (J132-J134). Boeing was issued two patents (J135, J136) for using this method to create magnetic recording media in a continuous process. The rejuvenation of a metal surface by this process can also activate and preserve catalytic activity. Enhancement of electrolysis during sonication has been known for years. Madigan et al. (J137) discuss conditions under which damage or particulate deposition occurs on electrodes. Lowering overvoltages in electrolytic processes reduces cost and byproducts. Tatsumoto et al. (J138) demonstrated a lowering of the oxidation current during voltammetry. Improved electrolytic recovery of metals was reported by Walker (J139). Ultrasound has also found a number of uses in the food industry. Mason et al. (J140) report on ultrasonic enzyme activation, extraction, emulsification, and several other processes. Bhatkhande and Samant (J141) describe the saponification of vegetable oils assisted by ultrasound and its benefits over traditional methods. The presence of trace contaminants in water is of great importance for the semiconductor industry. By studying sonoluminescent light from cavitating systems, qualitative information about a system can be determined. Kuhns et al. (J142) measured part per thousand concentrations of alcohols in high-purity water and identified them multivariately. Ashokkumar et al. (J143)

quantitatively measured the effects of trace surfactants and alcohols on light emission. An ultrasonic method for detecting submicrometer particles in ultraclean liquids is reported by Madanshetty (J144). Sonoluminscence may also prove useful for identifying and quantitating species at higher concentrations. Emission from alkali and alkaline earth metals is readily observed, as shown by Flint and Suslick (J145). They also showed that emission lines from the ultrasonic destruction of larger species can also be measured (J146). Sonoluminescence is extremely sensitive to a complex set of external parameters, which has hindered its more widespread use as an analytical method. However, advancements in this area continue. MISCELLANEOUS TECHNIQUES Nuclear Magnetic Resonance. Long a workhorse of the analytical laboratory, NMR has continued its slow development into a process analytical tool. McDonald described the use of NMR for on-line process control and quality assurance (K1). His review article provided general background to the technique and a history of on-line NMR. The Ph.D. dissertation of Skloss (under Professor Haw) described a proof-of-concept high-resolution NMR for industrial processes (K2). It was capable of monitoring liquid streams for dissolved analytes to limits of detection of ∼0.1 v/v %. Edwards and Giammatteo described the use of NMR in a slipstream to measure acid strength, hydrocarbon-to-olefin ratios, and other parameters important for control of a sulfuric acid alkylation process (K3). They indicated that the instrument has been in operation for over a year without downtime, while providing rapid data comparable to laboratory analyses. Proton NMR was also used by Gao et al. to determine gas oil composition in an N-methyl extraction process (K4). Noriyuki et al. patented an NMR device and method for a different kind of processs automatic selection of vegetables and fruits (K5). Microwave Spectroscopy. Process measurements using microwaves have existed for some time (e.g., moisture measurement instruments from Kay-Ray and Berthold), but the technique has not really “caught on”. Nevertheless, new commercial instruments, publications, and patents indicate continuing interest and developments in this area. Yasuhiko et al. described and patented a technique to follow fermentation processes by monitoring the attenuation of a microwave signal during the process and relating it to lactic acid concentration (K6). Shinichi et al. described a microwave apparatus to measure “molecular” orientation of paper or polymer sheets (K7). Varpula and Seppae looked at fiber orientation in paper and paperboards by using alternating electric fields in the radio or microwave frequency ranges (K8). Changes in the direction of the signal due to the material were related to fiber orientation. Kestner et al. developed a microwave method for measuring the water content of hydrocarbons over the range of 0-100% (K9). They used measurements of the amplitude, phase of transmission and reflection, and temperature to allow calibration over the full range. A unique fin design minimized problems from wax buildup on the waveguides. Board described use of a microwave instrument to measure moisture in an Eirich rotatingpan mixer in a concrete manufacturing process (K10). Marrelli et al. used radiation in the 10-12-GHz frequency band to determine solids-to-liquids ratio in a petroleum stream containing

solids, oil, and water (K11). Takeshi et al. described an apparatus to measure the freeness value of a paper pulp stream in real time (K12). Their device calculated the freeness from the pulp liquid (water) concentration obtained using a microwave measurement, other actual measurements, and additional stored data. Seichi described a technique to measure moisture content that used microwave attenuation at dual frequencies to correct for varying concentrations of nonwater components (K13). Goldberg et al. described a device that measured consistency (suspended solids concentration) and flow rate of a slurry (K14). The device used microwave propagation in a waveguide and cited examples included paper pulp and coal/water slurries. Another application of microwaves to paper pulp slurry concentration was described by Jakkula (K15). A wood grading and sorting instrument using microwave moisture measurement was described by Atsushi et al. (K16). Finally, Abernethy et al. described measurement of mass flow of pneumatically conveyed solids using microwaves and ultrasound (K17). Both approaches were successful in some applications, but the microwave technique was unable to determine solids flow velocity of coal powder at the low solids densities (