H.N. Koivo

Paper: A multivariable self-tuning controller

A multivariable self-tuning controller is derived extending the scalar version of Clarke and Gawthrop (1975). Because the control terms are penalized in the cost function, fluctuations and peaking in control signals are reduced compared with the multivariable minimum variance self-tuning controller (Borisson, 1979). Time-varying reference signals as well as certain nonminimum-phase systems can be handled without difficulty with the proposed controller. Several examples illustrate the power of the derived self-tuning controller.

Neural networks in process fault diagnosis

Fault detection and diagnosis is an important problem in process automation. Both model-based methods and expert systems have been suggested to solve the problem, along with the pattern recognition approach. A number of possible neural network architectures for fault diagnosis are studied. The multilayer perceptron network with a hyperbolic tangent as the nonlinear element seems best suited for the task. As a test case, a realistic heat exchanger-continuous stirred tank reactor system is studied. The system has 14 noisy measurements and 10 faults. The proposed neural network was able to learn the faults in under 3000 training cycles and then to detect and classify the faults correctly. Principal component analysis is used to illustrate the fault diagnosis problem in question. >

Application of artificial networks in process fault diagnosis

Abstract Fault diagnosis has been studied very actively during recent years. Estimation methods, rule-base reasoning and pattern recognition techniques are the most common methods used to solve problems. In recent years artificial neural networks have been used successfully in pattern recognition tasks and their suitability for fault diagnosis problems has also been demonstrated. However, the results presented in the literature usually consider very simple example situations. In this paper a realistic heat exchanger-continuous stirred tank reactor system is studied as a test case. The system with 14 noisy measurements and 10 fault situations is studied. The arrangement of different fault categories is visualized by the principal component analysis. The fault detection and diagnosis is based on the classification of process measurements and the classification is carried out using neural networks.

Brief paper: Multivariable tuning regulators for unknown systems

The problem of determining a multivariable robust PI-controller for an unknown linear multivariable stable plant is discussed. A method for constructing a multivariable P-controller, which uses interactions of the plant, is developed. These interactions are detected by observing the output of the plant subject to step-inputs. The developed P-controller form together with the robust multivariable I-controller (Davison, 1976) a robust multivariable PI-controller, the construction of which is also based on step-responses. The controllers are applied to a laboratory-scaled concentration-flow process.

Brief paper: Tuning of multivariable PI-controllers for unknown systems with input delay

A design method for on-plant tuning of multivariable PI-controllers for unknown systems is presented. The plant is assumed to be linear, open-loop stable with input delays and subject to step disturbances. The multivariable PI-controller may be tuned with the aid of open-loop step-responses. Finally an example is given to show the applicability of the controller proposed.

Inverse Nyquist array technique in the design of a multivariable controller for a solid-fuel boiler

Abstract This paper presents an application of the inverse Nyquist array (INA) technique to the design of a multivariable controller for a boiler subsystem. The boiler is a 1.6 MVV water boiler using solid fuel. The subsystem studied consists of the boiler underpressure and the flue-gas oxygen content. First a simple transfer function model based on step response measurements is constructed. The subsystem model is analysed and the interactions are decreased with a constant compensator using Rosenbroeks INA method. Finally, scalar PI controllers are tuned for the approximately decoupled system. A brief description of the microcomputer control system is given. The control system and the designed multivariable controller have worked well since their installation in February 1983.

Tuning of multivariable discrete-time PI-controller for unknown systems

Tuning of a multivariable discrete-time PI-controller, which uses interactions of the plant is discussed. Interactions are detected by observing the output of the plant subject to step inputs. The plant is assumed to be unknown but linear time-invariant and stable. The presented tuning method is an extension of the corresponding continuous-time method [1], [2].

An expert system for multivariable controller design

Abstract In this paper a software package for the design of decentralized and centralized multivariable controllers is described. The software package includes several numerical methods for multivariable controller design as well as an expert system to assist the user. The user interface of the expert system employs hypertext techniques. A continuous-time transfer function matrix is used as a process model. When the process model is known this software package can be used to select a control structure and to tune controllers. The paper first describes the different subtasks of multivariable controller design. Next a short description of the design and analysis methods included in the software package is given. A description of the expert system and a design example finish the paper.

Fault Diagnosis of Dynamic Systems Using Neural Networks

Abstract A method for detecting faults of nonlinear systems is developed using dynamic nonlinear predictor models which are realized with neural networks. Separate predictor models are identified for the process operating normally and for the situations where one of the potential faults has occurred. In monitoring the state of the process, the best fitting predictor of the bank is selected according to the Bayes rule. Radial basis function networks are used in identifying dynamic predictor models and the parameters of the networks are estimated with the orthogonal least squares algorithm. The performance of the proposed method is demonstrated in the simulation studies of a jacketed reactor.

Tuning of a Multivariable Fuzzy Logic Controller

ABSTRACTAlthough fuzzy logic control has created lots of interest in recent years, systematic tuning methods for fuzzy logic controllers have remained uncovered. The applications presented in the literature are usually tuned by trial-and-error methods.In this article a systematic off-line method for tuning a multivariable fuzzy logic controller for an unknown multivariable system is proposed. The system can include time delays, and it is assumed to be stable and linear. In order to keep the structure of the controller simple, the number of the system inputs and outputs are restricted to two. A model of the system is not needed. The rules of the controller and places of the membership functions for the controller outputs are determined from the static gain of the system which can be measured experimentally. Approximate knowledge of the time constants and the delays of the system can help in the selection of the widths of the membership functions. The controller eliminates the steady-state error and keeps t...