Structural Health Monitoring Using Statistical Pattern Recognition Techniques

TLDR

Structural health monitoring is framed as a statistical pattern recognition problem. The study aims to extract features and build a statistical model that differentiates signals from two structural conditions of a patrol boat. Two time‑series based pattern recognition methods—an AR/ARX two‑stage prediction model and a Mahalanobis‑distance outlier analysis—were applied to fiber‑optic strain‑gauge data from the boat to extract discriminative features. Both techniques successfully distinguished the data sets corresponding to the different structural conditions.

Abstract

This paper casts structural health monitoring in the context of a statistical pattern recognition paradigm. Two pattern recognition techniques based on time series analysis are applied to fiber optic strain gauge data obtained from two different structural conditions of a surface-effect fast patrol boat. The first technique is based on a two-stage time series analysis combining Auto-Regressive (AR) and Auto-Regressive with eXogenous inputs (ARX) prediction models. The second technique employs an outlier analysis with the Mahalanobis distance measure. The main objective is to extract features and construct a statistical model that distinguishes the signals recorded under the different structural conditions of the boat. These two techniques were successfully applied to the patrol boat data clearly distinguishing data sets obtained from different structural conditions.