ballman

5
Optimization of Methanol Process by Flowsheet Simulation S. H. Ballman and J. L. Gaddy' Department of Chemical Engineering, University of Missouri-Rolla, Rolla, Missouri 6540 1 Optimization of large complex chemical processes is a necessary step in economic design of these systems. Flowsheet simulation models afford an ideal mechanism for detailed optimization of these processes. The meth- anol process offers several interesting optimization variables (reformer temperature, pressure, and steam and carbon dioxide to hydrocarbon ratios; synthesis loop pressure, temperature, purge rate, and conversion) which require detailed models of the reaction kinetics. This paper presents the results of optimization of the intermedi- ate pressure process using the PROPS model and the adaptive random search routine. The costs associated with performing such a detailed process optimization study are easily justified by even nominal improvements in the process economics. Modular computer simulation programs are used exten- sively today for designing and studying chemical processes, and development of such programs is continuing (Motard et al., 1975). Sophisticated optimization algorithms have been developed and applied to unit processes, such as reactors (Barneson et al., 1970) and separators (Umeda and Ichikawa, 19711, or to simplified models of complex processes (Bracken and McCormick, 1968; Komatsu, 1968). Yet, few studies have been reported of the optimization of large complex processes, modeled with flowsheet simulators. Shannon et al. (1966) optimized a single variable of the sulfuric acid process with PACER. Seader and Dallin (1972) also used PACER to study the toluene dealkalation process, with a case study search procedure. Friedman and Pinder (1972) opti- mized the production rate of a gasoline polymerization unit, modeled with CHESS. This same process was studied with PROPS by Gaines and Gaddy (1976), with the objective of maximizing the profitability of the process. Each of the above studies was conducted at a university, and process optimiza- tion studies, using flowsheet simulators, if conducted in in- dustry, have gone unreported. The reluctance by industry to utilize simulators for detailed optimization can probably be explained by a combination of factors. There is considerable uncertainty over the amount of real time and computing time required to perform a detailed process optimization study. Industry may lack the experienced personnel to work in this area. Bounded by tight time con- straints during design and construction. the necessary ap- proach may be to come as close to optimal conditions as pos- sible using experienced engineers. and optimize the operation of the facility after it is operating. Little information IS available on the design improvements available with optimi- zation that might justify such studies. Most companies now use simulators during some phase of the design, perhaps for material and energy balance data. Therefore, it seems reasonable that additional simulation runs could be justified for optimization to improve the design. Such reasoning must be based upon the availability of a dependable process model, probabiy with economics, and a suitable op- timization algorithm. The purpose of this study is to develop a detailed simuiation model of a complex chemical process and to measure the in- vestment in real time and computer time for optimization of this process. The progress of the optimization is monitored so that the improvements obtained through optimization. even perhaps when starting very close to the optimum, can be determined. The process chosen for study is the intermediate pressure methanol process. This process is simulated with PROPS and optimized using the adaptive random search technique (Heuckroth and Gaddy, 1976). The Methanol Process Figure 1 is a typical MeOH flow diagram, showing the equipment used for producing methanol from natural gas. Natural gas is reacted with steam over a nickel based catalyst in a steam-methane reformer to produce a gas mixture con- taining hydrogen, carbon monoxide, and carbon dioxide. Since excess hydrogen is available from the reforming, carbon dioxide is added to the reformer feed to lower the ratio of hydrogen to carbon in the reformed gas. Reforming is carried out at temperatures between 1400 and 1700 OF. The steam methane reaction is endothermic, and the reactor is heated by burning a mixture of natural gas and process off gases. Energy is recovered from the reformed gas and the com- bustion gases as 900 psia steam (500 O F superheat) in a waste heat boiler. This steam is used to drive the synthesis and recycle gas compressors. Exhaust steam from the compressor turbines is used both as process steam for the reformer and as a source of heat in the purification section of the process. Centrifugal compressors are used to raise the pressure of the makeup and recycle gases to 200 atm. The synthesis gas (recycle plus makeup gas) is heated to 570 O F by the effluent gas from the methanol converter. Methanol is synthesized in the converter using a zinc oxide-chromium oxide catalyst with a copper promoter. The methanol synthesis reaction is highly exothermic. To control the temperature of the reaction, cool synthesis gas is injected between the catalyst beds. The gases from the converter exchange heat with the converter feed and are then cooled to atmospheric temperature to condense methanol. The pressure of the condensed liquid is reduced to 50 psia and dissolved gases are removed. A significant amount of the methanol is vaporized with the dissolved gases, and the MeOH is recovered by scrubbing with water. The crude methanol is then purified by distillation. Mehta and Ross (1970) and Shah and Stillman (1970) have reported optimization studies of the methanol process. Mehta and Ross studied only the effect of Cor, addition to the pro- cess. Shah and Stillman developed a FORTRAN modei of the major equipment items in the methanol process. They studied thp effect of several variables on the control of the process; however. only six iterations were used in the optimization procedure. Methanol Process Simulation The methanol process, shown in Figure 1, was modeled using PROPS (Process Optimization System), a modified version of CHESS. PROPS has the capability of computing the investment, operating cost, revenue, and profitability of the simulated process. A description of PROPS has been Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 3, 1977 337

Upload: asimilatrix

Post on 26-Dec-2015

214 views

Category:

Documents


0 download

DESCRIPTION

some

TRANSCRIPT

Page 1: Ballman

Optimization of Methanol Process by Flowsheet Simulation

S. H. Ballman and J. L. Gaddy'

Department of Chemical Engineering, University of Missouri-Rolla, Rolla, Missouri 6540 1

Optimization of large complex chemical processes is a necessary step in economic design of these systems. Flowsheet simulation models afford an ideal mechanism for detailed optimization of these processes. The meth- anol process offers several interesting optimization variables (reformer temperature, pressure, and steam and carbon dioxide to hydrocarbon ratios; synthesis loop pressure, temperature, purge rate, and conversion) which require detailed models of the reaction kinetics. This paper presents the results of optimization of the intermedi- ate pressure process using the PROPS model and the adaptive random search routine. The costs associated with performing such a detailed process optimization study are easily justified by even nominal improvements in the process economics.

Modular computer simulation programs are used exten- sively today for designing and studying chemical processes, and development of such programs is continuing (Motard et al., 1975). Sophisticated optimization algorithms have been developed and applied to unit processes, such as reactors (Barneson et al., 1970) and separators (Umeda and Ichikawa, 19711, or to simplified models of complex processes (Bracken and McCormick, 1968; Komatsu, 1968).

Yet, few studies have been reported of the optimization of large complex processes, modeled with flowsheet simulators. Shannon et al. (1966) optimized a single variable of the sulfuric acid process with PACER. Seader and Dallin (1972) also used PACER to study the toluene dealkalation process, with a case study search procedure. Friedman and Pinder (1972) opti- mized the production rate of a gasoline polymerization unit, modeled with CHESS. This same process was studied with PROPS by Gaines and Gaddy (1976), with the objective of maximizing the profitability of the process. Each of the above studies was conducted a t a university, and process optimiza- tion studies, using flowsheet simulators, if conducted in in- dustry, have gone unreported.

The reluctance by industry to utilize simulators for detailed optimization can probably be explained by a combination of factors. There is considerable uncertainty over the amount of real time and computing time required to perform a detailed process optimization study. Industry may lack the experienced personnel to work in this area. Bounded by tight time con- straints during design and construction. the necessary ap- proach may be to come as close to optimal conditions as pos- sible using experienced engineers. and optimize the operation of the facility after it is operating. Little information IS

available on the design improvements available with optimi- zation that might justify such studies.

Most companies now use simulators during some phase of the design, perhaps for material and energy balance data. Therefore, it seems reasonable that additional simulation runs could be justified for optimization to improve the design. Such reasoning must be based upon the availability of a dependable process model, probabiy with economics, and a suitable op- timization algorithm.

The purpose of this study is to develop a detailed simuiation model of a complex chemical process and to measure the in- vestment in real time and computer time for optimization of this process. The progress of the optimization is monitored so that the improvements obtained through optimization. even perhaps when starting very close to the optimum, can be determined. The process chosen for study is the intermediate pressure methanol process. This process is simulated with PROPS and optimized using the adaptive random search technique (Heuckroth and Gaddy, 1976).

T h e Methanol Process Figure 1 is a typical MeOH flow diagram, showing the

equipment used for producing methanol from natural gas. Natural gas is reacted with steam over a nickel based catalyst in a steam-methane reformer to produce a gas mixture con- taining hydrogen, carbon monoxide, and carbon dioxide. Since excess hydrogen is available from the reforming, carbon dioxide is added to the reformer feed to lower the ratio of hydrogen to carbon in the reformed gas. Reforming is carried out at temperatures between 1400 and 1700 OF. The steam methane reaction is endothermic, and the reactor is heated by burning a mixture of natural gas and process off gases.

Energy is recovered from the reformed gas and the com- bustion gases as 900 psia steam (500 O F superheat) in a waste heat boiler. This steam is used to drive the synthesis and recycle gas compressors. Exhaust steam from the compressor turbines is used both as process steam for the reformer and as a source of heat in the purification section of the process.

Centrifugal compressors are used to raise the pressure of the makeup and recycle gases to 200 atm. The synthesis gas (recycle plus makeup gas) is heated to 570 O F by the effluent gas from the methanol converter. Methanol is synthesized in the converter using a zinc oxide-chromium oxide catalyst with a copper promoter. The methanol synthesis reaction is highly exothermic. To control the temperature of the reaction, cool synthesis gas is injected between the catalyst beds. The gases from the converter exchange heat with the converter feed and are then cooled to atmospheric temperature to condense methanol.

The pressure of the condensed liquid is reduced to 50 psia and dissolved gases are removed. A significant amount of the methanol is vaporized with the dissolved gases, and the MeOH is recovered by scrubbing with water. The crude methanol is then purified by distillation.

Mehta and Ross (1970) and Shah and Stillman (1970) have reported optimization studies of the methanol process. Mehta and Ross studied only the effect of Cor, addition to the pro- cess. Shah and Stillman developed a FORTRAN modei of the major equipment items in the methanol process. They studied thp effect of several variables on the control of the process; however. only six iterations were used in the optimization procedure.

Methanol Process Simulation The methanol process, shown in Figure 1, was modeled

using PROPS (Process Optimization System), a modified version of CHESS. PROPS has the capability of computing the investment, operating cost, revenue, and profitability of the simulated process. A description of PROPS has been

Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 3, 1977 337

Page 2: Ballman

M E T H A N O L

PRODUCT FRACTIONATOR L E T DOWN TANK

Figure 1. Flow diagram of the 200 atm methanol process.

presented earlier (Gaines and Gaddy, 1974) and will not be reviewed here.

Standard equipment modules in PROPS were used for modeling exchangers, separators, compressors, etc. in the MeOH process. In addition, modules were developed for: steam-methane reformer, waste heat boiler, boiler feed-water heater, methanol converter, partial condenser, optimization procedure, and production rate control. The details of the simulation model are given by Ballman (1976) and are sum- marized in the supplementary material. (See the paragraph a t the end of the paper regarding supplementary material.)

As usual, modeling of the reactors required the greatest effort. The reformer model uses a modified Newton relaxation method to solve the algebraic material and energy balance equations to obtain the exit compositions and enthalpy. Re- actor conditions are also checked to avoid carbon deposition. The methanol converter consists of four catalyst beds with interstage injection of cool synthesis gas to control bed tem- peratures. The differential equations resulting from the ma- terial and energy balance are solved simultaneously using a fourth-order Runge-Kutta method.

As noted in Figure 1, the process has a recycle loop that must be converged. Other iterative loops within the process include selection of the proper synthesis gas splits between catalyst beds, selection of reformer conditions to avoid car- bonization, as well as numerous dew point, bubble point, and temperature from enthalpy determinations. Convergence acceleration using the secant method is used in each case.

In addition to the above iterative calculations, further computations were required to ensure a fixed production rate. A simulation study normally begins calculating in the first equipment item with a given input of raw materials. In opti- mization, where certain variables are being manipulated, the quantity of product varies for a specified feed rate. Therefore, optimization of a particular MeOH plant size imposes further iterative computations to determine the proper quantity of natural gas feed. These computations were guided by a pro- duction rate control module, which uses either a direct ratio for linear systems or a secant convergence acceleration for nonlinear systems.

Once all recycle calculations are converged, the economic calculations are made to determine equipment and operating cost and profitability. A complete simulation of this model for a fixed production rate requires about 40 s of CPU time (IBM 370/168). Development of the model required about six man-months of a semi-experienced engineer’s time.

Verification of the Process Model A study of this type, conducted in academia, is not privy to

the best industrial data, and must depend upon published

Table I. Results of Base Case Study

Independent Variables 1. Reformer operating temperature, O F (Vl ) 2. Reformer operating pressure, psia (V2) 3. H20/CH4 ratio reformer feed (V3) 4. CO*/CHd ratio reformer feed (V4) 5. Synthesis loop pressure, psia (V5) 6. Synthesis loop purge (V6) 7. Methanol converter feed temperature, OR (V7) 8. CO conversion methanol converter (V8)

Dependent Variables 1. Investment, MM$ 2. Working capital, MM$ 3. Earnings after taxes, MM$ 4. Payout period, years 5. Return on investment, % 6. Methanol production, tons/day 7 . Natural gas reformer feed, mol/h

1540.0 350.0

3.2 1.00

3000.0 0.01

1030.0 0.29

28.8 3.9 3.4 4.93

12.1 800.0

2600.0

sources. The methanol process is reasonably well documented, and no amount of detail was spared in using the available data to create a representative model. To establish that a realistic simulation model had been created, data obtained from the simulation were compared with data published in the litera- ture. The steam-methane reformer model was found to give equilibrium computations in agreement with those given by Hougen et al. (1966b), Morse, (19731, and Wellman and Kate11 (1963). The methanol converter model produced reactor temperatures and material flow rates which agree with those given by Natta (1955), Strelzoff (19701, and Shah and Stillman (1970).

Design conditions given by Shah and Stillman (1970) and Strelzoff (1970) were used to establish a base case design. The results of this run for a plant to produce 800 tons/day (TPD) are given in Table I. For the design conditions shown, a plant investment of $28.8 M M was computed and the return on investment was found to be 12.1%. Hedley (1970) gives a capital cost of an 800 T P D plant of $26.4 M M (corrected to 1976). The economic evaluation given by Hedley shows a 10.0% return on investment. Comparison of these data confirm the ability of the model to predict realistic values of the pro- cess economics.

Optimization Procedure Other studies of the methanol process reactions have used

the following independent variables: reformer operating temperature and pressure, the ratios of steam to methane and carbon dioxide to methane in the reformer feed, and methanol converter feed temperature (Shah and Stillman, 1970; Mehta and Ross, 1970). In addition, the following variables have been identified in this study: synthesis loop pressure, synthesis loop purge, and conversion of carbon monoxide in the methanol converter. These eight independent variables are labeled as V1 through V8 in Table I.

The return on investment, ROI, was used as the objective function in optimizing the methanol process. The optimization can be stated as:

Max ROI = f ( V1, V2, . . . , V8) (1)

subject to the constraints: 1600 I V1 I 2200; 100 I V2 I 500; 1 I V3 I 6; 0 I V4 I 1; 2500 I V5 I 3500; 0.01 I V6 I 0.15; 1000 5 V7 I 1030; 0.15 I V8 50.35; D 1 , 0 3 , D 5 , 0 7 I 1100; 1000 I D2,D4 , D6 I 1030; 0 8 , D9, D10 < 1.

The function, f, in eq 1 represents all the computations performed by the equipment modules, the equipment design module, and the economic evaluation module in the PROPS simulation. Constraints on the independent variables are given in eq 1. Implicit constraints involving dependent vari-

338 Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 3, 1977

Page 3: Ballman

Table 11. History of Optimization of 800 Ton/Day Methanol Process

are encountered in the search region. Also, this procedure, although random, has been found to have equal or better ef- ficiency when compared with other methods, and demon- strates good reliability in locating the optimum (Heuckroth et al., 1976; Kelahan and Gaddy, 1977).

CPU time, Function Objective function Step min evaluations improvements

1 2

24 15

11 9

3 2

3 10 9 4 4 10 12 0 5 15 13 5 6 20 14 2 7 8 5 1 8 15 8 1 9 10 5 2

10 10 6 1 11 20 13 1 12 20 8 3 13 8 4 1 14 15 8 3

Total 200 125 29 Total" 158 97 19

Results of search using a different starting point.

ables, D, in eq 24 represent temperatures in the methanol converter and carbon deposition ratios in the reformer. Con- straint boundaries were set from the usual values found in the literature.

The Adaptive Random Search procedure (Gaines and Gaddy, 1976; Heuckroth et al., 1976) was used to guide the selection of new values of the independent variables in the search for the optimal value of ROI. This procedure searches randomly in a restricted region about the best known objective function. The Adaptive Random Search was selected because of its ability to function effectively when implicit constraints

Results and Discussion Four optimization searches were performed in this study.

Plant sizes of 800 T P D (two searches from different starting points) and 1000 T P D were studied. In addition, a run was made with fixed feed rate to determine the variations in pro- duction rate that could be expected and the savings in com- puter time that might be possible if the feed rate were known, or could be estimated. Sensitivity runs were made about the optimum in each case.

Since a large amount of computer time was expected to be required for each study, the search procedure was halted pe- riodically to review the progress of the search. This procedure allows adjustment in the parameters of the search algorithm and allows a periodic judgement on the convergence of the search.

1. Optimization of 800 TPD MeOH Plant. Table I1 shows the history of the search for the optimum design conditions for an 800 T P D plant. The search was conducted in 14 steps requiring from 8 to 24 min per step and a total of 200 min to locate the optimum. A total of 125 evaluations of the ROI function (eq 1) were made, for an average of 1.6 min per evaluation. Twenty-nine improvements in ROI were found. Similar results are noted for the search from a different starting point.

The optimal results are given in Table I11 (last two rows). A t the optimum, several of the variables lie on or near their boundaries. Reformer temperature and pressure and meth-

Table 111. Results of Optimization of 800 Ton/Day Methanol Process

v 1 v2 v 5 V7

"R psia V3 x 10' psia x 102 "R X 10' Tons/day ROI x lo-,' x 10-2 V4 x 10-3 V6 x 10-3 V8

2.0000 3.5000 3.5000 5.0000 1.9997 3.5107 3.4998 5.5233 2.0830 4.7279 3.4345 5.1401 2.2000 5.0000 2.7837 3.4295 2.2000 5.0000 2.8909 3.4295 2.2000 5.0000 2.8906 4.4308 2.1941 3.8393 2.7739 8.2982 2.1941 5.0000 2.3804 8.3951 2.1938 5.0000 2.3674 8.3356 2.1266 5.0000 2.3259 7.8053 2.1266 4.9988 2.3751 5.1592 2.1265 4.9461 2.3751 5.1592 2.1266 4.9575 2.3751 5.1564 2.1266 4.9575 2.3817 5/73 2.1267 4.9570 2.5244 4.8012 2.1266 4.9570 2.7056 4.6308 2.1266 4.9570 2.7448 6.9891 2.1484 5.0000 2.7144 4.5092 2.1915 5.0000 2.7144 10.0000 2.2000 5.0000 2.5618 10.0000 2.2000 5.0000 2.3957 10.0000 2.2000 5.0000 2.5618 9.9699 2.2000 5.0000 2.5620 9.9700 2.2000 5.0000 2.5620 9.9700 2.2000 5,0000 2.5620 9.9700 2.2000 5.0000 2.5620 9.9700 2.2000 5.0000 2.5620 7.4700 2.2000 5.0000 2.5620 6.9700 2.2000 5.0000 2.5620 6.1700 2.2000" 4.9500" 2.2677" 6.2594"

" Results of search using a different starting point.

3.0000 3.1204 3.1787 3.4947 3.4947 3.4979 3.3500 3.3497 3.1145 3.1371 3.1354 3.1354 3.1255 3.1255 3.0413 3.0128 2.8411 2.8204 2.8451 2.8447 2.7496 2.8456 2.5009 2.6546 2.6555 2.5000 2.5000 2.5000 2.5000 2.5000"

4.0000 4.1968 4.1966 3.9441 3.6139 4.5111 4.5110 4.4878 4.4870 2.7331 3.0588 3.0591 3.0592 3.0582 3.3124 2.2251 1.9637 2.2232 2.2232 2.2201 2.7565 2.4591 1.2391 1.2391 1.4565 1.0000 1.0000 1.0000 1.0000 1.0000"

1.0150 1.0214 1.0149 1.0290 1.0300 1.0300 1.0300 1.0164 1.0163 1.0143 1.0147 1.0101 1.0164 1.0146 1.0146 1.0146 1.0146 1.0141 1.0279 1.0139 1.0139 1.0139 1.0130 1.0130 1.0130 1.0130 1.0130 1.0130 1.0130 1 .oooo "

~~~~ ~

2.5000 2.5132 2.6789 3.1173 3.1618 3.1766 3.1766 3.1765 3.1745 3.4603 3.4654 3.4654 3.4649 3.4652 3.4652 3.4652 3.4652 3.4420 3.4459 3.4459 3.4547 3.4459 3.4459 3.4459 3.4459 3.4459 3.4459 3.4459 3.4459 3.2640"

806 6.827 807 7.349 802 10.221 788 10.497 796 10.838 807 10.866 804 11.351 804 12.351 796 12.994 812 13.165 804 13.242 803 13.258 798 13.696 810 13.923 810 13.929 798 14.626 805 15.017 797 15.449 809 15.534 791 15.609 791 15.636 807 15.885 798 16.509 796 16.542 80 1 17.289 811 18.431 811 19.22 803 20.00 806 20.03 807" 19.94"

Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 3, 1977 339

Page 4: Ballman

Table IV. Sensitivity Study for 800 Ton/Day Methanol Process. Variable and ROI Values

v1 ROI V2 ROI v 3 ROI v4 ROI v5 ROI V6 ROI V7 ROI V8 ROI

2200.0

500.0 20.03

20.03 2.66

19.15 0.717

19.90

18.17

17.60

19.50

20.03

2900.0

0.040

1019.0

0.344

2160.0

480.0 19.35

19.63 2.56

20.03 0.667

19.92

18.06

18.25

19.28

20.03

2800.0

0.030

1016.0

0.334

anol conversion are a t their upper limit, while synthesis loop pressure and recycle purge are a t their lower limit. Converter temperature and steam and carbon dioxide to methane ratios are intermediate. No attempt was made to adjust the boundaries of these variables during this study, although this step would certainly be indicated for an industrial study.

The most significant cost items are raw materials and compression; therefore, i t is expected that the combination of variables would tend to minimize these costs. High tem- peratures and low pressures favor conversion in the reformer. However, low reformer pressure increases compression costs. Therefore, the combination of maximum temperature and pressure is not unexpected and in accordance with current design practice for reforming in ammonia synthesis, although somewhat different from other MeOH studies (Shah and Stillman, 1970). Case studies about the optimum, listed in Table IV, show the ROI to be particularly sensitive to a re- duction in temperature. Lower pressures reduce ROI, but less significantly, as expected.

Low synthesis pressure and high conversion per pass lower compression costs but increase the size of the converter. Low purge rates conserve raw materials, but slightly increase compressor cost. Therefore, optimal selection of the minimum pressure and purge rate and the maximum conversion are reasonable for the synthesis loop. Table IV shows the ROI to be particularly sensitive to changes in pressure and purge rate. The optimal result of intermediate values of steam and carbon dioxide to methane ratios and near minimal converter inlet temperature are in accord with the findings of Shah and Stillman (1970).

The values of the variables for the 29 improved search points are also given in Table 111. The ROI was improved from an initial value of 6.8 to an optimal value of 20.3. In this case, optimization resulted in additional net earnings of about $4.5 MM per year, for an investment in computer time of about $2500. As expected, the fastest improvement is obtained early in the search, the first half of the search reaching an ROI of 15 percent. The last hour and a half of the search resulted in improving the ROI about 4 percentage points. This im- provement represents a saving of about $1 MM per year for a minimal investment in computer time. For this plant, am improvement of 1% in ROI represents about $250 000 per year in net earnings. The engineering time (6 man-months) to create the model might cost as much as $20 000. Therefore, the engineering time and computing time can easily be justi- fied by even a slight improvement in ROI. Clearly, optimiza- tion with a detailed simulation model is one of the best in- vestment opportunities available to the chemical company.

2120.0 18.40

19.25 19.50 2.46 2.36 2.26

19.96 20.29 20.62 0.617 0.567 0.517

20.03 18.94 18.86

18.98 19.15 20.03

18.64 20.03

20.03 19.28 19.50

18.92 19.20 18.33

460.0 440.0

2700.0 2600.0 2500.0

0.020 0.010

1013.0 1010.0 1007.0

0.324 0.314 0.304

The experienced engineer may be able to design much closer to the optimum than the starting points in this study, possibly selecting the proper boundary for those variables which were optimal a t a boundary. Such a selection for all five bounded variables is somewhat questionable, however, in view of the findings of other studies and the usual design conditions listed in Table I. Observation of the progress of the search in Table I11 shows that even after all variables reached a bounded condition, the ROI was still improved about 2%. Table IV shows that very small changes in some variables have a pro- nounced effect on the ROI. Therefore, considerable process improvement may be possible, even when starting much closer to the optimum.

Optimization of a process after it is operating is certainly important. However, unless the initial design is first opti- mized, many opportunities may be lost. For example, once the compressors are ordered, the opportunity to change pressure, purge or reactor conversion may be lost, or a t least, tightly constrained. The greatest flexibility in optimization is avail- able during the early design period. Therefore, the preparation of the process simulation model should be one of the first objectives. Other advantages, of course, acrue from having the model, such as development of control strategies, studying synthesis alternatives, evaluating effluent problems, etc.

The accuracy of the simulation must be considered in de- ciding how far to carry the optimization procedure and in es- tablishing the credibility of the results. In all of the iterative computations, convergence errors are introduced. These errors can be reduced by specifying tight tolerances, of course, at the expense of computer time. The production rates shown in Table I11 indicate the variability allowed in this study. Tol- erances were specified to control the production rate within 1.5% (12 TPD). This same level of variability would be re- flected in the ROI; therefore, convergence of the optimization should not be attempted below the tolerance specified for the material and energy balances (0.3% ROI in this study). Looser tolerances might be used early in the search, with some saving in computer time.

The results of the second optimization run of the 800 TPD plant are shown as the last entry in Table 111. The objective functions agree within the specified tolerance, as do the values of the independent variables. The agreement of these two cases confirm location of the optimum.

2. Optimization of 1000 TPD MeOH Plant. The search history and sensitivity studies for a 1000 TPD plant are given in Tables V, VI, and VI1 of the supplementary microfilm. The history of this search shows that in the first hour, the ROI improved from 7 to 18%. Another 1.5 h of searching resulted

340 Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 3, 1977

Page 5: Ballman

in a maximum ROI of 22010. Again, the improvement in process economics clearly appears to justify the expenditure of engi- neering and computer time. As expected, the values of the independent variables a t the optimum are the same as for the 800 T P D plant. The improvement in ROI from 20 to 22% is, of course, due to the economies of scale of the larger plant.

3. Optimization of MeOH Plant with Fixed Feed Rate. The results of optimization of a MeOH plant with a fixed feed rate of 1350 mol of methane per hour are shown in Tables VIII, IX and X of the supplementary material. This study was performed without use of the production rate control module. The optimal values of the variables were nearly identical with those of the larger plants.

Variations in the production rate for the improved search points were found to be a t most f6%, indicating tha t a suc- cessful optimization might be performed without iterations to control the production rate. However, the CPU time for this run was about the same as for the other studies. The time for each ROI computation is reduced, as expected, for the fixed feed case (1 min compared to 1.6 min); but the search was much slower in locating valid search points. This can be ex- plained by the fact that the ROI is a much smoother function lor a fixed plant size. Therefore, it becomes easier to locate values of the variables that will improve the economics when the plant size is fixed.

I t may be concluded that while the computer time is re- duced in evaluating the function with a fixed feed rate, the time for optimization cannot be expected to improve. This is probably true, regardless of the optimization algorithm em- ployed, when ROI or earnings are chosen as the objective function.

Conclusions I t may be concluded that a detailed flowsheet simulation

model of a complex chemical process can be used for optimi- zation without use of unreasonable amounts of computer time. The expenditure of engineering time to prepare the model and computer time for optimization represents one of the best investment opportunities available. The lack of experienced personnel to do optimization studies may be a realistic barrier, but one that can only be overcome by practice.

Nomenclature DJ = exit temperature of J t h catalyst bed ( J = 1, 3, 5 , 7) ,

DK = inlet temperature of the K t h catalyst bed ( K = 2 , 4 ,

D8, D9, D10 = critical ratios of carbon deposition

Literature Cited Ballman, S. H., M. S. Theis, University of Missouri, Rolla, Mo., 1976. Barneson, R. A . , Brannock. J. C., Moore, J. G.. Morris, C., Chem. Eng., 77 (16),

Bracken. J., McCormick, G. P., "Selected Applications of Non-linear Program-

Friedman, P., Pinder, K. L., Ind. Eng. Chem., Process Des. Dev., 11, 512

Gaines, L. D., Gaddy. J. L. , Ind. Eng. Chem., Process Des. Dev., 15, 1 (1976). Gaines, L. D., Gaddy, J. L., "University of Missouri-Rolla PROPS User's Guide,"

Hedley. B., Powers, W., Stobaugh, R. B., Hydrocarbon Process, 50 (9), 275

Heuckroth, M. W., Gaines. L. D.. Gaddy, J. L., AIChE J., 22 (4), 744 (1976). Hougen. 0. A.. Watson, K. M., Ragatz. R. R., "Chemical Process Principles,"

Kelahan. R. C., Gaddy, J. L., "Variable Elimination for Optimization with Difficult

Komatsu, S., Ind. Eng. Chem., 60, 2 (1968). Mehta, D. D., Ross, D. E., Hydrocarbon Process., 49 (1 l), 183 (1970). Morse, P. L., Hydrocarbon Process., 52 (l), 113 (1973). Motard, R. L., Shacham, M., Rosen, E. M.. AlChEJ., 21 (3). 417 (1975). Natta, G., "Synthesis of Methanol," "Catalysis," P. H. Emmett, Ed., Voi. 3, pp

Seader, J. D., Dallin, D. E., Chem. Eng. Computing, Vol. 1. AlChE Workshop

Shah, M. J., Stillman, R. E., Ind. Eng. Chem., Proc. Des. Dev., 62 (12), 59

Shannon. P. T., Johnson, A . I., Crowe, C. M., Hoffman, T. W., Hamielec, A. E.,

Strelzoff, S., Chem. Eng. Prog. Symp. Ser., 66 (98), 5 5 (1970). Umeda. T., Ichikawa, A.. Ind. Eng. Chem., Process Des. Dev., IO, 229

Wellman, D., Katell. S.. Hydrocarbon Process Pet. Refiner, 42 (6), 135

I3

61, O R

132 (July 27, 1970).

ming," Wiley, New York, N.Y., 1968.

(1 972).

University of Missouri-Rolla, Mo., 1974.

(1970).

2nd ed, pp 986, and 1048. Wiley, New York. N.Y., 1966.

Inequality Constraints," submitted to AlChEJ., 1977.

394-41 1, Reinhold, New York. N.Y., 1955.

Series (1972).

(1 970).

Woods, D. R., Chem. Eng. Prog., 62 (6), 49 (1966).

(1 97 1).

(1 963).

Receit'ed f o r reciew June 2 2 , 1976 Accepted M a r c h 25, 1977

Presented a t t he 81st N a t i o n a l M e e t i n g of t he Amer i can I n s t i t u t e of Chemical Engineers, Kansas C i t y . Mo . , April 1976.

S u p p l e m e n t a r y M a t e r i a l Avai lab le. M o d e l flowsheet s imulat ion ( l i pages, i nc lud ing 6 tables). Orde r ing i n fo rma t ion is g iven o n any current masthead page.

The Crystallization and Drying of Polyethylene Terephthalate (PET)

Brian D. Whitehead

Zimmer Aktiengesellschaft, 6000 Frankfurt am Main 60, West Germany

Following a brief discussion of PET crystallization theory, the theory of PET drying is discussed in detail, together with a review of the required data. The application to practical processes and associated problems is then re- viewed, before summarizing fu ture developments.

Introduction This paper discusses the crystallization and drying stages

necessary before the final processing of polyethylene ter- ephthalate (PET) to finished products. The underlying theory of these two unit operations will be summarized, together with a review of the necessary physical property data. Then the problems encountered in drying processes will be discussed.

Finally, future developments in PET drying technology will be outlined.

The The production and processing stages of PET are still

mostly carried out at separate locations or even by separate companies. Even within one company, it is rare to find that

for Drying PET

Ind. Eng. Chem., Process Des. Dev., Vol. 16, No. 3, 1977 341