An outlier analysis framework for impedance-based structural health monitoring

Abstract

This paper presents the use of statistically rigorous algorithms combined with active-sensing impedance methods for damage identification in engineering systems. In particular, we propose to use statistical pattern recognition methods to address damage classification and data mining issues associated with the examination of large numbers of impedance signals for health monitoring applications. In this paper, in order to diagnose damage with statistical confidence, the impedance-based monitoring is cast in the context of an outlier detection framework. A modified auto-regressive model with exogenous inputs (ARX) in the frequency domain is developed and the damage-sensitive feature is obtained by quantifying the differences between the measured impedance and the output of the ARX model. Furthermore, because of the non-Gaussian nature of the feature distribution tails, extreme value statistics is employed to develop a robust damage classifier. This paper concludes with a numerical example on a 5 degree-of-freedom system and an experimental investigation on a multi-story building model to demonstrate the performance of the proposed concept.