Long-term stability of normal condition data for novelty detection

Abstract

As a technique of diagnosing failure in structures and systems, the method of novelty detection shows considerable merit. The basis of the approach is simple: given measured data from normal condition of the structure, the diagnostic system builds an internal representation of the system normal condition in such a way that subsequent departures from this condition can be identified with confidence in a robust manner. The success or failure of the method is contingent on the accuracy of the description of normal condition. In many cases, the normal condition data may have quite a complex structure: for example, an aircraft may experience a wide range of ambient temperatures in the course of a single flight. Also, the operational loads experienced by the craft as a result of flight manoeuvres may have wide-ranging effects on the measured states. The object of the current paper is to explore the normal condition space for a simple benchmark monitoring system. The said system uses Lamb-wave inspection to diagnose damage in a composite plate. Both short-term and long-term experiments are carried out in order to examine the variations in normal condition as a result of run-in of the instrumentation and variations in ambient temperature. The exercise is not purely academic as the fiber-optic monitoring system is a serious candidate for a practical diagnostic system.