Sunteți pe pagina 1din 9

Computers chem. Engng Vol. 20, No. 4, pp.

315-323, 1996

Pergamon
0098-135405)00022-4

Copyright ~) 1996 Elsevier Science Ltd Printed in Great Britain. All rights reserved ' 0098-1354/96 $15.00+0.00

O P T I M I Z A T I O N AS A TOOL F O R D E S I G N / C O N T R O L INTEGRATION
J. D. PERKINSt and S. P. K. WALSH Centre for Process Systems Engineering, Imperial College of Science, Technology and Medicine, London SW7 2BY, U.K. (Received 4 February 1994; received for publication 3 November 1994) Abstract--There is a long tradition of the use of optimization techniques to help solve process synthesis and design problems. Advances in optimization algorithms to handle problems involving integer variables, and problems involving dynamic systems with path constraints have recently been exploited to help address issues in the integration of process design and control. Methods, based on optimization, to assess controllability, to select control structures for a given process (the control system synthesis problem) and to develop integrated designs of process and control system for cases where dynamic performance is critical are presented in this paper.

INTRODUCTION

The challenge of process design is to develop a processing system that may be operated safely and reliably with acceptable environmental impact to give a reasonable return on investment. A key feature of any process design problem is the presence of uncertainties--in demand for the products, in economics, in the behaviour of the equipment once installed etc. In addition, the plant will be subject to disturbances during operation, causing variability in performance. It is the features of uncertainty and variability that motivate an integrated approach to process design and process control. These features may be analysed both from a steady-state and from a dynamic perspective. In the former case the corresponding requirement of the processing system is called flexibility, in the latter

employed by Morari, but will introduce two subcategories of process synthesis: the selection of control structure for a fixed process, and the integrated design of process and control scheme. In the next section, we will discuss controllability analysis. In subsequent sections, work on the control structure selection problem, and on the integrated design of process and control systems will be described, The bias will be towards presenting our own work in these areas. A recent review of the field by one of the authors is also available (Morari and Perkins, 1994). A unifying theme of the work presented will be the exploitation of recent advances in optimization techniques, both in mixed-integer programming and dynamic optimization.
2. CONTROLLABILITY ANALYSIS

controllability.
In this paper, we shall limit our attention to an analysis of dynamic effects. However, through the use of optimization as an analysis tool, it will be possible to establish some links with flexibility analysis. Morari (1992) presents a comprehensive review of work investigating the interactions between process design and controllability. He points out that the majority of work to 1992 had been concerned with controllability evaluation, that is the assessment of the dynamic characteristics of a given process system. Far less work had been reported on process synthesis taking account of dynamic requirements. In this paper, we shall adopt the same categories as t To whom correspondence should be addressed. 315

The assessment of the dynamic performance of a process and control system and of their ability to cope with variability and uncertainty in an effective way can be a time-consuming activity. This is particularly true if the "brute force" approach of developing the process and control system designs to a point where dynamic simulation can be used to explore the effects of disturbances and uncertainties on performance is employed. To reduce the amount of effort involved in this assessment, a number of tests have been developed which may be applied to process systems without detailed control designs in place. The rationale behind this approach is that there are inherent limitations built into the process equipment which no control system, however sophisticated, will be

316

J. D. PERKINSand S. P. K. WALS~t White et al. (1994) [following earlier work by Howell (1984)[ examine the effect of inherent plant characteristics on switchability (i.e. the ability of the plant to cope with large mode changes in an effective manner). Their approach is based on determining the optimal switching trajectory for the plant, by setting up and solving an optimal control problem. Howell's early work was limited by an inability of optimal control algorithms available at the time to handle path constraints on switches in an effective manner. Recent advances in these algorithms (e.g. Vassiliadis, 1993) permit path constraints to be handled directly. As a result, the ability of a plant to deliver satisfactory performance may be unambiguously assessed. White et al. show the results of a switchability analysis on a complex air separation unit which is required to cope with 100% changes in demand as quickly as possible. One feature of the optimization formulation of White et al. is the ability to include parameters characterizing the design of the plant as decision variables. In this way, it is possible to determine design parameter values leading to optimal switching. For example, White et al. present a binary distillation column design case study, where to maximize switching performance from low purity to highpurity products, an optimal design with minimum total column holdup is suggested. The effects of time delays and bounded parametric uncertainty on disturbance rejection capability may be assessed by analysing the performance of an idealized controller under worst-case conditions (Walsh and Perkins, 1992). The controller is idealized in that it is assumed to act perfectly as soon as knowledge of any disturbance becomes available to it. (Of course, such knowledge is delayed by the presence of the time delays in the system.) Any real controller will deliver performance worse than this idealized case, so inadequate performance of the idealized controller implies inherent limitations in the plant. The role of optimization in this approach is in the determination of the worst-case disturbance from the set of possible disturbances defined for the plant, again an optimal control problem. Walsh and Perkins (1992) present an analysis of a waste water treatment plant design where many process and control options could be rapidly screened by the application of this simple test. The approach of checking satisfaction of constraints with an idealized controller over a set of uncertain parameters can be considered as an extension of the approach to steady-state flexibility analysis of Grossmann et al. (1983) to include consideration of dynamics. In Grossmann et aL's approach

able to overcome. Analysing the effects of these limitations early enough in the design process gives the opportunity to modify the design should the effects be critical to performance. In 1983, Morari (1983) identified the following process characteristics which limit the achievable control performance independent of controller design: manipulated variable constraints, nonminimum phase behaviour (right half-plane zeros and time delays) and plant-model mismatch. Each of these effects has been analysed, and a number of techniques are now available to assess their potential impact on closed-loop performance. Morari (1992) gives a comprehensive review of the methods, and of published applications to that time. The majority of the methods are based on analysis of linear models; significant effort has been expended on validating the results of linear analysis on a variety of realistic (nonlinear) process systems. As a result, confidence in the ability of the methods to give useful results has grown, and several industrial applications have been published (e.g. Roat et al., 1986; Bouwens and K6sters, 1992; Oglesby et al., 1992). Whilst the usage of the existing techniques is growing, they all suffer from two basic weaknesses. Since the methods provide indicators of the likely effect of each attribute separately on closed-loop performance, at best they may be used to provide a relative ranking of a number of design alternatives. However, as pointed out by Morari (1992), in all but the simplest situations it is essentially impossible to provide a definitive rank-order, since the ranking depends in a complex way on the performance requirements for the system. An attraction of using optimization techniques as a tool to develop controllability tests is that these two weaknesses may be overcome. First, it has been shown that it is possible to devise absolute controllability tests by this means, that is tests which if failed imply that no real plant based on the tested partial design can meet the performance requirements. Second, it is possible to include the actual performance requirements into the test itself. As an added bonus, the use of optimization permits nonlinear models to be handled directly, obviating the need for linearization and removing the doubt introduced as to the validity of results based on linear approximation. Of course, the need to reliably solve the necessary optimization problems should be set against these advantages. It is only recently that dynamic optimization codes with the necessary attributes (reliable performance, ability to handle path constraints) have begun to appear, making possible the approaches to be described next.

Optimization as a tool certain process variables are designated as "operating variables", which are allowed to take different values for each set of uncertain parameter values. The feasibility condition becomes: min max min max [ck(p, v, o)] ~<0,
pc/" I,~V oEO
keK

317

(1)

where p is a vector of design variables, v is a vector of uncertain variables, ck is the kth element of the constraint vector c and o is a vector of operating variables which may be adjusted to maintain feasibility in the light of the value of v. The design variables are chosen to allow constraint satisfaction for all the uncertain variable values while the operating variables are adjusted for each value of the uncertain variables. The motivation for including operating variables is that there are variables which may be adjusted during plant commissioning or operation to give improved performance in the light of the actual plant behaviour. Requiring such variables to be chosen so as to accommodate all possible uncertain variables, i.e. as design variables, introduces an element of conservatism to the design. This is particularly the case if the operating variables include process inputs which would be adjusted by a control scheme to maintain satisfaction of constraints and if the uncertain parameters are fixed but unknown rather than variable. On the other hand, including operating variables in the problem formulation assumes an ideal adaptation of the operating variables to all the uncertain variables. Arguably, this is a reasonable assumption in a steady-state context. However, once the uncertain parameters are allowed to be time-varying, it is difficult to formulate the equivalent of equation (1) in such a way as to disallow the adaptation of operating variables based on perfect knowledge of the future variation of these parameters. If the natural generalization of equation (1) is employed, the conclusions from a dynamic analysis will be optimistic, since any real adaptation mechanism will have only partial knowledge of the past and little knowledge of the future behaviour of the uncertain parameters. Failure of a design problem with operating variables indicates that no control scheme for adjusting the operating variables can achieve feasible operation for all the uncertain parameters. Success of such a design problem does not imply that an implementable control scheme exists which can achieve feasible operation. Operating variables should not therefore be used in the final determination of design parameters, though they may be useful in certain screening tests if the resulting problem can be solved efficiently.

To go beyond the potential conservatism of having all the variables as design variables and the optimism of using operating variables, it is necessary to include the adaptation mechanism (control scheme) within the model. If desired the parameters controlling this adaptation can be made design variables. Grossmann et al. (1983) consider this "would make the problem virtually unmanageable" at the design stage. However, including basic control information may be as simple as requiring that certain variables remain at their setpoints, which may be added to the design variables, and eliminating the operating variables o using the extra equality constraints. The optimization problem without operating variables is easier to solve as one level of optimization is eliminated, and it gives a measure of the ability of a particular control structure to suppress the effect of uncertainty. The benefits of this approach are achieved at the expense of either having to select manipulated and measured variables prior to using this analysis (e.g. using techniques discussed in Section 3) or having to optimize over the possible structural choices for the control scheme.
3. CONTROL STRUCTURE SELECTION

The choice of suitable sets of manipulated variables, control objectives, measurements and control algorithms interconnecting the two sets is a challenging process synthesis problem, and a variety of techniques and tools have been suggested to help solve it (Stephanopoulos, 1983). In contrast to other process synthesis problems, the control structure selection problem has until recently not been attacked using a systematic optimization approach. One problem has been to quantify the economic costs and benefits of control. To circumvent this, various authors have resorted to a multi-objective problem formulation (Lenhoff and Morari, 1982; Palazoglu and Arkun, 1986, 1987; Luyben and Floudas, 1992). Recently, Narraway et al. (1991) have formulated a model which permits the estimation of the economic benefits of a given control scheme for a continuous process working in a given disturbance regime. The method is based on standard techniques of control benefits analysis (e.g. Marlin et al., 1991), but can be used during design to analyse the relative benefits of different plant options. The basic concept is illustrated in Fig. 1, The size of the "ball" around chosen steady state operating conditions for the plant depends on the disturbance regime, and the control system implemented. The steady-state operating point should be chosen so

318

J. D. PERKINSand S. P. K. WALSH

~"OptJmm"StmdyState ]

ady State

Feasible d

were any variable from these sets to be selected for the control structure. A disturbance regime is defined, and an optimal control structure is sought which minimizes the sum of the economic give away implied by the ball-size analysis (Fig. 1) and the costs of instrumentation associated with the control system. Narraway et al. (1993a, b) have presented two special cases within this general framework. The first is based on a linear(ized) analysis, and approximates the performance of the controllers using the perfect control hypothesis. With this approach, whatever measurements are chosen are assumed to be perfectly regulated; disturbance effects do not appear in these variables. Further, all disturbances are assumed to be sinusoidal in nature. With these assumptions, the problem is almost a mixed-integer linear programming problem. All nonlinearity is localized in the calculation of the ball-size for a fixed configuration. However, this nonlinearity includes discontinuities in first derivative due to the calculation of moduli of complex variables. A special algorithm, making use of the properties of this nonlinearity, is presented by Narraway et al. (1993a) and tested on a number of case studies. Not surprisingly, it has been found that the control structure chosen by the method is critically dependent on the disturbance regime assumed. However, the relationship between disturbances and optimal control structure is not simple. While there are cases which conform to the intuitive idea that increasing the "richness" of the disturbances motivates increased levels of instrumentation and control (Heath and Perkins, 1994), there are other cases (Narraway, 1992) where including more disturbances leads to less control! An attempt to address nonlinear systems directly, and to include realistic controller models, has been presented by Narraway and Perkins (1993b). Here, a multiloop proportional plus integral control structure was assumed, which increases the combinatorics of the problem since pairings between measurements and manipulated variables must be selected in addition to the variables themselves. A highly simplified disturbance regime, consisting of a single sinusoidal disturbance was assumed in this study. The formulation results in a mixed-integer optimal control problem, which was tackled by a version of the outer-approximation augmented penalty algorithm (Viswanathan and Grossmann, 1990). An MILP MASTER problem was set up and solved at every iteration to provide a new candidate control structure. This structure was evaluated by solving an optimal control.problem to determine the best controller tunings and the resulting economics.

DEPENDS ON Disturbances Controller

Fig. 1. The economics of process control.

that all points in the bali are feasible with respect to operating constraints, and so that the predicted economic performance is as favourable as possible. The ideal situation is illustrated in Fig. 1, where the centre of the ball is as close to the steady-state optimum in the absence of disturbances as possible. The calculation of a suitable operating point for a given ball size and disturbance regime may itself be formulated as an optimization problem (an LP in the case of linear objective, plant and constraints and a fixed disturbance). However, the ability to quantify economic benefit permits a more ambitious optimization to be a t t e m p t e d - - t h e optimal control structure for a fixed plant and disturbance regime (Narraway et al., 1993a, b). The problem formulation is illustrated in Fig. 2. A set of candidate manipulated variables and measurements is given, together with costs of the instrumentation that would be necessary

Problem specification: Select control structure such that disturbances are rejected without process constraint violation in an economically optimal manner
Potential disturbances

manipulated
variables

Process

measured variables

Fig. 2. Control structure selection: problem definition.

Optimization as a tool

319

~ -----~ Air supply

~ Concentrate

Fig. 3. Three cell froth flotation circuit.

The algorithm performed well for the three-cell froth flotation problem of Narraway et al. (1991), shown in Fig. 3. The optimal control structure (confirmed by exhaustive enumeration) was found from 9 out of 10 initial structures. A more demanding case study is shown in Fig. 4. Here, the performance of the algorithm was less satisfactory, with 10 different solutions being found from 20 starting points! The fact that all solutions gave economic performance within 1.2% of the best solution found is reassuring. Nevertheless, this behaviour suggests significant nonconvexity in the formulation, and that improvements in the formulation or in MINLP algorithms' ability to handle nonconvexities are needed before this approach can be relied on.
4. INTEGRATED PROCESS DESIGN

designers manage the trade-off between risk and return on investment and how well they anticipate potential operational problems. Research on inte-. grated design aims to provide tools, procedures and environments to aid engineers in tackling these issues in a systematic and well-coordinated manner. In our work on integrated design, we use an approximation to the design problem above which is amenable to computer aided design. The objective is taken to be: maximize return on investment while ensuring that performance constraints are satisfied for all possible plant parameters and disturbances. A number of assumptions are made in adopting and implementing this approach: 1. The return on investment can be estimated adequately. This is a key problem in process design, common to all design approaches. In many cases a simplified target is adopted, e.g. minimize annualized cost while maintaining the capacity to meet a range of production and quality targ:ets. 2. The performance requirements can be represented by either constraints or a contribution to the operating costs. Representing the design requirements in this form is clearly an approximation. For example, the requirement that the plant must operate safely may be approximated by a set of constraints on process variables, but these will never precisely capture the safety requirement. This representation of the problem must therefore be open to review in the light of design results. 3. Uncertainty and disturbances can be described in terms of bounds on process characteristics. This assumption is necessary as our formulation requires that constraints be satisfied for all possible plant parameters and disturbances. In many cases this is a natural way of specifying the

Engineers are required to produce designs which will operate safely and legally at all times and provide an adequate return on investment. They must meet these requirements despite incomplete knowledge of the process and equipment characteristics and the operating demands on the plant. Success or failure will depend on how well the

T,P

Steam

N-Tc, Pc VV1

VV2

VV3

Feed Liquor

LV2

Ploducl ~ Is, Fw, X

Fig. 4. Double effect evaporator.

320

J. D. PERKINSand S. P. K. WAt.SH

requirements, e.g. maximum feed variability or maximum measurement error will often be part of a design specification. In other cases the choice of bounds involves careful judgement, e.g. possible values for rate coefficients or tray efficiencies. As with constraints, the uncertainty description must be reviewed in the light of design results. We assume the value of each uncertain parameter is limited only by upper and lower bounds. Time-varying uncertain variables constrained to lie within an envelope over time are handled by parameterizing the time variation as a piecewise linear function and translating the envelope to bounds on the individual parameters defining this function. By reviewing the problem definition, the project team can manage the trade-off between risk and return on investment at a very high level. Our experience is that defining the uncertainty in terms of bounds on process parameters provides an intuitive approach to dealing with uncertainty with which project engineers are quite comfortable. The approach also facilitates the development of clear operational specifications in terms of process constraints, required feed properties and equipment performance. There are design problems for which this objective is particularly challenging, and to achieve it involves a careful analysis of the combined dynamics of process and control system. A well-known example is the design of chemical effluent treatment facilities, which have traditionally involved proper utilization of the skills of process and control engineers to ensure effective designs. This particular design problem has become more challenging recently with a move towards 100% compliance to environmental constraints on discharges, As a result, not only is the disturbance regime for such plants particularly harsh, but "product" specifications have become tight and hard constraints. Also, the system is notoriously nonlinear, and there is significant uncertainty in the process parameters (e.g. associated with mixing, and titration characteristics of the mixtures to be processed). A systematic procedure to handle waste water neutralization problems has been proposed by Waish and Perkins (1994) and is summarized in Fig. 5. The method consists in successive refinement of the process and control system design based on application of a sequence of increasingly complex analyses. The controllability analysis involves application of the time delay test discussed in Section 2. Later

Select reagent based

on

rulesl )

Estimate numberand size of tanks for disturbance rejection(controllability)

conslraintsand use reactorengineering to modifyas needed

Check steady-state 1

for detailedanalysisusing I worst c a s e dynamic [ optimisafion j "


I Modifydesign

usingrulesand judgement

Fig. 5. Integrated design procedure for waste water neutralization systems. stages involve application of worst-case design methods based first on steady-state, and later on dynamic models. The worst-case design problems are addressed using an outer approximation algorithm (Fig. 6). In this context outer approximation refers to the use of a number of uncertain parameter scenarios to define a feasible region for design which contains (is an outer approximation to) the true feasible region. The algorithm alternates between two optimization problems:

Choose initial scenario~


of the uncertain parameters | J

Optimise design for all scenarios identified

1_

J-

Add constraint 1 maximiser to design scenarios

i Maximise constraint l violation over the uncertain parameters

[Solution

Fig. 6. Worst-case design algorithm.

Optimization as a tool

321

.----,
f

feedforward trim
i- . . . . . . . . . . . . . . . . .
- ' ~

~ .......

,' ,". . . . . . . . . . . . . . . "~ F B ,~. . . . . . . ', ""1. . . . '---~ . . . . ,,-~-, ',, - - .I---:- ~ ca-.,......... . . . 1~ . . . ', , F , I
. . . . . . .

"r"
__l__j

;'- --:--I FI3 ,"--'


,
........

. . . . .

i
o. . . .

u o
_1 s-I" ~

: - - I F B :---

f ....

",
I~
Ca(OH)2 PFR

"" r " /

NaOH

v ~

"r"

I><]
effluent CSTR 1 CSTR 2

Fig. 7. Final process and control system for the example.

1. O p t i m i z i n g the design variables while satisfying the c o n s t r a i n t s associated with all scenarios selected. 2. Finding the values of the p a r a m e t e r s , within t h e i r defined ranges of uncertainty, which maximize the m a x i m u m c o n s t r a i n t violation. T h e initial scenarios c o n s i d e r e d in the design m a y be chosen based on engineering judgement.

Alternatively, local maximizers of individual constraints may be f o u n d using c o n v e n t i o n a l local search techniques a n d included in the initial set of scenarios if the constraint is violated. T h e second optimization p r o b l e m is used to update the scenarios c o n s i d e r e d in the design p r o b l e m until no c o n s t r a i n t violation can be found. It should be n o t e d t h a t the c o n s t r a i n t m a x i m i z a t i o n p r o b l e m

pH
7.50 7.40 7.30 7.20 7.10 7.00 6.90 6.80 6.70 6.60 6.50 II scenario 2i ,, ............................ :""tfieF;................. 1si...................... :- -"
.... ~ , ..............
il r ........................... 'I|%1' ' blip' ~'vm"*"--'-i~lll

iiiiiiiiiiiiilliiiiiiiiiiiiiiiiiiiiiiiiiii iii
. . . . . . . . . . . . . . . . . . . . . . . . . . , . . . . . . . . . . . . . . . . . . . . . . . .

, ........................

o...

~
r,# ~, . .......................

,,
,

lll'~'h~,t~'~':--

~ ........................

,...

hours

0.00

1.00

2.00

3.00

Fig. 8. Extremal scenarios considered in dynamic worst-case design.

322

J . D . PERKINS and S . P . K . W

LSH

Table 1. Problem specification for a waste water neutralization system design Performance specification Final pH between 6.5 and 7,5 at all times (ideal for downstream precipitation) pH sctpoints between 2 and 12 (avoids degraded measurement) Total residence time less than 30 min (allows treatment system to bc sited in main plant structure) Steps in flow from 0-50%, 50-100%, 100-25% at 1 h intervals at concentration of about 12 M acid (worst-case disturbance scenario based on judgement) Annualized cost is minimizing number of CSTRs and total working volume for a given number of CSTRs. Uncertainty specification +0.25 pH bias pH measurement lag between 5 and 30 s (CSTR mounted probes) Titration curve slope between 100 and 250 pH/M near target _+40% uncertainty in Ca(OH)2 reaction rates in CSTR Additional 2:1 uncertainty in reaction rate in in-line reactors + 1% error on flow and concentration measurements + 2.5% error or reagent composition

S P A R C II workstation are s h o w n for each analysis. The final d y n a m i c worst-case design which optimizes e q u i p m e n t sizes and controller tunings o v e r a range of scenarios for disturbances a n d uncertain parameters typically c o n s u m e s h o u r s of C P U time, a n d covers a wide range of o p e r a t i n g conditions (see Fig. 8). T h e dynamic model used in the example presented has 115 state variables and 307 equations. A n u m b e r of industrial design case studies have b e e n carried out using this p r o c e d u r e , o n e of which has led to a p a t e n t application (Walsh et al., 1992). 5. CONCLUSIONS T h e r e is a n o t a b l e t r e n d towards the use of optimization as a tool for d e s i g n / c o n t r o l integration. This t r e n d is e n a b l e d by advances in c o m p u t a t i o n a l h a r d w a r e a n d optimization m e t h o d s a n d driven by the n e e d to place control design decisions on the same basis as process design decisions, i.e. economics and feasibility, a n d by the n e e d to systematically address the effect of u n c e r t a i n t y a n d o p e r a t i n g variability on design. In controllability analysis, the n e e d to i m p r o v e on the relative and a m b i g u o u s n a t u r e of earlier measures can be a d d r e s s e d by posing a range of optimization p r o b l e m s with varying degrees of idealization of the controller r e p r e s e n t a t i o n . Ideal delay-limited control and optimal control can be used to give estimates of, or b o u n d s on, the achievable control p e r f o r m a n c e in t e r m s of feasibility or the e c o n o m i c cost associated with m o v i n g away from the o p t i m u m steady-state to satisfy the constraints. T h e ideal delay-limited control and optimal control a p p r o a c h e s can b o t h be readily i n t e g r a t e d with analysis of the effect of uncertainty. As control systems are m e c h a n i s m s for a d a p t i n g m a n i p u l a t e d variables to c o m p e n s a t e for u n c e r t a i n t y a n d variability, the c o m b i n e d analysis can give more realistic estimates of p e r f o r m a n c e in the face of u n c e r t a i n t y t h a n e i t h e r treating m a n i p u l a t e d variables as fixed (design variables) or allowing the m a n i p u l a t e d variables to be optimally adjusted for each particular realization of the uncertain variables. Idealized controller r e p r e s e n t a t i o n s can also be applied to control structure selection based o n economics by f o r m u l a t i o n of an optimization p r o b l e m including the structural decision directly. For perfect control and linearized models the resulting optimization p r o b l e m can be solved effectively using M I L P techniques. In general, the control structure selection p r o b l e m will require M I N L P techniques. M I N L P techniques can also be applied to control structure selection with a p a r a m e t e r i z e d class of i m p l e m e n t a b l e controller such as multiloop PI control r a t h e r than an idealized controller.

will typically have multiple local m a x i m a with m a n y of the u n c e r t a i n variables lying at t h e i r b o u n d s for each m a x i m u m . This o b s e r v a t i o n implies t h a t a p p r o x i m a t e global o p t i m i z a t i o n must be used a n d provides the basis for heuristics to i m p r o v e efficiency of the optimization. Systematic overdesign can be used in the design optimization to reduce the n u m b e r of scenarios r e q u i r e d to o b t a i n a feasible design. Typically 2 - 3 m a j o r iterations, each adding a new scenario, are r e q u i r e d to find a solution. This algorithm is discussed in detail in Walsh (1993). A typical p r o b l e m f o r m u l a t i o n is shown in T a b l e 1, a n d the evolution of the design for this case is illustrated in T a b l e 2. T h e final design is s h o w n in Fig. 7. To give an idea of the complexity of the analysis at each stage, typical C P U times on a S U N

Table 2. Evolution of the design Reagent selection ~arbonate-based reagents were eliminated based on final pH and foaming problems --Sludge formation was not a problem as suspended solids removal was availablc --Slaked lime was taken for further analysis as the cheapest reagent likely to perform adequately Controllability analysis --Delay of 10 s associated with mixing in CSTRs --At least two CSTRs required ~ P U timc,~ 1 s Steady-state worst-case design --2 CSTRs inadequate (85 min residence time) --Initial plug-flow reactor (up to 2 min) with feedforward and 2 CSTRs adequate (20 min residence time) --CPU time ~ 15 min Dynamic worst-case design --Ratio and feedforward control included based on rules --PI feedback adequate (27 min residence time) ~ P U time ~ 30 h

Optimization as a tool Preliminary experience with these techniques highlights the need for care in describing the disturbances to be considered and the need for improved methods of dealing with nonconvexity. Integrated process design attempts to deal systematically with the central design trade-off of risk and cost. The approach adopted centres on optimizing design parameters while ensuring constraint satisfaction for a bounded set of uncertain parameters. This worst-case design problem can be tackled using successively more realistic representation of the control, e.g. perfect steady-state control, ideal delay-limited control and optimal control, to screen designs. In the final stage of the design an implementable process/control scheme can be optimized with both process and control parameters being adapted simultaneously. This approach has been applied successfully to industrial waste water treatment system design problems which provide a simple but extreme e m b o d i m e n t of the issues of dynamics and uncertainty at the core of integrating process design and control. The use of optimization based on idealizations of the control design problem to allow integration of process and control system design is still in its infancy, but early results and trends in computational resources and algorithms suggest it is an approach which will have real impact on the design process. Improvements in algorithms for global optimization of both NLP, and MINLPs would enhance the effectiveness of the approach, as would improved efficiency and robustness of dynamic optimization methods. Certainly, it is a fertile area for future work.
REFERENCES

323

Bouwens S. M. A. M. and P. H. Kosters, Simultaneous process and system control design: an actual industrial case. In Interactions Between Process Design and Process Control (J. D. Perkins, Ed.), pp. 75-86. Pergamon, Oxford (1992). Grossmann I. E., K. P. Haleman and R. E. Swaney, Optimization strategies for flexible chemical processes. Computers chem. Engng 7, 439-462 (1983). Heath J. A. and J. D. Perkins, Control structure selection based on economics. In The 1994 IChemE Research Event, Vol. II, pp. 841-843. IChemE Publications, Rugby (1994). Howell J. M., The effect of energy integration on the switchability of chemical processes. PhD Thesis, University of London (1984). Lenhoff A. M. and M. Morari, Design of resilient processing plants I. A general framework for the assessment of dynamic resilience. Chem. Engng Sci. 37, 245-258 (1992). Luyben M. L. and C. A. Floudas, A multiobjective optimization approach for analysing the interaction of design and control. In Interactions Between Process Design and Process Control (J. D. Perkins. Ed.), pp. 101-106. Pergamon, Oxford (1992).

Marlin T. E., J. D. Perkins, G.W. Barton and M.L. Brisk, Benefits from process control: results of a joint industryuniversity study. J. Proc. Cont. 1, 68-83 (1991) Morari M., Design of resilient processing plants--Ill. A general framework for the assessment of dynamic resilience. Chem. Engng Sci. 38, 1881-1891 (1983). Morari M., Effect of design on the controllability of chemical plants. In Interactions Between Process Design and Process Control (J. D. Perkins, Ed.), pp. 3-16. Pergamon, Oxford (1992). Morari M. and J. D. Perkins, Design for operations. Presented at FOCAPD-94, Snowmass, CO (1994). Narraway L. T., Selection of process control structure based on economics. PhD Thesis, University of London (1992). Narraway L. T., J. D. Perkins and G. W. Barton, Interaction between process design and process control: economic analysis of process dynamics. J. Proc. Cont. 1, 243-250 (1991). Narraway L. T. and J. D. Perkins, Selection of control structure based on economics. Computers chem. Engng 18, $511-515 (1993a). Narraway L. T. and J. D. Perkins, Selection of process control structure based on linear dynamic economics IEC Res. 32, 2681-2692 (1993b). Oglesby M. J., T. I. Malik, S. Fararooy and V. Geake, Early stage process controllability assessment. In: Interactions Between Process Design and Process Control (J. D. Perkins, Ed.), pp 145-150. Pergamon, Oxford (1992). Palazoglu A. and Y. Arkun, Design of chemical plants in the presence of process uncertainty: a multiobjective approach. Computers chem. Engng 10, 567-575 (1986). Palazoglu A. and Y. Arkun, Design of chemical plants with multiregime capabilities and robust dynamic operability characteristics. Computers chem. Engng 11,205216 (1987) Roat S., J. Downs, E. Vogel and J. Doss, The integration of rigorous dynamic modelling and control system synthesis for distillation columns: an industrial approach. In Chemical Process ControI--CPC 11I (M. Morari and T. J. McAvoy, Eds), pp. 99-138. Elsevier, Amsterdam (1986). Stephanopoulos G., Synthesis of control systems for chemical plants--a challenge for creativity. Computers chem. Engng 7, 331-365 (1983) Vassiliadis V., Computational solution of dynamic optimisation problems with general differentialalgebraic constraints. PhD Thesis, University of London (1993). Viswanathan J. and I. E. Grossmann, A combined penalty function and outer approximation method for MINLP optimization. Computers chem. Engng 14, 769-782. Walsh S. P. K., Integrated design of chemical waste water treatment systems. PhD Thesis, University of London (1993). Walsh W. P. K., D. Annels and J. Harkin, International Patent W094/04467. Process for neutralisation of acid strength (1992). Walsh S. P. K. and J. D. Perkins, Integrated design of effluent treatment systems. In Interactions Between Process Design and Process Control (J. D. Perkins, Ed.), pp. 107-112. Pergamon, Oxford (1992). Walsh S. P. K. and J. D. Perkins, Integrated design of waste water neutralisation systems. In ESCAPE 4--4th European Symposium on Computer Aided Process Engineering, pp. 135-141. IChemE Publications, Rugby (1994). White V., J. D. Perkins and D.M. Espie, Switchability analysis. In Integration of Process Design and Control (E. Zafiriou, Ed.), pp. 254-259. Pergamon, Oxford (1994).

S-ar putea să vă placă și