Guanzhong Dai

Fault Detection and Identification of Dynamic Systems using Multiple Feedforward Neural Networks

Abstract Combining feedforward neural network (FNN) and multiple model adaptive estimator (MMAE), a new approach for fault detection and identification (FDI) of nonlinear systems as well as linear systems is proposed in this paper. Instead of Kalman filters, a bank of FNNs is used in the MMAE which are trained for the normal operation and possible fault situations. In order to overcome the drawbacks of the traditional BP training algorithm for FNN, singular value decomposition is used for the selection of hidden neurons, and then a new fast learning algorithm for training FNN by using a variable time-varying forgetting factor technique and U-D factorization baaed extended Kalman filter (EKF) is proposed. The new approach is then used for FDI of nonlinear systems as well as linear systems. The effectiveness of the method proposed is demonstrated by two simulation examples.

Expert System Approach of Industrial Process Adaptive Control

This paper develops intelligent control from the view of integrating artificial intelligence with control theory, especially expert system with adaptive control. The general structure of intelligent control based on the principle of expert system is presented and is called Expert Intelligent Control (EIC). Expert intelligent control is the harmonious combination of dynamic inference, feature recognition, self-learning and control theory under knowledge base environment. In this paper, we also discuss the application of the real-time expert intelligent control for the tension control of an optical fibre drawing process. The EIC system is implemented in a microcomputer. The good operating results of the closed-loop control system show that our EIC scheme satisfies the requirements of the tension control system perfectly.

Intelligent Multisensor State Information Fusion with a Stochastic, Fuzzy, Neural Network

Abstract The intelligent multisensor state information fusion based on a new stochastic, fuzzy, neural network is investigated. This network carries out the parameter and structure learning to obtain the optimal fuzzy membership functions and the optimal number of fuzzy rules of stochastic dynamic systems. The state information fusion with radar and infrared sensors based on the network is simulated in an uncertain environment. The numerical results show that the proposed intelligent method is effective and superior to fuzzy neural network based method

Performance analysis of interacting multiple model algorithm

Abstract A new non-simulation method for performance analysis of Interacting Multiple Model Algorithm is proposed. Firstly, how input-interaction effects model-conditional estimation is analyzed. Four conclusions are made qualitatively. Besides this, the compression ratio of model-conditional error is defined. So the parameters and modeling can be chosen quantitatively. Secondly, how input-interaction effects model probability is analyzed. Input-interaction is found not only to decide the upper and lower limits of model probability but also to lessen the difference among model probabilities, then in the sense of model probability, decaying-memory filtering, damping coefficient and regulating-time are defined. All these works may be useful to choose optimal parameters and design new adaptive filters.

Decentralized Stabilization for Nonlinear Similar Composite Systems with Uncertainty

Abstract A kind of discontinuous decentralized robust control scheme for a class of nonlinear large-scale composite systems with similar subsystems is proposed by using the Lyapunov Max-Min methods. Both matched and unmatched uncertainties are considered, and the interconnections with more general bounding functions are dealt with. By exploiting the similar structure, a discontinuous nonlinear controller is designed to stabilize the composite system. Finally, a numerical example is given to demonstrate the effectiveness of our result.

Development of an Expert System for Computer-aided Control Systems Design

Abstract We propose a rule-based expert system for computer-aided control systems design (CACSD). The main goals of this expert system are to reduce the users’ knowledge needed for operating CACSD package as much as possible, and help users to obtain satisfying design results. The expert system acts as an intelligent front end between the users and CACSD package. The expert system embodies knowledge not only about all the detailed syntax rules required for forming a CACSD running program, but also about particular problem-solving, command interpretation, control theory and alogrithms. The expert system decides what kinds of input are required for a problem-solving, and then prompts the users in a friendly fashion for any additional information. The expert system is a rule-based system, using data list structure to model the approaches and the attributes (i. e., problem attributes, approach attributes, and so on). The concepts, the thoughts and the new ideas we have described in this paper may be of great value in the development of the third-generation of CACSD package.

An Optical Piber Drawing Self-tuning Control System

Abstract This paper discusses the design and implementation of an optical fiber drawing process self-tuning control system. According to the design objective and requirements of the system, a self-tuning control algorithm is presented based on the optical fiber drawing process dynamic properties. The self-tuning control algorithm has been implemented on the dual-microcomputer systea composed by IBM-PC and DJS-041 single board computer, and the successful application has been achieved. The diameter of optical fiber can be controlled wi thin 125 ± 1 µm, coming up to advanced standard.

New Recursive Smoothing Algorithms for Bernoulli-gaussian Input Sequence

Abstract It is well known that the computational requirement of the traditional optimal smoothing algorithms for Bernoulli-Gaussian input sequence is directly proportional to 2k, where k is the steps of the computation, In this paper we develop some smoothing algorithms for Bernoulli-Gaussian input sequence whose computational requirement are not as such as 2k. First, we view the process of the estimation as a tree structure with its trajectories represent the progress of the Bernoulli sequence. Then we define a function S[*] and give the way to compute it recursively. With the determination of S[*], we develop a procedure to find the trajectories which are not the subsequence of the optimal trajectory. So these trajectories can be pruned, We also develop another algorithm which reduces the computational load further. These two algorithms are simulated through an example, The results show that both algorithms give good results and considerably reduce the computational load.