DAMAS COS Damage Location Demonstrator for Structural Health Monitoring

摘要

The problem of damage detection and identification has a natural hierarchical structure. At the higher levels, one might require the diagnostic to return say, information about the expected time to failure of a structure, while at the lowest level, the question is simply of whether a fault is present or not. In many ways, the latter is the most fundamental. In response to the need for robust low-level damage detection strategies, the discipline of novelty detection has recently evolved (Bishop, 1994) (Tarassenko et al., 1995). The problem is simply to identify from measured data if a machine or structure has deviated from normal condition, i.e., if the data are novel. The method requires a bank of normal condition data against which the possible damage condition data are compared. Recent work (Worden et al., 2000) has shown that it is possible to extend these ideas to allow damage to be located as well as detected. The idea uses the output from a network of novelty detectors to train a neural network classifier to be able to indicate the area of damage in a structure. This paper describes a demonstrator system that first constructs a network of novelty detectors, which are trained to be sensitive to change in the Lamb wave excitations in a composite panel. The output from these novelty detectors are then used to train an artificial neural network to recognise which area of the composite panel is being subjected to damage. The layout of the paper is as follows: Section 2 gives the main objectives of the DAMASCOS consortium and the partners involved whilst Section 3 describes the Lamb wave data. Section 4 discusses the demonstrator hardware and software whilst Section 5 rounds off with a discussion and some conclusions.