Novel support vector regression for structural system identification

Abstract

Structural parameter identification using vibration data is a challenging topic, because of the noise in I/O measurement, incomplete measurement, large DOF of structures and ill-condition nature of inverse analysis. A novel structural identification method is proposed, by using the support vector regression (SVR) technique, which is a promising machine learning technology. Due to the 'Max-Margin' idea of the SVR, the suggested method produces accurate and robust results, even when vibration data are polluted by high-level and non-Gaussian noise. Moreover, unlike common machine learning technologies applied in the structural health monitoring, the SVR is utilized to make structural identification by means of the auto-regressive moving average (ARMA) or auto-regressive (AR) model derived from governing equations of motion, thus it is able to identify structural parameters directly. Numerical structural identification examples are provided to verify the efficiency of the proposed approach. Copyright © 2006 John Wiley & Sons, Ltd.