Automated Characterization of Structural Disbonds by Statistical Examination of Bond-line Strain Distribution

Abstract

This article outlines a statistical damage detection engine for the automated characterization of disbonds in bonded structural joints. Unlike other complex pattern recognition algorithms, the statistical technique described in this article is model-independent. This avoids the need to generate accurate numerical models of the structure which is time and resource intensive. Instead, the training data for the damage detection engine is derived from actual measurements of the virgin (undamaged) structure using embedded sensors. The sensor measurements are normalized to remove the effect of varying loads acting on the structure and desensitized to reduce the probability of false alarm before being used to train the statistical engine. Damage detection is based on box plot outlier analysis, and damage location is determined by knowing the absolute spatial positions of the embedded sensors. This technique has been applied to both simulated and experimental data for joints made of glass fiber reinforced polymer composite, and has been shown to be effective in the automated characterization of disbonds with an acceptable level of accuracy.