Learning Machines for Structural Damage Detection

Abstract

The complexity of civil infrastructure systems and the need to keep essential systems operating after unexpected events such as earthquakes demands the development of reliable health monitoring techniques to detect the existence, location, and severity of any damage in real time. This chapter presents an overview of structural health monitoring techniques. Furthermore, it describes a methodology for damage detection based on extracting the dynamic features of structural systems from measured vibration time histories by using independent component analysis. Based on this analysis, statistical learning theory is used to determine and classify damage. In particular, an illustrative example is presented within which artificial neural networks (ANNs) and support vector machines (SVMs) are compared. ANNs and SVMs are of important value in engineering classification problems. They are two applications of the principles of statistical learning theory, which provides a great variety of pattern recognition tools. The results show that data reduction from acceleration time histories using independent component analysis (ICA), followed by an implementation of learning machines applied to pattern recognition provide a grounded model for structural health monitoring.