On the use of the Mahalanobis squared-distance to filter out environmental effects in structural health monitoring

Abstract

This paper discusses the possibility of using the Mahalanobis squared-distance to perform robust novelty detection in the presence of important variability in a multivariate feature vector. The application of interest is vibration-based structural health monitoring with a focus on data-based damage detection. For this application, the Mahalanobis distance can be used to detect novelty using a multivariate feature vector extracted from vibration measurements from a structure at regular intervals during its lifetime. One of the major problems is that changing environmental conditions induce large variability in the feature vector under normal condition, which usually prevents detection of smaller variations due to damage. In this paper, it is shown that including the variability due to the environment in the training data used to define the Mahalanobis distance results in very efficient filtering of the environmental effects while keeping the sensitivity to structural changes.