Multiple-site Damage Location Using Single-site Training Data

Abstract

The identification of multiple-site damage is a challenging problem in data-based structural health monitoring (SHM). It is generally accepted that higher level damage identification via statistical pattern recognition requires the adoption of a supervised learning approach, with the need for data to be gathered from the structure in all damaged states of interest. The number of states for which data would be required to cover all damage combinations grows exponentially with the number of locations at which damage may occur. Damage state data sets of this extent are unlikely to be available in practical applications. The objective of this paper is to explore an interesting approach to the problem of multiple-site damage location. It is postulated that if sufficient information can be gleaned from single-site damage data to allow identification of multiplesite damage, then the requirement to gather data for all combinations of damage location may be circumvented. In the present study this possibility is assessed using data from an experimental structure. The experimental structure used is a full-scale, laboratory-based aircraft wing section. Damage sensitive features identified using single-site data are shown to perform well when applied to the multiple-site location problem.