Sujit V. Gaikwad

Control-Relevant Input Signal Design for Multivariable System Identification: Application to High-Purity Distillation

Abstract Input signal design issues associated with a discrete-time MIMO control-relevant identification methodology are the focus of this paper. Using a priori information such as the open-loop dominant time constants and desired closed loop speed-of-response, guidelines are presented for the design variables in two periodic, deterministic inputs: the PRBS and Schroeder-phased signals. The guidelines are illustrated on the Weischedel-McAvoy high purity distillation column, which represents an ill-conditioned, highly interactive system. The case study clearly demonstrates that the sensible use of open-loop experimental design and nonparametric estimation, followed by control-relevant parameter estimation, naturally results in a low-order model description capturing the directionality information important for control.

Systematic techniques for determining modelling requirements for SISO and MIMO feedback control

Abstract This paper presents a fundamental methodology for assessing modelling requirements of SISO and MIMO linear control problems. The main result is the formulation of a control-relevant parameter estimation problem (CRPEP), which suitably captures the interplay that occurs between controller sophistication, speed and shape of the closed-loop response, and set-point/disturbance directions affecting the closed-loop system. The CRPEP is used to explain the apparent dilemma between emphasis on low-frequency, steady-state behaviour versus high-frequency, initial time behaviour in modelling for SISO feedback control. For multivariable systems, solutions to the CRPEP are presented using prefiltered estimation of MIMO ARX models (model reduction case) and a state-space frequency-weighted estimation method (system identification case). The superior performance of reduced order and model predictive controllers obtained from control-relevant models is demonstrated on a subset of the Shell heavy oil fractionator problem.

Digital PID controller design using ARX estimation

This paper describes a comprehensive methodology to obtain reduced-order models that satisfy the Prett-Garcia digital PID tuning rules, using prefiltered AutoRegressive with eXternal input (ARX) estimation as a basis. The Prett-Garcia tuning rules possess the advantage that they systematically relate all the controller parameters to the plant model and a low-pass filter with a single adjustable parameter, which directly influences the closed-loop speed-of-response. Furthermore, these rules avoid the problems of intersample rippling, excessive overshoot and undershoot that are a consequence of sampling. The essential aspect of the estimation method is the selection of the prefilter, which allows the reduced model to retain those plant characteristics that have the most effect on closed-loop system behavior. The design of the prefilter is performed systematically using the engineers desired control requirements and the setpoint/disturbance characteristics of the problem. The benefits of this method are shown for a variety of simulated plants, which include a fourth-order system, plants with time delay and integrators, and a plant with zeros outside the unit circle.

Adaptive/self-tuning PID control by frequency loop-shaping

Addresses issues arising in the online adaptation of PID controller parameters. With frequency loop-shaping principles as the underlying controller design approach, the PID parameter adaptation can be performed directly by minimizing a suitable estimation error with standard least-squares algorithms. A variant of such an algorithm is also proposed in an effort to approximate the minimization of the H-infinity norm of the error operator. A numerical example is used to illustrate the implementation of the algorithm.

Systematic Techniques for Determining Modeling Requirements for SISO and MIMO Feedback Control Problems

Abstract This paper presents a methodology by which the issue of modeling requirements for SISO and MIMO control of linear time-invariant plants is analyzed in a fundamental manner. The main result is the formulation of a Control-Relevant Parameter Estimation Problem (CRPEP) which explictly incorporates control considerations such as set point /disturbance directions and desired closed-loop transfer functions. The CRPEP is solved via prefiltered estimation of MIMO ARX models (model reduction case) and a state-space frequency-weighted estimation method (system identification case). Reduced-order and model predictive controllers obtained from these estimation techniques are demonstrated on a subset of the Shell Heavy Oil Fractionator problem.

Digital PID controller tuning using prefiltered ARX estimation

The use of prefiltered ARX (autoregressive with exogenous input) estimation to obtain reduced-order models that satisfy the Prett-Garcia (1988) digital PID (proportional plus integral plus derivative) tuning rules is described. The design of the prefilter is performed systematically using the engineers desired control requirements and the setpoint/disturbance characteristics of the problem. The benefits of this method are shown for a fourth-order system.<<ETX>>

CONTROL-ID: a demonstration prototype for control-relevant identification

We present CONTROL-ID, a SIMULINK-based package that accomplishes control-relevant identification geared for the needs of the chemical process industries. The major components of the package are described and illustrated using a delayed plant model subject to a substantial nonstationary disturbance.

Modeling for Control Design in Combined Feedback/Feedforward Control

This paper presents control-relevant parameter estimation as a means to address the problem of modeling requirements for a combined feedback-feedforward control system. The key element in the estimation procedure is prefiltering of the input and output time series obtained from the plant. This insures that the estimated model retains those plant characteristics that are most significant with regards to the users control requirements. An example model reduction problem is presented which uses the IMC design procedure to design a reduced-order feedback-feedforward control system. Its performance is compared to both full-order IMC and model-predictive control via DMC.

CONTROL-ID: an integrated framework for system identification and process control

CONTROL-ID is a computer aided control engineering (CACE) tool serving as a support environment for computer aided control system design (CACSD) in the chemical process industry. The fundamental basis for this tool is the theory of control-relevant system identification, which takes advantage of the interplay between identification and control design. CONTROL-ID is implemented using MATLAB on a VAXStation 4000 cluster, which is integrated in real-time to an industrial-scale Honeywell TDC 3000 plant information and control system. Control action is computed on the TDC 3000 system using low-order difference equations, which yield superior performance over traditional PID control while resembling the behavior of model predictive control systems. Results from a simulation study using a gas/oil furnace are reported.<<ETX>>