Plantwide Control: Recent Developments and Applications

Part 1

Overview and Perspectives

1

Introduction

Gade Pandu Rangaiah¹ and Vinay Kariwala²

¹Department of Chemical and Biomolecular Engineering, National University of Singapore, Singapore

²School of Chemical and Biomedical Engineering, Nanyang Technological University, Singapore

1.1 Background

Industrial chemical plants and processes usually involve many types of operations and numerous items of equipment operating at different temperatures and pressures. Consequently, these plants are complex and also often large in size. The safe and optimal operation of industrial chemical plants requires the maintenance of critical operating conditions such as temperature, pressure and composition at their respective optimal values as well as within safe limits. This challenging task has to be achieved in the presence of known disturbances such as throughput and product specification changes arising from variations in the market demand and requirements, as well as unknown and unmeasured disturbances in raw material composition, catalyst activity, equipment conditions and environment. Hence, a reliable and extensive monitoring and control system is essential for the safe and optimal operation of modern chemical plants.

The monitoring and control requirements from the chemical plants have led to the development of process control as an important area within the Chemical Engineering discipline. Accordingly, the majority of undergraduate programs in Chemical Engineering throughout the world have a compulsory course on process dynamics and control. Further, many of these programs include an optional course on advanced process control. Many textbooks on process dynamics and control are available, a number of them into their second or even third editions (e.g., Ogunnaike and Ray, 1994; Marlin, 2000; Bequette, 2003; Romagnoli and Palazoglu, 2005; Riggs and Karim, 2006; Svrcek et al., 2006; Seborg et al., 2010). Advanced and specialized courses in process control such as model predictive control, digital control, robust control and nonlinear control can be found in the graduate programs in Chemical Engineering.

Numerous equipment in industrial chemical plants are inter-connected and operate together in order to achieve the desired process objective such as optimal production of a valuable product of desired quantity and quality from the raw materials. In effect, there are complex interactions between the equipment in chemical plants; these are increasing with energy and material integration and safety and optimization requirements (with consequent reduction in intermediate storage). A plantwide perspective is therefore crucial for synthesis and design of control systems for chemical plants, and this in turn has led to the development of plantwide control (PWC) as a sub-area within the broad topic of process control. This can be seen from the inclusion of one or two chapters in the more recent textbooks related to process control (e.g., Marlin, 2000; Skogestad and Postlethwaite, 2005; Svrcek et al., 2006; Seborg et al., 2010; Seider et al., 2010). There is also one book dedicated to plantwide control by Luyben et al. (1998). Another book on plantwide dynamic simulators by Luyben (2002) is also relevant and useful for PWC applications.

Figure 1.1 Biodiesel manufacture by transesterification of vegetable oil.

1.2 Plantwide Control

As an example of a typical chemical plant, consider the biodiesel production from vegetable oil by trans-esterification. The process flow diagram for this process is shown in Figure 1.1. This process has three continuous stirred tank reactors (CSTRs), two liquid-liquid phase separators, two distillation columns, a neutralization unit, a wash vessel and several heat exchangers. The process features a material recycle of un-reacted methanol and an energy recycle stream for energy conservation. The liquid-liquid phase separators can have very slow dynamics due to their large inventories. A suitable thermodynamic model is necessary for predicting phase behavior in the phase separators and distillation columns. Besides the product specifications, there are upper limits on the maximum temperature (i.e., in the reboiler) of the two columns in order to avoid decomposition of biodiesel and glycerol byproduct, which also necessitate vacuum operation. A plantwide control system needs to be synthesized and designed for the complex biodiesel process for its safe and optimal operation. It should consider and maintain product purities and operating constraints as well as smoothly change the throughput in response to the variations in the feed availability and/or product demand. In fact, a control system for this plant is synthesized and tested in Chapter 14 of this book.

Accordingly, PWC refers to the synthesis and design of a control system for the complete plant considering all aspects such as throughput changes and interaction between units affecting the safe and optimal operation of the entire plant. Interaction between units has been increasing with increasing energy and mass recycling due to process optimization and with reducing inventories due to safety concerns. The main focus in PWC is on the control system synthesis considering these interactions within the plant, and not on the design of a feedback controller (although it is one part of PWC). The key questions in the control system synthesis are: which variables should be controlled, which variables should be manipulated and how should these be paired? In other words, what kind of controllers are required and where should they be placed for safe, economic and sustainable operation of the plant? In a complete plant, there are numerous choices for both controlled and manipulated variables; PWC system synthesis is therefore a large combinatorial problem. It is also a complex problem since it should consider the dynamics of all equipment in the plant.

PWC typically deals with the synthesis and development of the regulatory layer of the control system and can include supervisory layer. The former consists of ubiquitous proportional-integral-derivative (PID) controllers which directly manipulate mass and energy flow to the equipment, for example, through control valves. For complete PWC design, parameters of these feedback controllers, ratio/cascade control loops and so on also need to be specified. Complexity of PWC is also evident from the numerous PID controllers in a typical plant. On the other hand, the supervisory layer has one or more model-based/predictive controllers providing set points for some of the PID controllers in the regulatory layer.

Interest, research and development in PWC can be traced back to Buckley (1964), who developed the first procedure for PWC. Most of the developments in PWC have occurred during the last two decades. Figure 1.2 shows the number of articles published in each year during the period 1990–2010. These data were obtained by searching by topic on Web of Science for the important keywords (plantwide control, plant-wide control and reactor separator recycle control) in the subject area of Chemical Engineering. The search has found many PWC papers known to us, but it has missed some related papers (e.g., on controlled and manipulated variables selection and pairing). Note that the data shown in Figure 1.2 include conference papers. In any case, Figure 1.2 gives a good indication of the research in the area of PWC. It is clear that PWC papers have been increasing since mid-1990s, with 30–35 papers published in each of the years 2008, 2009 and 2010.

Figure 1.2 Number of PWC articles published during the period 1990–2010.

1.3 Scope and Organization of the Book

PWC covers selection and pairing of controlled and manipulated variables, degrees of freedom, comprehensive methodologies, realistic applications and performance assessment of control systems designed. Obviously, it requires enabling techniques and tools for these such as steady-state/dynamic simulation and controller tuning. All these are covered in this book, with emphasis on recent research and development.

This book is broadly divided into five parts. Part I (Chapters 1 and 2) provide an overview and perspectives on research and development in PWC. Several tools and heuristics for carrying out subtasks of PWC design are presented in Part II (Chapters 3–8). Part III (Chapters 9–12) deals with systematic methodologies for design and evaluation of PWC systems. Various application studies are used to illustrate the wide applicability of these approaches in Part IV (Chapters 13–17). Some emerging topics within the scope of PWC are described in Part V (Chapters 18–20). Brief overviews of these chapters are presented next.

In Chapter 2, Downs provides an industrial perspective on the past and ongoing research activities in the area of PWC. It is emphasized that industrial acceptance requires design of control strategies, which are easy to understand and can be devised in a time-efficient fashion with limited information (e.g., steady-state model). These requirements often limit the application of analytical methods based on a detailed dynamic model in process industries. Furthermore, Downs highlights the need to develop tools for the important issue of identifying the most difficult disturbances to be handled by the PWC system.

Chapters 3–5 deal with the identification and pairing of controlled and manipulated variables; these decisions are collectively known as control structure design. In Chapter 3, Konda and Rangaiah point out that the traditional method of computing control degrees of freedom (CDOF) by subtracting the number of equations from number of variables is tedious and error-prone for large-scale processes. A simple method based on the concept of restraining number for identifying CDOF is discussed in detail and illustrated using several case studies ranging from simple units to industrial processes, including a carbon capture process.

In Chapter 4, Umar, Hu, Cao and Kariwala present the self-optimizing control (SOC) based method for systematic selection of controlled variables (CVs) from available measurements. The general formulation of SOC methodology and the local methods for quick pre-screening of CV alternatives are presented. Branch and bound methods, which allow the application of local methods to large-scale systems, are discussed. The detailed case study of the forced-circulation evaporator is used to illustrate the CV selection method.

In Chapter 5, Moaveni and Kariwala provide an overview of the key methods available for selection of pairings of controlled and manipulated variables. Pairing selection methods for linear time-invariant systems are classified as relative gain array (RGA) and variants, interaction methods, and controllability- and observability-based methods. Some recent methods for pairing selection for uncertain and nonlinear processes are also discussed. Several examples are presented in tutorial fashion to aid the reader’s understanding of the application of different methods.

In Chapter 6, Luyben presents some ‘common-sense’ heuristics which can aid the design of practical PWC systems for complex chemical processes. In particular, heuristics are presented for dealing with recycle streams and determining effective ways to feed the fresh reactant streams into the process. Some guidelines for tuning the PID controller for different loops (e.g., flow, pressure, level, temperature and composition) with a plantwide perspective are also provided. The toluene hydrodealkylation (HDA) process is used to illustrate the application of these heuristics.

In Chapter 7, Jagtap and Kaistha discuss the choice of the throughput manipulator (TPM). A heuristic for selecting the TPM for tight bottleneck/economically dominant constraint control and designing the PWC system around the selected TPM is suggested. The effect of the TPM choice on the economic performance of two realistic chemical processes is evaluated. It is shown that the suggested heuristic provides better economic performance than the conventional practice of using the fresh process feed as the TPM.

In Chapter 8, Downs and Caveness highlight that the PWC system is a mechanism to shift process disturbances and process variability from harmful locations to other locations that have less risk, harm or cost to the overall plant. Thus, viewing the process control system as a variability change agent can provide insights into PWC system development and analysis. Theoretical analysis and realistic examples are presented to signify the effect of choosing inventory location and size, TPM and strategies for managing recycle streams or the management of process variability.

In Chapter 9, Vasudevan and Rangaiah present a review of PWC design methodologies and applications. The available PWC methodologies are classified based on their approach and their brief overview is provided. The structure-based classification of PWC methodologies is also presented. The industrial processes considered in the reported PWC studies are listed together with their main features. Finally, PWC comparative studies performed to date are reviewed.

In Chapter 10, Vasudevan, Konda and Rangaiah present the integrated framework of simulation and heuristics (IFSH) as an effective and practical PWC system design method. The main emphasis of this methodology is the use of steady-state and dynamic simulations of the plant throughout the procedure to make the right decision from those suggested by heuristics. The IFSH procedure is illustrated on the modified HDA process featuring a membrane separator in the gas recycle loop. Analysis of the results indicates that the integrated framework builds synergies between the powers of both simulation and heuristics, to yield a stable and robust PWC structure.

Chapter 11 is on the PWC procedure of Skogestad. An important feature of this procedure is to start with the optimal economic operation of the plant and then attempt to design a control structure that implements optimal operation, while also considering the more basic requirements of robustness and stability. The procedure is split into a top-down part, based on plant economics, and a bottom-up part. The bottom-up parts aims to find a simple and robust ‘stabilizing’ or ‘regulatory’ control structure, which can be used under most economic conditions.

In Chapter 12, Vasudevan and Rangaiah present reliable quantitative criteria for comprehensively analyzing and comparing the performance of different PWC structures. These criteria include dynamic disturbance sensitivity, deviation from the production target, total variation in manipulated variables, process settling time and steady-state economic measure. These measures are applied to the PWC system developed for the modified HDA process in Chapter 10. The authors also provide some recommendations for comprehensive performance assessment of PWC systems.

In Chapter 13, Luyben considers control of an ammonia process containing multiple adiabatic reactors with ‘cold-shot’ cooling. It is demonstrated that a cooled ammonia reactor is much more economical because of lower-pressure operation (less feed compressor work), smaller recycle gas flow rates (less recycle compressor work) and recovery of the exothermic heat of reaction by generating steam. A PWC system is developed and shown to provide effective regulatory control for large disturbances.

In Chapter 14, Zhang, Rangaiah and Kariwala consider a biodiesel production plant. Different alternative designs for the production of biodiesel through alkali-catalyzed transesterification of vegetable oil are considered and a suitable design is selected. A complete PWC structure is then designed using the IFSH procedure and is shown to give stable and satisfactory performance in the presence of expected plantwide disturbances.

In Chapter 15, Huang, Chien and Lee discuss the design and control of reactive distillation processes. Two important operations (reaction and separation) are carried out in a single vessel in reactive distillation, which makes the control of this process difficult. For reactive distillation of ethyl acetate with homogeneous and heterogeneous catalysts, optimal designs are developed and PWC systems are designed systematically. The performance of the homogeneous catalyst process is considerably inferior as compared to that of the heterogeneous catalyst process due to slow reaction rate, which highlights the effect of process chemistry on the control performance.

In Chapter 16, Seki, Amano and Emoto design a control system for a multistage crystallization process that is part of the product recovery section of an industrial para-xylene production plant. Multiloop PID and model predictive controllers (MPCs) are designed for this process. Closed-loop simulations show the superior performance of MPC. The possibility of constraint switching using a steady-state optimizer to enlarge the feasible operation region is evaluated.

The economic PWC procedure discussed in Chapter 11 is applied to an off-gas system by Shang, Scott and de Araujo in Chapter 17. Dynamic models for the off-gas systems of a smelter’s roasters and furnaces are developed using fundamental principles. It is shown that the PWC system allows near-optimal economic operation of this process, while complying with environmental regulations by avoiding emission of hazardous off-gases to the atmosphere.

In Chapter 18, Bao and Xu study PWC from a network perspective. The process is modeled as a network of process units interconnected via mass and energy flow, and a network of distributed controllers is employed to control the process network. Modeling of the process and controller networks is discussed. The effects of the interactions between process units on plantwide stability are analyzed. Lastly, an approach is presented for control network design to achieve plantwide performance and stability, even when the communication system breaks down.

In Chapter 19, Seck and Forbes discuss approaches for distributed PWC. It is highlighted that co-ordinated distributed schemes provide a good trade-off between the advantages of the centralized and decentralized approaches. For co-ordinated PWC, overviews of price-driven resource allocation and prediction-driven schemes are provided. Two case studies, namely, a pulp mill process and a forced circulation evaporator, are used to illustrate the advantages and disadvantages of the different approaches.

In Chapter 20, Munir, Yu and Young propose eco-efficiency as a way to integrate process design and control. The thermodynamic concept of exergy is used to analyze the process in terms of its efficiency. The focus of this chapter is on input-output pairing selection using relative exergy array (REA), which measures both the relative exergetic efficiency and controllability of a process. Case studies involving distillation columns are used to show that the combination of RGA and REA can guide the process designer to reach the optimal control design with low cost.

Rigorous process simulators are being increasingly used in PWC studies. In the Appendix of this book, Vasudevan, Konda and Zhang share their experience on the use of Aspen HYSYS as part of their extensive PWC studies. Selected problems faced by them and the different solutions that they tried and employed to overcome the problems are presented. In addition, some general problems together with possible solutions are also discussed.

In summary, this book provides researchers and postgraduate students with an overview of the recent developments and applications in the area of PWC. It will also allow industrial practitioners to adapt and apply the available techniques to their plants. Contents of this book can be readily adopted as part of the second course on process control aimed at senior undergraduate and postgraduate students. The reader can also study chapters of interest, independent of the rest of the book.

References

Bequette, B.W. (2003) Process Control: Modeling, Design and Simulation, Prentice Hall, Upper Saddle River.

Buckley, P.S. (1964) Techniques of Process Control, John Wiley & Sons, New York.

Luyben, W.L. (2002) Plantwide Dynamic Simulators in Chemical Processing and Control, CRC Press, New York.

Luyben, W.L., Tyreus, B.D. and Luyben, M.L. (1998) Plantwide Process Control, McGraw-Hill, New York.

Marlin T.E. (2000) Process Control: Designing Processes and Control Systems for Dynamic Performance, 2nd edn, McGraw Hill, New York.

Ogunnaike, B.A. and Ray, W.H. (1994) Process Dynamics, Modeling and Control, Oxford University Press, New York.

Riggs, J.B. and Karim, M.N. (2006) Chemical and Bio-Process Control, Prentice Hall, Boston.

Romagnoli, J. and Palazoglu, A. (2005) Introduction to Process Control, CRC Taylor & Francis, Boca Raton.

Seborg, D.E., Edgar, T.F., Mellichamp, D.A. and Doyle, F.J. III (2010) Process Dynamics and Control, 3rd edn, John Wiley & Sons, Hoboken.

Seider, W.D., Seader, J.D., Lewin, D.R. and Widagdo, S. (2010) Product and Process Design Principles: Synthesis, Analysis and Evaluation, 3rd edn, John Wiley & Sons, New York.

Skogestad, S. and Postlethwaite, I. (2005) Multivariable Feedback Control: Analysis and Design, 2nd edn, John Wiley & Sons, Chichester.

Svrcek, W.Y., Mahoney, D.P. and Young, B.R. (2006) A Real-Time Approach to Process Control, 2nd edn, John Wiley & Sons, Chichester.

2

Industrial Perspective on Plantwide Control

James J. Downs

Eastman Chemical Company, Kingsport, TN 37662, USA

2.1 Introduction

There is a general notion that when a new plant is designed and built there is an orderly flow of information, an organized Gantt chart illustrating who will have done what by when and that properly skilled human resources needed to accomplish the design are available. If only the design process could be so neatly described! In reality, the industry and market demands on time are so intense that the design process is rarely so organized. When faced with the choice of ‘build it in limited time with limited knowledge’ or ‘don’t build it at all’ because of a rapidly changing marketplace, the engineering opportunity of studying a problem and developing the optimum answer may not be available. Even when faced with designing a world-scale facility that may be in operation for years to come, the prospect of coming on-stream a few months later than a competitor may destine a plant to be operated at less than full capacity for several years. Such risks drive the plant design process to omit steps that, to an otherwise intelligent engineer, may seem folly.

It is in this environment that methodologies for plantwide control must function. Theoretically correct but intractably time-consuming approaches will languish in their adoption by the process control design community. The tradeoff between optimality and practicality is a difficult assessment to make, especially when the pressures of a project timeline are in force. For plantwide control design guidance to be beneficial in an industrial design environment, that guidance must not only address time-efficiency constraints but must also be tolerant of limited data and information. Tom McAvoy was a proponent of the relative gain array partly because of its information efficiency; that is, with limited information input, the measure provided significant understanding and guidance.

The purpose of this chapter is to briefly describe the challenges and needs of the industrial engineering environment in the design of plantwide control systems. The fast-paced flow of engineering information during a project design may not allow for time-consuming studies once a project begins. For well-understood plants based on existing technology, such interruptions may be deemed unnecessary. At the same time for plants based on new technology, design data, process understanding and process objectives may be constantly updated and changed. The establishment of control objectives, identification of probable disturbances and the measures of fitness-for-use are often assumed as known entities. However, even this basic data may be unclear. The impact of the sensitivity of plantwide design methods to this uncertainty is discussed in Section 2.2. Section 2.3 highlights process disturbances as one of the most important challenges facing a plantwide control designer. Finally, academic contributions and the need for academically trained engineers as critical ingredients to moving this technology forward in coming years are addressed in Section 2.4.

Migrating the plantwide control design function from ‘this is how we’ve always done it’ to ‘this is a better way and here’s why’ is a huge step, but an important one. It will demand highly skilled engineers with a convincing argument to change the design methods in the face of historical precedence. The tools we develop to equip industrial designers must be robust to input information uncertainty, be relatively easy to use and be quick to deploy. Obtaining the technically correct solution will only be a part of the battles fought to change plantwide control designs.

2.2 Design Environment

Over the last few decades, the chemical industry has changed in both the demand for new plants and the design requirements for new plants. In the 1960s and 1970s, there was significant development of products based upon new chemistry. The growth of the petrochemical industry expanded using both existing chemistry and technology and newly developed chemistries to feed the growth of new products. The petroleum industry developed new processes to utilize their oil resources to the maximum extent while the chemical industry evolved into providing new products, the largest volume being primarily plastics. New plant designs could not be copied or based upon existing facilities, but instead were designed with new unit operations and new chemistries. This resulted in a need for the design technology to incorporate the control and operation of these plants into the design function. Today, large-scale plant designs are more often than not based upon existing plant operations or existing technologies. These designs use similar chemistries, unit operations and design philosophies as their predecessors. Product development efforts in today’s market are usually based upon modifying the properties of existing products or developing products based on existing market-available chemicals.

This industry environment drives company ‘new facility’ expansion plans that fall into two broad categories. One category is the building of large world-scale facilities based upon known chemistry and existing plant designs. This type of expansion is fueled by economies of scale rather than by new chemistry, new plant design technology or even new products. This capital is investing to simply meet a growing demand for existing market-proven products. Costs for these plants are large and the risk must be commensurately low.

The other category of new plants is for products that have no current production. Often specialty chemicals that fall into this category are designed to fill a niche market with a hope that the market will grow. For this type of design, there is no existing plant or operation on which to base design choices. Because of the unknown market demand, these plants are usually much smaller scale and may be designed by incorporating existing idle equipment. These plant designs are usually driven by ‘time-to-market’ and ‘robustness to unknown process and chemistry characteristics’. Once chemistry is verified in the laboratory to produce a product with the desired properties, then the speed with which that can be replicated in an industrial process is paramount. There may not be time to fully explore and develop optimal processing conditions or needed design information. Plant designs that are robust to missing or incomplete information are needed to get the product to market. Additional design information can be gathered from the non-optimal but operating facility should the market bear out the need for more production.

The duration for which a new facility may produce a product is also changing. For world-scale plants that are designed and built to squeeze out older, more expensive, plants, the design expectation is that the new plant will be operating for many years. However, for smaller plants designed based on new technology with the hope of a developing market, there is much less certainty that production will last. Consequently, there is much more desire to economize on the design and base design decisions on a much shorter time horizon. This leads to less investment into process development beyond that which simply guarantees a product can be produced. For plantwide control design, this means that less process information may be available. The process control design engineer must be able to quickly identify the most critical issues that need to be addressed for the development of a control strategy that stabilizes the process and allows for safe operation.

This design environment drives the plantwide control question more towards the use of heuristics and guidelines and away from rigorous (perhaps more analytical) design methods. The time investment required to develop and study complex dynamic models of a prospective process is often not available. In addition, the perceived benefits compared to the time and costs of more detailed and rigorous approaches are often in question. The sensitivity of a control system design to the process uncertainties may lead one to prefer a less than optimal, but more conventional, plantwide control design. Compare the current industrial practice of simply tuning PID controllers. While there is a wealth of literature detailing optimal tuning procedures, analysis of loop performance and stability analyses, the vast majority of PID loops are tuned using heuristics and relatively simple-to-apply guidelines. This behavior is driven by concerns not much different than those driving plantwide control designers to base their designs on heuristics and experience.

Control strategy designs that are understandable are also desired. During the rapid conceptual design of a process and its control philosophy, the need for straightforward easy-to-understand control strategies is usually the dominant driver. A simpler strategy that can be easily explained has much less prospect of being changed as the design is carried forward. This is especially important for plant designs that may be outsourced to a third party for detailed design or to a joint venture partner who carries a design forward. Once the plant is up and running, support of the control system can depend upon the simplicity of the strategy. Because of changing product specifications or even changes in the products themselves, control strategies must often be changed from the original design. These changes may need to be done by local contractors or support organizations that have little formal process control training. If the location of the new plant is far away from the process control experts, then the likelihood of a complex plantwide strategy becoming unworkable is high.

2.3 Disturbances and Measurement System Design

The identification of the disturbances that a control system must handle is one of the most important yet least addressed issues. Occasionally it is important for the control system to track targets that move. Much more often, the control system is required to reject disturbances that affect the process. Plantwide control design is highly dependent on the disturbances that are assumed. The characteristics of the disturbances such as frequency, magnitude, point of entry and measurability are often unknown and often unknowable. If the answer obtained from a plantwide control system design procedure can be changed by simply assuming a different disturbance scenario, as is often the case, then the design problem has only been transformed from one difficult problem to another. This highlights the need for the assessment of the disturbance portfolio that the control system must handle.

The process control engineer has the responsibility to work with the design engineer to develop plant designs that are robust in absorbing, redirecting or rejecting plant disturbances. We often design process purges to remove trace components, even though we may not know what those components are or where they come from. Measurement systems capable of identifying the amount of purge needed to remove the offending impurities while not wasting valuable product is an example of design and control efforts working together to design for disturbances. Plant designs routinely have in-process inventory to provide natural process breaks. The sizing and use of these inventories or the strategic placement of additional inventory is clearly a joint decision among project stakeholders including the plantwide control designer. Such inventories are included most often because the flow and composition disturbances traveling through a process are unknown at the design stage; however, it is this uncertainty that prompts us to add such items to the process design and to include them in the plantwide control strategy design.

The capability to identify the most difficult disturbances for the control system to manage would be a valuable step in the design process. If the most offending disturbances can be mitigated by additional equipment or if they can be measured, then options can be generated to minimize the effects of the disturbances. The effort to detect and take preemptive action to handle a disturbance cannot be evaluated if we cannot assess its impact on the operation of the plant. At the steady state, the control system structure determines how the disturbance is transformed into a new operating point. Regardless of the dynamics of the transformation, if the new operating point is not a desirable one then the response of the control strategy to that particular disturbance would be considered weak. One cannot overstate the importance of confidence that the plantwide control strategy will always take the plant to an acceptable, perhaps optimal, operating point after a persistent disturbance enters the plant.

Understanding the character of the expected disturbances leads to a need to include the development and design of the process measurement system. This is important on three levels: (1) to strategically locate conventional measurements to reliably monitor the state of the plant and maintain stable operation; (2) to assess whether the plant is producing a product with the desired properties; and (3) to detect disturbances before they negatively impact operation. For new plant designs, the placement and availability of measurements may be based upon what was available in the laboratory rather than by an assessment of control system needs. The graceful degradation of control performance as measurements are removed and the definition of a minimal set of measurements needed for plant operation are issues seldom addressed in a procedural fashion. Similarly, the improvement in plant operation if the (n + 1)th measurement is added is difficult to capture. A design philosophy of ‘each item purchased must be justified’ can lead to the need for such information.

Inference of process information based upon combining several measurements can also play an important role in the plantwide design function. The assessment of product properties seldom relies upon a simple continuous measurement. More often a combination of measurements, augmented by sampled off-line data, provides needed product properties. Plantwide control systems that cannot cope with delayed or variable off-line measurements are destined to be ineffective. Just as the cost of adding additional measurements is questioned, so is the frequency of off-line data analyses. Not only are the one-time capital costs of measurement systems important, but the ongoing costs of the staffing needed to obtain off-line laboratory measurements can also come into question.

2.4 Academic Contributions

The development of plantwide control design methodologies has been an area of research in its own right during the last few decades. Many traditional process control technologies can also be brought to bear on the plantwide problem. The main thrusts have been a heuristic approach and a mathematical programming approach. Recently Konda et al. (2005) proposed combining these two basic approaches. The heuristic approach is based upon the experience of what has worked and gives good guidance on how to think about the plantwide issues in a formal way. The mathematical programming approach (e.g. Kookos and Perkins, 2001; Chen and McAvoy, 2003) casts the plantwide problem as a mathematical problem statement and leverages known technology to find a mathematical solution. Each of these approaches has its strengths. The applicability of plantwide control design technology depends on the nature of the plantwide problem. While complex, some problems demand simplicity and the time-to-solution is the overriding concern; other problems allow more time for study and can justify a more detailed analysis.

A key issue in the application of plantwide control design technology is the availability of a model. While it is often taken for granted that a model of some sort exists or can be generated, this one issue will often tip the scales toward a heuristic approach. For most plant designs, steady-state models suitable for equipment design will be available. However, extending the steady-state model to become a dynamic model is usually a large effort. Extending a basic dynamic model to include the detail needed for dynamics that occur with a time constant of less than 0.25 hr is very time consuming because of the amount of detail that must be added to the model. In addition, simply getting the model to run usually requires some control strategy to be generated during the model building phase.

The steady-state model used for design is sometimes little more than a basic material balance model with assumed reactor conversions, heat exchanger outlet temperatures and splitter blocks with assumed component split fractions. These models are of limited use for control design. If richer models based upon more first principles are available, then there are plantwide steady-state analyses that are relatively easy and useful. For example, if reaction kinetics are known then steady-state recycle policies based upon the chemistry can be determined (Ward et al., 2004, 2006). This can significantly improve a plantwide control strategy over a heuristics-only design. If additional model rigor incorporating rate- and composition-dependent performance of unit operations is available, then the concepts of self-optimizing control (Skogestad, 2000, 2004) can be applied using the steady-state-only model. These technologies allow designers to use the incremental process knowledge to develop incrementally improved strategies without having to resort to an all-or-nothing approach.

For a small number of unit operations, such as a subset of a larger plant, a dynamic model can be developed and explored. This is often necessary when unit operations are sufficiently unique to make heuristics inapplicable. There are numerous control techniques to successfully address subsets of the plantwide problem. Often much of the control strategy design can be accomplished by using only simple rules and then applying a mathematical technique to address a control loop interaction or other difficult control issues.

Clearly the heuristic approach makes the most sense for process design efforts that are short on information and detail, demand a rapid design-to-startup timeline and may only produce the designed-for product for a limited time. Contributors (Buckley, 1964; Luyben et al., 1998) have outlined the steps to arrive at suitable strategies that are relatively robust to unknown issues. In fact, a more widespread application of these procedures would be welcome and would be an improvement to the often ad hoc designs that may result from short design-to-startup timelines. For the design of large-scale facilities that employ known technology and chemistry but have perhaps a tighter knit recycle structure, the time and effort of an analytical study may be of benefit. For existing facilities for which production is limited by plantwide variation, the inability to determine the optimum process conditions given current disturbances or the need to adjust many process targets simultaneously, an analytical approach can be justified. An analytical approach can take into account, in an organized manner, process interactions that can be distributed throughout a process. Such interactions can be modeled and verified from the existing process operation. Furthermore, once a process is operational opportunities for improvement can be much more evident than in the design phase. Verification can also provide needed confidence in problem solutions that may be difficult to fully understand or explain.

Demonstrating analytical approaches on small problems is useful for demonstrating the concepts; however, the benefit comes from their application to large problems for which heuristics may be incomplete. Application to such large plant designs requires a high degree of skill and training. For these techniques to be of benefit, there must be a simultaneous meeting of: (1) the plantwide control design need; (2) the availability of the people trained to apply the analytical approach; and (3) the corporate support for the time and costs of completing the plantwide control assessment and design.

This confluence of need, skill and support is probably unlikely for most companies. As these more mathematical techniques become more defined and understood, their use will increase. The need for process simulator suppliers to embrace such technology is a prerequisite. The natural progression from concept to steady-state simulator model to decisions on plantwide control is centered upon the modeling of the process. This modeling requires physical properties, kinetic descriptions, equipment performance descriptions and so on. The assembling of this process data and development of simply a steady-state model are huge tasks that can stretch the design process. The additional step of extending this basic design to the dynamic level can be overwhelming.

2.5 Conclusions

The challenges and needs of the design environment for the design of industrial plantwide control systems have been discussed. The design environment for new plants can be characterized by short time lines, conservative design policies and limited talent. The development of plantwide control design technology that can function in this environment is important to the continued improvement of control systems. A renewed focus on the disturbances for which we must design can provide a rallying point for the appropriate time and energy to devote to the plantwide control design problem. The confidence to ensure disturbance scenarios will be gracefully managed and that the control system will be ‘right the first time’ is of clear value. Training engineers to understand the strengths and weaknesses of the strategies they put in place will allow them to discriminate between what can be accomplished quickly with limited information and what demands a more in-depth study. With this information to hand, the relentless tide of time pressure can be stemmed when needed.

References

Buckley, P.S. (1964) Techniques of Process Control, John Wiley & Sons, New York.

Chen, R. and McAvoy, T. (2003) Plantwide control system design: methodology and application to vinyl acetate process. Industrial & Engineering Chemistry Research, 42, 4753–4771.

Konda, N.V.S.N.M., Rangaiah, G.P. and Krishnaswamy, P.R. (2005) Plantwide control of industrial processes: an integrated framework of simulation and heuristics. Industrial & Engineering Chemistry Research, 44, 8300–8313.

Kookos, I.K. and Perkins, J.D. (2001) An algorithm for simultaneous process design and control. Industrial & Engineering Chemistry Research, 40, 4079–4088.

Luyben, W.L., Tyreus, B.D. and Luyben, M.L. (1998) Plantwide Process Control, McGraw-Hill, New York.

Skogestad, S. (2000) Plantwide control: the search for the self-optimizing control structure. Journal of Process Control, 10, 487–507.

Skogestad, S. (2004) Control structure design for complete chemical plants. Computers & Chemical Engineering, 28(1–2), 219–234.

Ward, J.D., Mellichamp D.A. and Doherty, M.F. (2004) Importance of process chemistry in selecting the operating policy for plants with recycle. Industrial & Engineering Chemistry Research, 43, 3957.

Ward, J.D., Mellichamp, D.A. and Doherty, M.F. (2006) Insight from economically optimal steady-state operating policies for dynamic plantwide control. Industrial & Engineering Chemistry Research, 45, 1343.

Part 2

Tools and Heuristics

3

Control Degrees of Freedom Analysis for Plantwide Control of Industrial Processes

N.V.S.N. Murthy Konda¹ and Gade Pandu Rangaiah²

¹Centre for Process Systems Engineering, Department of Chemical Engineering, Imperial College London, South Kensington Campus, London, SW7 2AZ, UK

²Department of Chemical and Biomolecular Engineering, National University of Singapore, Engineering Drive 4, Singapore 117576

3.1 Introduction

Control degrees of freedom (CDOF) refers to the number of variables that can be manipulated independently in a given process. It is one of the first steps in the design of plantwide control systems. To understand CDOF, it is important to understand the concept of degrees of freedom (DOF) within the context of mathematical modeling. In principle, any system can be represented by a set of mathematical equations containing a number of variables. DOF of a system is the number of system variables, whose values must be specified before the remaining variables can be calculated (Felder and Rousseau, 2005). It is given by:

(3.1)

If DOF is zero, the model equations can be solved in principle; even then, it may not be possible to solve the equations for certain specifications, which result in unequal number of unknown variables and equations for one or more subsystems. On the other hand, if DOF is greater than zero there are more unknown variables than independent equations relating them, and values of DOF variables must be specified before values of all the remaining variables can be determined. In this case, the problem has infinitely many solutions. Hence, it is possible to perform process optimization. Finally, if DOF is less than zero, there are more independent equations than unknowns. In this case, the problem is over-specified and cannot be solved. In the DOF analysis, it is important to ensure that the equations are independent. They are called ‘independent’ if any of the equations cannot be obtained by any combination of the other equations (adding, subtracting, multiplying, etc.).

Within the context of chemical processes such as petrochemical processes, petroleum refineries, mineral processes, pharmaceutical processes, food processing, and fine and specialty chemicals, types of equations involved in the model include mass, energy and momentum balances, chemical/thermal equilibrium equations and heat/mass transfer-based rate equations. For instance, if two streams are in equilibrium, the chemical potential of each component in both the streams is equal. If the streams are not in equilibrium, they will obey mass-transfer relationships. Regarding the variables, types of variables depend on the purpose and scope of the model. Three common situations that may arise are discussed below.

If the purpose of the model is to design a process/equipment, typical DOF include physical specifications of the equipment (e.g., reactor volume, number of trays in a distillation column) and operating conditions (e.g., temperature, pressure, composition, flow rates). In this case, these DOF are referred to as design degrees of freedom (DDOF).

If the model is built for control purposes, typical DOF include flow rates (i.e., control valve openings), electrical power and mechanical speed-related inputs. These DOF are referred to as CDOF.

On the other hand, if the purpose of the model is integrated design and control, both the types of DOF considered in the earlier cases are included.

Depending on the context, DOF analysis can be helpful in answering the following types of questions on design and control (Pham, 1994).

How many operating conditions or desired outputs can we specify?

How many control loops are required?

Can we optimize the process by changing some design and/or operating conditions?

DOF analysis can therefore be useful in a number of ways in different contexts (i.e., during the design and/or operational stages). Since this chapter deals with CDOF, it is discussed in further detail in the following sections; DDOF is mentioned briefly whenever deemed appropriate. The next section describes CDOF in more detail. Section 3.3 presents a critical review of methods for determining CDOF available in the literature. Section 3.4 presents the method based on restraining number proposed by Konda et al. (2006) and applies it to many standard units encountered in chemical processes. Section 3.5 presents the application of this method to relatively complex units such as distillation columns with reboiler and condenser, and discusses the concept of redundant variables. Some large-scale applications involving processes of varying complexity, including a CO2 capture process, are presented in Section 3.6. Finally, conclusions are given in the last section.

3.2 Control Degrees of Freedom (CDOF)

The main purpose of control is to operate the process efficiently and safely at the desired steady state, which is usually inferred from the controller set points for various process variables. This is achieved by manipulating some variables. However, not all process variables can be varied simultaneously, as they need to abide certain constraints imposed by the governing equations (such as mass and energy balances). Thus, only a certain number of variables can be manipulated. In mathematical terminology, these variables are independent variables while the rest are dependent variables decided by the governing equations. Thus, from a control point of view, these independent process variables are the only