Elsevier

Applied Soft Computing

Volume 8, Issue 1, January 2008, Pages 261-273
Applied Soft Computing

Modelling plant control strategies and their applications into a knowledge-based system

https://doi.org/10.1016/j.asoc.2007.01.008Get rights and content

Abstract

The high complexity of a plant control system related structuring the domain expert knowledge into a knowledge base could decrease task. This paper presents a strategy adopted to model the application of control strategies employed in surveyed companies. Control engineering plays an important part in any industrial plant. Good control and optimisation of correct control strategies is therefore very crucial for the effective running of a control task related establishment of any kind. The control strategy approaches, must be followed right from system identification, modelling, validation, test and implementation. These stages are not always transparent when dealing with a system whose mathematical model does not exist or is difficult to obtain. When faced with this kind of problem, control enhancement techniques, such as knowledge-based, and intelligent system are always an obvious alternative.

Introduction

This paper presents findings of a study on the availability and application of modern control techniques on industrial plants. For many industry automation projects, speed, reliability repeatability together with simple structure and operation, easy maintenance and low component cost are the most important characteristics of a plant system. System modelling plays an important role in the characterisation of any plant behaviour. It is one of the main parts of the plant system research work. For control of plant systems, no matter which approaches are going to be employed, a sufficient insight into and knowledge of the behaviours of the system will definitely facilitate the control design. Many research efforts have been made to come up with competitive control plant. These efforts include but not limited to the following:

  • Seeking for advanced component structure and low component imperfection.

  • Seeking for sophisticated but still practical control techniques.

  • Seeking for better modelling and understanding of overall plant systems.

With the progress in research and development, cost effective plant control systems have been used in robot manipulators, truck air braking, truck clutch control, and packaging and rehabilitation machines. The motivation of this research is the need to know the types of industrial control plant used in the developing world, especially Botswana. System identification and modelling play an important role in the development of a plant model. For control of industrial plants, no matter which approaches are employed, sufficient insight and knowledge of the behaviour of the system facilitate good control system design. The identification and modelling of the dynamics any real plant is a non-linear problem. Some critics argue that it is not even an engineering problem. However, with the application of heuristic control strategies, especially knowledge-based and expert systems, solutions to such problems can be realised. The trend analysis procedure employs a number of moving average views (weighted or non-weighted) that give the user flexibility in customising the analysis according to his/her expertise. Non-expert users can have default views generated automatically by the knowledge base [1]. The mathematical model of such plant is difficult to get or may not exist, hence the use of intelligent methods is unavoidable. The best approach to such a problem is the black box system identification method, using input/output data [2]. The data has been collected by means of a questionnaire and oral interviews of engineers in the field. Survey questions form the input to the system and answers from these engineers and technicians will form the output to the system. This input/output data will then be fed into the expert system and the output from the expert system will be the best control strategy for the plant. Like any other control problem, there are bound to be disturbances in the plant. These are also highly non-linear, such as political decisions, environmental issues, conflicts of interest, corruption and unavoidable inherent socio-economic problems from neighbouring countries. Nowadays, fuzzy systems and fuzzy neural networks are being extensively used in data analysis. Fuzzy systems and fuzzy neural networks are based on the extension of classical rule-base systems, because they deal with linguistic rules instead of classical logic rules. Furthermore, fuzzy neural networks concern the integration of fuzzy systems and neural networks and they are often used to construct corresponding fuzzy systems in fuzzy modelling [3]. To understand this research problem well, one has to bear in mind that, plants in industry are controlled via actuators such as electrical, hydraulic and pneumatic actuators. Solution of problems associated with these actuators would then warrant efficient plant control [4].

To put this research work into perspective, the existing information retrieval methods are discussed for critical review and comparison purposes. The reviewed approaches to which the proposed cybernetic balance search (CBS) technique conforms to the human affect recognition, textual affect modelling and online virtual training environment (VTE). Formal methods to incorporate such information in the development of models and controllers have been developed. Among these methods, techniques based on fuzzy sets and fuzzy logic represents a promising approach which has been developed considerably in recent years [5], [6], [7]. The rule-base nature of fuzzy models allows the use of information expressed in the form of natural language statements and makes the models transparent to interpretation and analysis. At the same time, at the computational level, fuzzy model can be regarded as flexible mathematical structure, similar to neural networks that can approximate large class of non-linear system to a desired degree of accuracy [8], [9]. Fuzzy rule-base system can be used as knowledge-based models constructed by using knowledge of experts in the given field of interest [10], [11]. Many real world systems are inherently non-linear and cannot be represented by linear models used in conventional system identification [12]. Artificial neural networks and fuzzy models belong to the most popular model structures used. From the input–output view, fuzzy systems are flexible mathematical functions which can approximates other functions or just data, measurements with a desired accuracy. This property is called general function approximation. Different methods have been developed using fuzzy set theory to model systems such as rule-base fuzzy system [13] fuzzy linear regression methods [14] fuzzy based on cell structures [15]. Two main approaches to the integration of knowledge and data in a fuzzy model can be distinguished as:

  • 1.

    The expert knowledge expressed in a verbal form translated into a collection of if–then rules. In this way, a certain model structure is created. The particular tuning algorithm may exploit the fact that at the computational level, a fuzzy model can be seen as a layered structure (network), similar to artificial neural networks, to which standard learning algorithms can be applied. This approach is usually termed neuro-fuzzy modelling [16].

  • 2.

    No prior knowledge about the system under study is initially used to formulate the rules, and fuzzy model is constructed from data. It is expected that the extracted rules and membership functions can provide an a posteriori interpretation of the system's behaviour. An expert can confront this information with his own knowledge, can modify the rules or supply new ones and can design additional experiments in order to obtain more informative data. This approach is termed rule extraction. Fuzzy clustering is one of the techniques that are often applied [17], [18], [19]. In order to have both the capabilities of leaning and interpretability in a single system, hybridization of neural network and fuzzy systems often called neuro-fuzzy systems is a powerful designing approach [20].

In order for an adaptive information retrieval system to respond to human affect, the system must first be able to sense and recognize users’ affective state [21]. The sensing medium can be broken into three categories of textual, audio/visual and physiological. In [22], it has been demonstrated that affect signals maybe derived in many ways.

This modelling approach analyses text to estimate the effective state of the user. The model approach ranges from the simple keyword spotting to more sophisticated techniques such as assigning probabilistic affinity of these key words and statistical natural language processing approaches.

The potential of VTE technology for supporting education is widely recognized. This use centered education delivery approach has been implemented in [23] for the aid of people with disabilities. The approach demonstrated superior benefits compared with the two mentioned approaches because it:

  • Limits the cost constraints in ‘learn-by-doing’ institutions.

  • Provides a unique vantage point for learning by placing the user in a simulated world.

  • Provides different framework to scaffold education.

  • Provides quantitative and qualitative information from which to distil the tutorial strategies necessary for developing intelligent agents (IAs).

This paper is organised as follows: Section 2 describes the conduct of the survey, system identification and plant modelling procedure. Section 3 describes the application of the neuro-fuzzy model strategy. Section 4 gives the results of the survey and implementation of survey result into a knowledge base and an expert plant model. Section 5 gives a discussion of feed back and conclusion is drawn in Section 6.

Section snippets

Intelligent based system identification and modelling

In fuzzy modelling, the identification method used is very important. Identification for fuzzy modelling has two aspects: structure identification and parameter identification. In general, structure identification is a difficult and extremely ill defined process and not readily amenable to automated techniques. The problem of parameter identification is closely related to the estimation of the membership functions of the fuzzy sets or alternatively, the fuzzy relation associated with the fuzzy

Design of a neuro-fuzzy controller

To analyse our system, the study is based on the Telecom Literature Question Answering System (TeLQAS) [32]. TeLQAS is an ontology-based domain specific QA system which has been developed for the field of Telecommunications. It is composed of two major processes, namely, online and off-line. Both processes use a domain ontology accompanying with a local text warehouse. The ontology accommodates knowledge of the system. It is created initially by experts in the domain of telecommunications. It

Results and implementation

In this section experimental results demonstrating the performance of the developed control strategies as applied to an information flow system are presented. To demonstrate the characteristic behaviour PD- and PI-type FLCs, these were individually tested with control education survey data model. It has been noted in Fig. 10 that the linear fuzzy system response with PD-type (predicted output) controller has larger rise time, relatively large overshoot but both have equal settling time and

Discussion

The crisp analysis showed a slight influence of the percentage of plant use on the experience and employees awareness. The rule-base itself uses a simple technique. It starts with a rule-base which contains all of the appropriate knowledge encoded into if–then rules and a working memory, which may not initially contain any data, assertions or initially known information. Which rule is chosen to fire is a function of the conflict resolution strategy. Which strategy is chosen can be determined by

Conclusion

In this paper, it is described a strategy of applying an adaptive neuro-fuzzy based controller to make the control parameter to follow the desired pattern. Simulation results show how well the neuro-fuzzy based control method performs. The effectiveness of this strategy can be validated using standard model validity and correlation tests. It has been shown that neural-fuzzy networks can be used to design a controller for typical non-linear plants like industrial behaviour. The difficulty in

Acknowledgements

The results of this work are obtained within the project “The potential use of modern control techniques in plant in Botswana”. The authors thank the University of Botswana, office of research and development for financial support of this work and more gratitude goes to Dr M.O. Tokhi at the University of Sheffield, England, UK who spent his precious time helping with the pre-review of the manuscript before initial submission.

References (34)

  • C.T. Lin

    Neural Fuzzy Control Systems with Structure and Parameter Learning

    (1994)
  • L.-X. Wang

    Fuzzy systems are universal approximators

  • B. Kosko

    Fuzzy systems are universal approximators

    IEEE Trans. Comput.

    (1994)
  • X.J. Zeng et al.

    Approximation theory of fuzzy systems—MIMO case

    IEEE Trans. Fuzzy Syst.

    (1995)
  • B.G. Buchanan et al.

    Rule-based Expert Systems

    (1984)
  • D.W. Pattertson

    Introduction to Artificial Intelligence and Expert Systems

    (1990)
  • L. Ljung

    System Identification Theory for the User

    (1987)
  • Cited by (4)

    • Predicting users’ preferences by Fuzzy Rough Set Quarter-Sphere Support Vector Machine

      2021, Applied Soft Computing
      Citation Excerpt :

      A recommender system involves the prediction of a set of items that will be users’ interests and will achieve high ratings by users. In the literature, a variety of strategies are presented for making recommendations, e.g., the content-based methods [1,2], knowledge-based methods [3,4], collaborative filtering methods [5,6], demographic methods [7,8], and hybrid methods [9–11]. Content-based methods use item features to recommend other items similar to what the user likes, regarding the user’s previous ratings.

    • Asphaltene deposition prediction using adaptive neuro-fuzzy models based on laboratory measurements

      2013, Fluid Phase Equilibria
      Citation Excerpt :

      Learning of a neuro-fuzzy system incorporates the two basic steps of structure and parameter identifications [25,26]. The number of rules is treated as the measure of model structural complexity [27,28] to be properly tuned on the basis of available data as clearly been addressed by the self-structuring scheme adopted in online identification and adaptive control applications based on fuzzy rule-base (FRB) models [29,30]. In this research, an adaptive neuro-fuzzy algorithm is introduced to recursively realize a compact number of fuzzy rules.

    • Online affine model identification of nonlinear processes using a new adaptive neuro-fuzzy approach

      2012, Applied Mathematical Modelling
      Citation Excerpt :

      Thus, model structure identification reduces to realization of input-output relations in the form of if-then rules [7] for which automated methods can be used to add or remove rules [12]. In a neuro-fuzzy modeling, once available a satisfactory structure, the parametric identification searches for the optimal adjustment of the parameters relating to membership functions in the antecedent part and the associated constituent in the consequent part of a fuzzy rule [2,3,7]. Juang and Lin [13] presents a self-constructing neural fuzzy inference network that recursively extracts fuzzy rules from the training data based on the distance criterion between the new input data and the center of the Gaussian membership function in the existing fuzzy rules.

    View full text