Article pubs.acs.org/IECR
Control of a Heat-Integrated Pressure-Swing Distillation Process for the Separation of a Maximum-Boiling Azeotrope William L. Luyben* Department of Chemical Engineering, Lehigh University, Bethlehem, Pennsylvania 18015, United States ABSTRACT: Methanol and trimethoxysilane form a maximum-boiling homogeneous azeotrope whose composition changes fairly significantly with pressure, so separation into high-purity products is viable using pressure-swing distillation. Heat integration of the two columns can be used to reduce energy costs since they operate at different temperatures. The pressure in the high-pressure column is not controlled but “floats” depending on conditions in the base of the low-pressure column. An auxiliary reboiler is required to balance the heat duty in the low-pressure column. Since pressure affects the composition of the azeotropic, this variable pressure increases the complexity of the required control structure. The purpose of this paper is to develop a control scheme that can effectively handle large disturbances in feed flow rate and feed composition.
1. INTRODUCTION Many papers have discussed the steady-state design of pressureswing distillation processes for minimum-boiling azeotropes, and several papers have explored the dynamic control of these systems. A recent example is a paper by Yu et al.1 in which the minimum-boiling azeotrope formed by methylal and methanol is studied. A control structure for a heat-integrated pressure-swing distillation process is developed that uses pressure-compensated temperature control to account for the variable pressure in the high-pressure column. However, the design of pressure-swing distillation systems for maximum-boiling azeotropes has received much less attention,2 and the dynamic control of these systems appears to be unexplored in the literature. That is the purpose of this paper. A paper3 dealing with the control of an extractive distillation process for separating a maximum-boiling azeotrope has appeared, but no studies of a pressure-swing system have been found. A recent paper4 discussed the steady-state design of a pressureswing distillation process for the separation of the maximumboiling methanol−trimethoxysilane azeotrope. Figure 1 gives the flowsheet of the economic optimum process for this separation. Details of the nonideal phase equilibrium, equipment sizes, and process conditions can be found in the previous paper. Figure 2 shows that the xy-curves and azeotropes change with pressure. Figure 3 shows the temperature profiles in the two columns. The high-pressure column has a significant temperature break in the rectifying section, so temperature control may be possible. However, the pressure in the high-pressure column is not constant, so simple temperature control will not be effective. There is very little temperature change in the low-pressure column, so direct composition control is necessary. The reflux ratio is the high-pressure column is modest (RRHP = 2.69), but in the low-pressure column the reflux ratio is somewhat larger (RRLP = 5.65). Column diameters are 1.36 and 2.35 m in the high and low-pressure columns, respectively.
and reflux drums are specified to provide 5 min of liquid holdup when 50% full. Pump heads and control valves pressure drops are selected to provide the required rangeability for changing flow rates away from steady-state values in the face of disturbances. Of particular importance is the pumping of the bottoms from the low-pressure column at 0.25 bar back into the high-pressure column at 7 bar. The sizing of the pump and control valve in this recycle stream are made complex by the variable pressure in the high-pressure column. As demonstrated in the simulation results presented later in this paper, the pressure in the high-pressure column changes quite significantly from the design value of 7 bar. Pressures as high as 9 bar and as low as 5 bar must be handled by the plumbing as flow rates also change by up to 20%. Proportional level controllers are used with KC = 2. Composition and temperature controllers are tuned by using relay-feedback tests and Tyreus−Luyben tuning rules. Deadtimes of 1 and 3 min are used in temperature and composition loops, respectively. The impurity levels in the product streams are the important variables. In the high-pressure column, the specified value of the mole fraction of trimethoxysilane impurity in the distillate is xDHP = 0.01. In the low-pressure column, the specified value of the mole fraction of methanol impurity in the distillate is xDLP = 0.01. 2.2. Heat Integration Implementation in Aspen Dynamics. The heat-transfer rate in the condenser of the high-pressure column QCHP (and in one of the reboilers in the low-pressure column) is determined by the overall heat-transfer coefficient (U = 0.85 kW m−2 K−1), the area (27.67 m2), and the differential temperature driving force (ΔT). This ΔT, which changes with time, is the temperature in the high-pressure column reflux drum minus the temperature in the low-pressure column base. Both of these temperatures vary with time as compositions and pressures vary dynamically.
2. DYNAMIC SIMULATION 2.2. Plumbing and Controller Tuning. The steady-state Aspen Plus file is exported into a pressure-driven Aspen Dynamics simulation file after the sizes of the column bases
Received: July 31, 2014 Revised: September 13, 2014 Accepted: October 28, 2014
© XXXX American Chemical Society
A
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 1. Heat-integrated pressure-swing flowsheet for methanol/trimethoxysilane.
Figure 2. xy diagram for methanol/trimethoxysilane at 0.25 and 7 bar.
At any point in time the heat-transfer rate is calculated from the fixed area and U, using the current temperatures. This heat-transfer rate is the heat removed in the
high-pressure condenser (QCHP) and must be equal to a portion of the heat added in the two reboilers of the lowpressure column. The remaining portion of the heat added in B
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 3. (A) Temperature profile in HPC at 7 bar; (B) temperature profile in LPC at 0.25 bar.
3. CONTROL STRUCTURE CS1
the reboiler of the low-pressure column comes from the auxiliary reboiler (Qaux). The Flowsheet Equations function in Aspen Dynamics is used to achieve these relationships. Several examples are presented in the next section. In the following sections we explore several alternative control structures that are tested for their effectiveness in handling disturbances in feed flow rate and feed composition.
The columns under study have fairly high reflux ratios, and in this situation control of reflux-drum level is usually achieved by using reflux flow rate. The use of temperature to infer composition appears to be problematic because of the flat temperature profile in the low-pressure column and the variable pressure in the highpressure column. C
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
condenser, which drives up the pressure in the high-pressure column PHP. The control structure is unable to stabilize the system even with no disturbances. Tuning the two composition controllers was found to be very difficult since the initial steady state is unstable. Alternative structures are explored in the next sections.
A control structure that is frequently used under these conditions is to control distillate composition by manipulating reboiler duty. In the process under study, there is partial heatintegration since there is an auxiliary reboiler on the low-pressure column. This provides two independent heat inputs that can be used to control the two distillate compositions. With reboiler duty used for distillate composition control and reflux used for level control, the control structure must be set up such that the distillate flow rate can vary. One commonly used approach is to ratio the distillate flow rate to the reflux flow rate. Figure 4A shows this common control structure (CS1). The various loops are listed as follows. 1. Feed is flow controlled. 2. Pressure in the low-pressure column is controlled by manipulating condenser duty. 3. Pressure in the high-pressure column is not controlled. 4. Trimethoxysilane impurity in the high-pressure distillate is controlled by manipulating reboiler duty in the highpressure column. 5. Methanol impurity in the low-pressure distillate is controlled by manipulating the duty in the steam-heated auxiliary reboiler. 6. Reflux-drum levels in both columns are controlled by manipulating reflux flow rates. 7. Distillate flow rates in both columns are ratioed to their corresponding reflux flow rates. Reflux flow rate is measured and the signal is sent to a multiplier whose other input is the reciprocal of the design reflux ratio. The output signal of the multiplier is the remote setpoint signal of a distillate flow controller, which is on “cascade”. 8. Base levels in both columns are controlled by manipulating bottoms flow rates. Note that the bottoms streams recycle between the columns. This structure does not have a flow controller somewhere in the recycle loop, so the potential for “snowballing” exists. The simulation results presented below show that this occurs. The Aspen Dynamics flowsheet equations for this control structure are given in Figure 4B. The first equation calculates the heat duty in the reboiler/condenser using a fixed area, a fixed U and the temperature difference between the top of the highpressure column and the base of the low-pressure column. The second equation sets the total heat input to the lowpressure column QRLP to be the sum of the heat duty in the condenser/reboiler and the heat duty of the auxiliary reboiler. The auxiliary reboiler duty is the output signal of the composition controller “CCLP” that is controlling methanol impurity in the low-pressure distillate. Dynamic simulation results for this control structure are shown in Figure 4C. There is no disturbance. Both composition controllers are on manual, which means that the high-pressure reboiler duty and the auxiliary reboiler duty are fixed. Distillateto-reflux ratios are controlled. Reflux-drum and base levels are controlled. Pressure in the low-pressure column is controlled. Somewhat surprisingly, this commonly used control structure does not work. Snowballing occurs. The flow rates of both bottoms streams (BHP and BLP) slowly decrease, which increases the high-pressure distillate DHP. This increases the composition of the heavy component trimethoxysilane (xDHP) in the high-pressure distillate, which increases the temperature at the top of the high-pressure column (TCHP). The larger temperature differential increases heat transfer in the reboiler/
4. CONTROL STRUCTURE CS2 The fundamental snowballing problem can be eliminated by placing a flow controller somewhere in the liquid recycle loop. Figure 5A shows a control structure in which the recycle bottoms from the low-pressure column (BLP) is flow controlled and ratioed to the feed flow rate. This structure should eliminate snowballing. Since the bottoms cannot be used for base level control in the low-pressure column, the heat duty of the auxiliary reboiler is selected. The composition of the low-pressured distillate is controlled by manipulating the low-pressure distillate flow rate. The control structure in the high-pressure column is modified so that the composition loop is similar to that used in the lowpressure column; that is, the composition of the high-pressure distillate is controlled by manipulating the high-pressure distillate flow rate. Tuning of the composition controllers was straightforward. The high-pressure reboiler duty is set by ratioing the steam flow rate to the feed. So there are now two ratios: QRHP/F and BLP/F. Figure 5B gives the flowsheet equations that are modified to make the heat duty of the auxiliary reboiler the output signal from the base level controller (LCLP1) in the low-pressure column. Thus, auxiliary reboiler duty is now controlling base level. The dynamic performance of this control structure for a 10% increase in feed flow rate is shown in Figure 5C. Stable control is achieved. Note the change in the pressure PHP (bottom right graph in Figure 5C) for this increase in throughput. The heat transfer in the condenser/reboiler must increase so the differential temperature must increase. This means the temperature in the top of the high-pressure column must increase, so pressure must be higher. Although the CS2 structure is stable, there is a very large dynamic transient in the composition of the high-pressure distillate (xDHP). This occurs because the 3 min composition deadtime limits the dynamic response. In addition, manipulating distillate flow rate involves the dynamics of the reflux drum. The composition controller is quite slow. A relay-feedback test and Tyreus−Luyben tuning give KC = 0.13 and τI = 115 min using a composition transmitter range of 0 to 5 mol % methanol and a controller output range of 0 to 100 kmol/h. If a more rapid detection of disturbances could be used, the transient deviation could be reduced. Since temperature measurement has smaller deadtime (1 min), it is desirable to use temperature instead of composition. However, the variable pressure means that simple temperature control would be ineffective. In the next section, the use of “pressure-compensated temperature control” is explored. 5. CONTROL STRUCTURE CS3 The composition of a binary mixture depends on both temperature and pressure. The basic concept of “pressurecompensated temperature” control is to measure both temperature D
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 4. (A) CS1, dual composition with QR, fixed RR; (B) CS1, flowsheet equations for heat integration; (C) CS1, openloop wit RR.
and pressure at an appropriate location so that the composition can be estimated at that location. The method has been used
for many years in distillation columns that experience significant pressure changes. Vacuum columns and heat-integrated E
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 5. (A) CS2, dual composition, QRHP/F, BLP/F; (B) CS2, flowsheet equations for heat integration; (C) CS2, 10% feed flow rate increase.
systems are important examples. Yu et al.1 present a recent application.
Luyben5 summarized a procedure for developing the required equations that have been used in this paper. The resulting F
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Another indication of the easier separation is the smaller changes in pressure in the high-pressure column. Apparently smaller changes in column vapor rates and heat-transfer rates were required when disturbances occurred.
equations for estimating the composition on Stage 4 in the highpressure column (with temperature 148.62 °C, pressure 7.02 bar, and composition 61.44 mol % methanol) are x = m(P)T + b(P)
6. CONTROL STRUCTURE CS4 A cascade composition/temperature control structure is shown in Figure 7A. The composition of the high-pressure distillate is measured (with a 3 min deadtime) and controlled by manipulating the setpoint of the xCC controller, which is on “cascade”. The composition controller is tuned by using a relay-feedback test and Tyreus−Luyben tuning to give KC = 0.29 and τI = 40 min using a composition transmitter range of 0 to 5 mol % trimethoxysilane and a controller output range of 40 to 80 estimated mol % methanol on Stage 4. Figure 7B shows that a 10% increase in feed flow rate is well handled with both products quickly brought back to their specified values. However, Figure 7C shows what happens when a feed composition disturbance is imposed on the process. At 0.5 h the feed composition is changed from 50 mol % methanol to 40 mol % methanol, with a corresponding change in trimethoxysilane. More of the heavy component trimethoxysilane is being fed and must be taken overhead in the low-pressure column. The total feed flow rate does not change, so the CS4 structure does not change the low-pressure bottoms flow rate. The base level in the low-pressure column climbs as indicated by lower left graph in Figure 7C. At about 7 h, the pressure in the high-pressure column rises rapidly and the system fails. The problem is corrected by ratioing the flow rate of the lowpressure bottoms not to the fresh feed but to the total reboiler duty in the low-pressure column (QRLP), which is the sum of the heat duties
where the slope m and the intercept b are functions of pressure. m(P) = c1P + c 2 b(P) = c3P + c4
c1 = 0.0006905 c 2 = −0.018806 c3 = −0.01475 c4 = 2.7816
Note that we are calculating an estimate of the light component composition on Stage 4, not an adjusted temperature signal. This composition signal is fed as the process variable into a composition controller whose output signal changes the QR/F ratio and whose setpoint signal is 0.6144 mole fraction methanol. Figure 6A shows the new control structure. Both temperature and pressure on Stage 4 are measured, and a methanol composition is estimated using Aspen Dynamics Flowsheet Equations, as shown in Figure 6B. This signal is fed to a composition controller xCC. To get faster response, instead of manipulating the flow rate of the high-pressure distillate, we switch to manipulating the reboiler duty. Thus, the estimated composition is controlled by manipulating the reboiler duty in the high-pressure column (using a steam-to-feed ratio structure). The controller is “Direct” acting since more heat input should be used if the composition of the light component is increasing. The composition controller is tuned by using a relay-feedback test and Tyreus−Luyben tuning to give KC = 0.41 and τI = 16 min using a composition transmitter range of 40−80 mol % methanol and a controller output range of 0−0.2 ratio. A second major structural change is to switch the control of the high-pressure reflux-drum level from reflux to distillate. Since the reflux ratio in this column is modest, either configuration should be workable. A reflux-to-feed ratio is used to set the reflux flow rate in the high-pressure column. The dynamic performance of control structure CS3 is shown in Figure 6C for a 10% increase in throughput. Dynamics are much improved with more rapid responses in the high-pressure column since the composition controller integral time has been cut from 115 to 16 min. However, as the upper left graph in Figure 6C shows, the composition of the high-pressure distillate (xDHP) does not return to the desired value of 1 mol % trimethoxysilane. It levels off at about 1.2 mol %. Thus, in this maximum-boiling azeotropic distillation system, the use of pressure-compensated temperature control does provide fast dynamic control, but it does not maintain distillate composition close to is specification. This problem can be solved by using a cascade composition/temperature control structure, as discussed in the next section. Yu et al.1 demonstrated that pressure-compensated temperature control was effective in a heat-integrated pressureswing process, but the system they studied was a minimumboiling azeotrope. The separation was much easier than in the maximum-boiling azeotrope system studied in this paper as indicated by the much lower reflux ratios used in their columns.
7. CONTROL STRUCTURE CS5 Figure 8A shows the modified control structure with the flow rate of the low-pressure bottoms set by the total reboiler duty (QRLP) in the low-pressure reboilers. In the Aspen Dynamics simulation, this heat duty is available and in used as the input signal to a multiplier, which is set at the design ratio of 116.6 kmol/h to 7.503 GJ/h (using metric units as required by Aspen Dynamics). In a real plant application, as shown in Figure 8A, the total heat duty can be calculated from the flow rates of the auxiliary steam and the overhead vapor from the high-pressure column. Dynamic performance results for feed composition disturbances are shown in Figure 8B. Solid lines are for changes from 50 to 40 mol % methanol. Dashed lines are for changes from 50 to 60 mol % methanol. These disturbances are well handled. Dynamic performance results for feed flow rate disturbances are shown in 8C. Solid lines are for 15% decreases. Dashed lines are for 15% increases. These disturbances are well handled. Note that the responses for the increases in throughput are somewhat oscillatory. Larger increases in throughput increase the level of these oscillations. 8. CONTROL STRUCTURE CS5 WITH NONLINEAR CONTROLLER The performance observed suggests that the process is nonlinear and requires different controller tuning constants at different throughputs. Relay-feedback tests are run at different steady-state feed flow rates. At the design flow rate of 100 kmol/h, the CCHP G
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 6. (A) CS3, pressure-compensated TC; (B) CS3, flowsheet equations; (C) CS3, 10% feed flow rate increase.
tuning constants are KC = 0.29 and τI = 40 min. At a flow rate of 120 kmol/h, the CCHP tuning constants are KC = 0.076 and
τI = 49 min. At a flow rate of 80 kmol/h, the CCHP tuning constants are KC = 0.64 and τI = 48 min. H
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 7. (A) CS4, pressure-compensated TC with cascade CC; (B) CS4, cascade CC, 10% feed flow rate increase; (C) CS4, cascade CC, feed composition 50 to 40 mol % MeOH.
used to develop a “gain-scheduled” controller whose gain varies with the setpoint of the feed flow controller (see Figure 9A.)
The controller gain changes by almost a factor of 10 over this range of feed flow rates. Aspen Dynamics flowsheet equations are I
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 8. (A) CS5, BLP/QRLP, pressure-compensated TC with cascade CC; (B) CS5, BLP/QRLP, cascade CC, feed composition 50 to 40/60 mol % MeOH; (C) CS5, BLP/QRLP, cascade CC, 15% feed flow rate disturbances.
J
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Figure 9. (A) Flowsheet equations for gain-scheduled nonlinear controller; (B) task to ramp feed flow controller setpoint; (C) nonlinear controller with feed ramped 20%.
An Aspen Dynamics Task (see Figure 9B) is used to make ramp changes in the setpoint of the feed flow controller, increasing from 100 kmol/h at 0.5 h to 120 kmol/h over a 1 h period.
Dynamic results for both 20% ramp increases and decreases in throughput are shown in Figure 9C. These very large disturbances are well handled with stable regulatory control achieved. K
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
Industrial & Engineering Chemistry Research
Article
Note the very large changes in the pressure in the highpressure column. At the design throughput of 100 kmol/h, the pressure is 7 bar. At 120 kmol/h, the pressure is over 9 bar. At 80 kmol/h, the pressure is about 5 bar.
9. CONCLUSION A control structure has been developed for a heat-integrated, pressure-swing distillation process used to separate a maximumboiling azeotrope. The variable pressure in the high-pressure column makes control more complex. Placing a flow controller in the bottoms recycle loop is required to prevent snowballing.
■
AUTHOR INFORMATION
Corresponding Author
*E-mail:
[email protected]. Tel.: 610-758-4256. Fax: 610-7585057. Notes
The authors declare no competing financial interest.
■
REFERENCES
(1) Yu, B.; Wang, Q.; Xu, C. Design and Control of Distillation System for Methylal/Methanol Separation. Part 2: Pressure Swing Distillation with Full Heat Integration. Ind. Eng. Chem. Res. 2012, 51, 1293−1310. (2) Luyben, W. L. Pressure-Swing Distillation for Minimum and Maximum-Boiling Homogeneous Azeotropes. Ind. Eng. Chem. Res. 2012, 51, 10881−10886. (3) Luyben, W. L. Control of Maximum-Boiling Acetone/Chloroform Azeotropic Distillation System. Ind. Eng. Chem. Res. 2008, 47, 6140− 6149. (4) Luyben, W. L. Methanol/Trimethoxysilane Azeotropic Separation Using Pressure-Swing Distillation. Ind. Eng. Chem. Res. 2014, 51, 10881−10886. (5) Luyben, W. L. Distillation Design and Control Using Aspen Simulation, 2nd ed.; Wiley: New York, 2013.
L
dx.doi.org/10.1021/ie502395h | Ind. Eng. Chem. Res. XXXX, XXX, XXX−XXX
