System identification via orthogonal arrays sampled genetic algorithms

Abstract

System identification methods as an important branch of structural health monitoring (SHM) have been extensively applied in model updating and structural damage detection. Based on the measurements of input and output, the identification process is translated into optimization problems. The structural parameters are identified by minimizing output-formulated objective function. In this procedure, genetic algorithms (GAs) have proved to be robust in finding the approximate global optima. However, due to its random search nature, original GA often requires considerable computer time to achieve an acceptable result. The performance will be further deteriorated with an increasing number of variables to be identified. In this study a search space reduction method by orthogonal arrays sampling (SSRMOA) is presented. Using small samples to explore preliminarily in hyper plane of the solution space, the search limit in the present iteration is predefined. To balance the tradeoff between the space reduction and exploration, a jump-back procedure is proposed to guarantee a much safer search range in the successive runs. Numerical examples show that the proposed sampling-based search space reduction method significantly enhances the GA performance in identifying the unknown structural systems. © 2009 Society for Experimental Mechanics Inc.