Novelty Detection in Airframe Strain Data

The structural health of airframes is often monitored by analysis of the frequency of occurrence matrix (FOOM) produced afer eachjight. Each cell in the matrix records a stress event of a particular severity. These matrices are used to determine how much of the aircrafr’s life has been used up in each jiight. Unfortunately, the sensors that produce this data are subject to degradation themselves, resulting in corruption of FOOMs. This paper reports a method of automating detection of sensor faults. It is the only known method that is capable of detecting such faults. The method is in essence a dimensionality reduction algorithm coupled to a novelty detection algorithm that produce measures of unusual counts of stress events at the level of the individual cell and unusual distributions of counts over the entire FOOM. Cell-level error is detected using a probability threshold and a sum of standard deviations. FOOM-level error is detected using a novel application of the Eigenface algorithm. Novelty is measured using a mixture of Gaussian model of the data, fitted using the ExpectationMaximisation algorithm. Our approach to the detection of sensor fault is to examine the variability of FOOMs from normal flights both at the level of the individual cell and at the level of the whole FOOM. Our strategy is to apply a number of statistical measures to sets of example FOOMs and use these as inputs to a novelty detector. The measures include a Gaussian model of each cell in the FOOM coupled with an Eigenface representation of data developed by Turk and Pentland [5 ] . These measurements provide a compact representation of FOOM data without making any assumptions about the underlying distributions of stress cycle counts. In an earlier paper [3], the authors examined the use of a Multi-Layer perceptron (hILP) for the classification of sensor faults. This scheme required data from faulty sensors to be available for training. Here, we circumvent this problem by using a novelty detection scheme as our flight classifier. The novelty detector algorithm uses the Expectation-Maximization algorithm of Dempster, Laird and Rubin [2 ] to fit a Gaussian basis function model to the parameterised data. The advantage of using a novelty measure is that any unusual flight will be detected, providing changes are evident in the measurements we are using.