"A Practical Multiple Model Adaptive Strategy for Single-Loop MPC"
Danielle Dougherty and D. J. Cooper
Control Engineering Practice, 11, 141 (2003)

This paper details a multiple model adaptive control strategy for Model Predictive Control (MPC). The internal model of the process employed in MPC is linear, but most chemical processes are nonlinear. Hence, the performance of MPC will degrade as the operating level moves away from the original design level of operation. To maintain performance of the controller over a wide range of operating levels, a multiple model adaptive control strategy for Dynamic Matrix Control (DMC), which is the process industry’s standard for MPC, is presented. The method of approach is to design multiple linear DMC controllers. The tuning parameters for the linear controllers are obtained using novel analytical expressions. The controller output of the adaptive DMC controller is a weighted average of the multiple linear DMC controllers. The capabilities of the multiple model adaptive strategy for DMC are investigated through computer simulations and an experimental system. The work provides an adaptive DMC strategy that is simple to implement and use, requires minimal computation for updating model parameters, relies on the linear control knowledge of plant personnel, and is reliable for a broad class of process applications.

Return to Top of Title List

"Design and Tuning of PID Controllers for Integrating (Non-Self Regulating) Processes"
Robert Rice and D. J. Cooper
Proc. ISA 2002 Annual Meeting, 43, 424 (2002)

This work explores an easy to use and broadly applicable method for tuning PID controllers for integrating processes. Details are presented on the requirements for collecting closed loop dynamic process test data near the design level of operation, the fitting of an integrating dynamic model form to this test data and correlations for computing controller tuning values based on the parameters from the resulting model fit. The method presented is applicable to PID control algorithms in both the interacting and non-interacting derivative forms. The work builds on the work of Chien and Fruehauf [8] and their use of the internal model control (IMC) structure to derive tuning correlations for integrating processes. One novel contribution of this work is the extension of the tuning correlations to include the PID with derivative filter forms. The design and tuning method is demonstrated on process simulations for both set point tracking and disturbance rejection. Results show that the methods described here compare favorably with other more computationally intensive approaches.

Return to Top of Title List

"A Training Simulator for Computer-Aided Process Control Education"
D. J. Cooper and Danielle Dougherty
Chemical Engineering Education, 43, 252 (2000)

A training simulator offers an alluring method for providing students with the significant hands-on practice critical to learning process control. The proper tool can provide virtual experience much the way airplane and power plant simulators do in those fields. It can give students a broad range of focused engineering applications of theory in an efficient, safe and economical fashion. And it can work as an instructional companion as it provides interactive challenges that track along with classroom lectures. Process control is a subject area well suited to exploit the benefits of a training simulator. Modern control installations are computer based, so a video display is the natural window through which the subject is practiced. With color graphic animation and interactive challenges, a training simulator can offer experiences that literally rival those of the real world. These experiences can be obtained risk free and at minimal cost, enabling students to feel comfortable exploring nonstandard solutions at their desk. If properly designed as a pedagogical tool with case studies organized to present incremental challenges, we believe learning can be enormously enhanced for process control with such a training simulator. This paper presents example lessons drawn from the Control Station process control software for education and training.

Return to Top of Title List

"Enhancing Process Control Education with the Control Station Training Simulator"
D. J. Cooper and Danielle Dougherty
Computer Applications in Engineering Education, 7, 203 (1999)

A process control training simulator can enhance learning by integrating the theoretical abstraction of textbooks with the tactile nature of the lab and plant. The primary objective of a training simulator is education. It can motivate, help with visualization, and provide hands-on practice and experience. This paper explores the use and benefits of the Control Station training simulator for process control education. Examples presented illustrate how the standard curriculum can be enhanced with a series of hands-on exercises and study projects.

Return to Top of Title List

"A Tuning Strategy for Unconstrained Multivariable Model Predictive Controllers,"
Rahul Shridhar and D. J. Cooper
Industrial and Engineering Chemistry Research, 37, 4003 (1998)

Move suppression coefficients serve a dual purpose in the Model Predictive Controller (MPC) architecture. These include suppressing aggressive control action and conditioning the system matrix prior to inversion. The work presented here exploits this dual effect in deriving an analytical expression that computes appropriate move suppression coefficients as a function of process model parameters, other MPC design parameters and partitioned block condition numbers of the system matrix. The development is based upon an approximate mosaic Hankel matrix structure of the multivariable system matrix. The primary contribution of this work is the derivation of the analytical expression for computing move suppression coefficients and its demonstration in an overall MPC tuning strategy (Table 1). The examples presented show that the move suppression coefficient remains properly scaled as the other MPC design parameters and process characteristics change to produce a consistent closed loop performance. This tuning method is applicable to multivariable processes, including non-square systems.

Return to Top of Title List

"A Novel Tuning Strategy for Multivariable Model Predictive Control,"
Rahul Shridhar and D. J. Cooper
ISA Transactions, 36, 273 (1998)

Model predictive control (MPC) has established itself as the most popular form of advanced multivariable control in the chemical process industry. However, the benefits of this technology cannot be realized unless the controller can be operated with desirable performance for an extended period of time. The objective of this work is to present an easy-to-use and reliable tuning strategy that enables the control practitioner to maintain multivariable MPC at peak performance with minimal effort. A novel analytical expression that computes the move suppression coefficients, guidelines to select the additional adjustable parameters, and their demonstration in an overall tuning strategy are some of the significant contributions of this work. The compact form for the expression that computes the move suppression coefficients is derived as a function of a first order plus dead time (FOPDT) model approximation of the process dynamics. With tuning parameters computed, MPC is then implemented in the classical fashion using an internal model formulated from step response coefficients of the actual process. Just as a FOPDT model approximation has proved a valuable tool in tuning rules such as Cohen-Coon, ITAE and IAE for PID implementations, the tuning strategy presented here is significant because it offers an analogous approach for multivariable MPC.

Return to Top of Title List

"A Tuning Strategy for Unconstrained SISO Model Predictive Control,"
Rahul Shridhar and D. J. Cooper
Industrial and Engineering Chemistry Research, 36, 729 (1997)

This paper presents an easy-to-use and reliable tuning strategy for unconstrained SISO Dynamic Matrix Control (DMC) and lays a foundation for extension to multivariable systems. The tuning strategy achieves set point tracking with minimal overshoot and modest manipulated input move sizes and is applicable to a broad class of open loop stable processes. The derivation of an analytical expression for the move suppression coefficient and its demonstration in a DMC tuning strategy is one of the significant contributions of this work. The compact form for the analytical expression for the move suppression coefficient is achieved by employing a first order plus dead time (FOPDT) model approximation of the process dynamics. With tuning parameters computed, DMC is then implemented in the classical fashion using a dynamic matrix formulated from step response coefficients of the actual process.

Return to Top of Title List

"Automated Rule-Based Model Parameter Estimation and Controller Design"
Carlos Velazquez-Figueroa and D. J. Cooper
Proc. ISA Tech97 Annual Conf. Anaheim, CA, ISA Publications (1997)

Proper design of automatic process controllers is key to the efficient operation of an industrial process. The design of such controllers follows a specific procedure comprised of three steps. These steps include generation of proper dynamic data, regression of a low order linear dynamic model, and use of the resulting model to complete the controller design. The objective of this work is to provide the practitioner with guidelines for generating dynamic data appropriate for controller design. These guidelines are obtained by studying the impact of design variables on a sum of squared errors (SSE) regression surface. An additional objective is to automate the model parameter estimation and controller design using a rule-based approach. The rule-based system automatically computes initial values of the model parameters to start the estimation procedure. It also incorporates penalty functions that ensure the physical meaning of the model parameters. Finally, it employs the model parameters in appropriate algorithms to estimate controller tuning values.

Return to Top of Title List

"Pattern-Based Closed-Loop Quality Control for the Injection Molding Process"
Suzanne L. B. Woll and D. J. Cooper
Polymer Engineering and Science, 37, 801 (1997)

The basis for a novel pattern-based closed-loop control strategy for the injection molding process is presented. The strategy uses artificial neural networks (ANNs) embedded within a cascade design to analyze sensor patterns, identify process character and control part quality. The platform for this work, the injection molding process, is an industrially significant, cyclic manufacturing operation. Final part quality of this process is a non-linear function of many machine and polymer variables. Part quality control of this process is currently attained via single-input single-output machine controls supervised by human operators. Presented is a method that employs ANN technology to improve upon this approach and provide the basis for closed-loop part quality control. In the cascade design, machine controller set points of an inner loop are updated based on ANN analysis of mold cavity pressure patterns. The controller action maintains the desired pressure pattern set point of the outer loop associated with desired part quality. Control strategy details are provided along with set point tracking demonstrations that support feasibility of this pattern-based approach.

Return to Top of Title List

"A Dynamic Injection Molding Process Model for Simulating Mold Cavity Pressure Patterns"
Suzanne L. B. Woll and D. J. Cooper
Polymer Plastics Technology and Engineering, 36, 809 (1997)

The quality of injection molded parts is typically controlled in the plant using statistical techniques that involve measuring parts as well as monitoring processing parameters. Part quality is also controlled by machine operators who adjust processing conditions in response to trends in process behavior. To achieve direct on-line monitoring and automatic control of part quality, a multivariable, nonlinear process model must be developed that relates process behavior to machine controllable parameters. Presented in this work are the details of such a model derived from first principles and proven correlations. Since recent work has shown that complete mold cavity pressure patterns are good indicators of part quality, the focus of this lumped parameter model is to simulate mold cavity pressure patterns observed during the filling, packing and cooling stages of the process given machine set points for barrel temperature and holding pressure. The model is validated experimentally using a production injection molding machine and parameter sensitivity case studies are presented.

Return to Top of Title List

"On-line Pattern-Based Part Quality Monitoring of the Injection Molding Process"
Suzanne L. B. Woll, B. Souder and D. J. Cooper
Polymer Engineering and Science, 36, 1477 (1996)

The quality of injection molded parts is often monitored in the plant using techniques that focus on the statistical analysis of discrete data and, in particular, peak values. This paper presents an alternative on- line technique for part quality monitoring that focuses on the analysis of complete data patterns. Specifically, this paper discusses the application of artificial neural networks (ANNs) as part quality monitoring tools. The method of approach is to train a back propagation network (BPN) to associate part quality with the corresponding data pattern produced during injection. In Part I of this work, the data pattern consists of a series of discrete values and the part quality measure is defined as part weight. In Part II, the data pattern is the measurement profile observed from a pressure sensor placed in the mold cavity and the part quality measure is defined as part length. Results show that ANNs are successful in predicting part quality based on data patterns when an entire sensor profile is analyzed. Furthermore, demonstrations show that the approach is superior in predicting part quality when compared to statistical techniques now widely practiced by the injection molding process industry.

Return to Top of Title List

"A Unified Excitation and Performance Diagnostic Adaptive Control Framework"
Ralph F. Hinde, Jr., and D. J. Cooper
AIChE Journal, 41, 110 (1995)

Model based controllers contain two elements that must be adjusted to maintain desired performance: parameters of the process model and a tuning parameters in the controller design equation. A unified framework is presented where vector quantizing networks are used in pattern-based methods for diagnosing process excitation and controller performance. Excitation diagnostics analyze sufficiently excited dynamic process data for model updating. Performance diagnostics analyze set point response data and determine appropriate updates to the tuning parameter. Supervisory adaptation logic enables these two adaptive mechanisms to work together to maintain model accuracy and desired controller performance. The method is general to a number of model based control algorithms and process model forms. Demonstrations employ a first order plus dead time model form as well as both PI and DMC algorithms for set point tracking and disturbance rejection in a simulated and a bench scale application.

Return to Top of Title List

"A Neural Network Strategy For Disturbance Pattern Classification and Adaptive Multivariable Control"
Lawrence Megan and D. J. Cooper
Computers and Chemical Engineering, 19, 171 (1995)

This paper presents a neural network approach to adaptive control through pattern recognition techniques, extending previously published results on single-input/single- output systems to two-input/two-output systems. Two vector-quantizing neural networks are used to analyze both the input and output patterns resulting from a perturbation to the process. The results of these analyses are then used to update the model gain of the first order plus dead time model that describes each input/output pair. This work focuses primarily on making model adaptations following load disturbances as opposed to set point changes, as load disturbances present by far the greatest adaptation challenge to chemical process applications. The results are compared to a more traditional modeling technique, batchwise model regression, with respect to both accuracy and computational load. The adaptive strategy is demonstrated using a variety of disturbances on two challenging multivariable process simulations.

Return to Top of Title List

"Pattern Recognition Based Adaptive Control Of Two-Input/Two-Output Systems Using ART2-A Neural Networks"
Lawrence Megan and D. J. Cooper
Industrial and Engineering Chemistry Research, 33, 1510 (1994)

This paper details an applied investigation of pattern recognition based adaptive control for two-input/two-output systems. Two ART2-A neural networks perform a concurrent analysis of controller error and manipulated input patterns resulting from a set point change or an unmeasured disturbance to the system. This information is then used to adapt the models that describe each input/output relationship. The adaptive strategy is demonstrated on two challenging processes: a pilot scale continuous distillation column and a simulation of the Shell Fundamental Control Problem. The distillation column demonstrates the applicability of the adaptive strategy to both set point changes and disturbances in a challenging real-world process, while the Shell problem demonstrates the ability of the strategy to handle irregular disturbance dynamics.

Return to Top of Title List

"A Pattern Based Approach to Excitation Diagnostics for Adaptive Process Control"
Ralph F. Hinde, Jr. and D. J. Cooper
Chemical Engineering Science, 49, 1403 (1994)

To maintain desired controller performance in the presence of process nonlinearity and nonstationarity, linear model based control strategies become dependent upon the regular updating of a process model. This paper explores the use of a passive adaptive algorithm which updates the process model in closed loop by taking advantage of naturally occurring dynamic events rather than by injecting perturbations into the system to create dynamic events. Such closed loop identification is possible, but it requires that these events contain process information that is not masked by measurement noise or unmeasured disturbances. Presented here is pattern-based excitation diagnostic tool (EDT) that determines when sufficient excitation exists for model updating. The EDT consist of vector quantizing neural networks (VQNs) similar to the ART2-A and a decision maker that is a simple set of rules. The VQNs are trained to recognize local dynamic behavior in the recent histories of each process variable. The decision maker uses the outputs from the VQNs to diagnose when sufficient dynamics exist for model updating. Details of the EDT are presented along with several challenging demonstrations on both simulated and real single-input single-output processes.

Return to Top of Title List

"Using Pattern Recognition in Controller Adaptation and Performance Evaluation"
Ralph F. Hinde, Jr. and D. J. Cooper
Proceedings of the 1993 American Control Conference, IEEE Publications, NJ, 74 (1993)

This work presents pattern recognition based methods for controller adaptation and performance evaluation. These methods comprise a passive model-based adaptive control algorithm that is simple to use, easy to understand, stable, and fairly robust in a wide variety of applications. Controller adaptation in this work uses excitation diagnostics to initiate batch-wise regression of a process model to dynamic closed-loop process data. The process model is then employed in model-based controller tuning relations to update the controller's character. Controller performance evaluation is used to determine appropriate adjustments to the tuning relations such that an accurate process model will produce desired controller performance. These adaptive techniques are implemented using vector quantizing neural networks as efficient pattern recognition tools. The adaptive algorithm is presented in a structure that allows for the implementation of these advanced techniques without requiring the replacement of an existing feedback controller. This is demonstrated using a simulated nonlinear third order process and an IMC tuned PI controller with Smith Predictor.

Return to Top of Title List

"Neural Network Based Adaptive Control Via Temporal Pattern Recognition"
Lawrence Megan and D. J. Cooper
Canadian Journal of Chemical Engineering, 70, 1208 (1992)

This paper presents a neural network approach to pattern recognition based adaptive control. Two interconnected back propagation networks are trained to translate error patterns resulting from sustained set point changes into predictions of mismatch between current internal model parameters, model gain and model time constant, and those which restore desired performance. The network predictions are then used to update a model based PI controller. The strategy is demonstrated on two simulations and a pilot scale process which are undergoing severe changes in model gain and time constant. The strategy compares favorably against a more traditional rule based pattern recognition approach.

Return to Top of Title List

"Comparing Two Neural Networks For Pattern Based Adaptive Process Control"
D. J. Cooper, Lawrence Megan and Ralph F. Hinde, Jr.
AIChE Journal, 38, 41 (1992)

An adaptation strategy based on an analysis of the patterns exhibited in the recent history of the controller error and manipulated input variable is presented. The strategy is a two parameter adaptation, where the gain and time constant of the controller's internal model are adjusted to make the closed loop error response match a desired or target pattern. Both a back propagation network and a vector quantizing network (VQN) are compared as pattern analysis tools. This strategy is established for a number of model based controllers and is demonstrated here using the generalized predictive control algorithm. Details of this set point tracking strategy are presented along with demonstrations on both simulated and real single loop processes that experience significant changes in process gain and time constant. Results show both networks to be equally capable at pattern recognition with the VQNs ease of training and implicit ability to assess the accuracy of the pattern match as deciding factors in network selection.

Return to Top of Title List

"Modeling Combustion Efficiency in a Circulating Fluid Bed Liquid Waste Incinerator"
Douglass W. Sevon and D. J. Cooper
Chemical Engineering Science, 46, 2983 (1991)

An incinerator's combustion efficiency (CE) indicates the effectiveness of the incinerator to completely oxidize waste. Circulating fluidized beds (CFB's) show promise as a viable hazardous waste incineration process, but to fulfill this promise, an understanding of the interaction between CE and important operating parameters is still needed. To gain this understanding, an experimental and a theoretical investigation of CFB incineration was performed. A CFB incinerator was constructed to study the destruction of an organic liquid waste. An experimental program on incineration was conducted with propanol as the simulated waste. Operating data on the dependence of CE as a function of major operating parameters such as excess air, average particle size, and primary air/total air ratio for this facility is presented. A theoretical process model specific to CFB incineration of an organic liquid waste is developed, and the numerical implementation of the model is presented. Model constants are fitted with experimental data so the process model specifically describes the CFB pilot plant. Process sensitivity to major operating parameters is investigated in series of simulations. Design configurations slightly different from the experimental CFB such as the column height and preheat temperature of combustion air are also studied with the model.

Return to Top of Title List