THRESHOLD MODELS FOR NONLINEAR TIME SERIES ANALYSIS.

The discrete representation of the nonlinear random vibratory system using experimental data is a problem of considerable importance in many engineering fields. A new modeling technique is presented to meet this need. The system is identified in terms of a discrete structure, known as the threshold nonlinear autoregressive moving average (TNLARMA) model. The physical properties of the system is characterized by the analysis of the model. A nonlinear van der Pol process is used to illustrate the modeling procedures. It is shown that the TNLARMA approach leads to a satisfactory result and is applicable to many engineering systems.