Identification of Non-Linear Dynamic Systems using an Expert Approach

Abstract

An Expert System approach for the identification of non-linear systems is presented. The Expert System is an attempt to bring order into the non-linear system identification process. The final objective is to deliver a parsimonious mathematical model of the dynamical system under investigation. It works by defining the key stages of the procedure and iterating between them if necessary. It takes advantage of numerous methodologies to accomplish the tasks in each of the stages and uses the results from many of them. The Expert System applies the excitation forces appropriate to each method and analyses the responses. Each stage ends with a set of recommendations that can be used to begin the next stage. The Expert System is applied to a simple non-linear dynamic system. It is shown that the Expert System procedure can automatically detect, locate and quantify the nonlinearity using its array of techniques. Furthermore, it can choose a suitable model structure, select appropriate terms and estimate the model parameters using an extensive set of rules. The resulting identified model is validated and shown to be an accurate representation of the experimental system.