Condition-Based Machinery Maintenance

Caution of faults in mechanical systems such as the gear boxes and bearings onboard helicopters and ships. Current problem solutions use relatively simple metrics to characterize changes in recorded data. The approaches are model-based and of limited utility. In many cases, the interaction of fault conditions with the mechanical system is time-varying and highly nonlinear. Specifying a good analytic model is at best difficult and most likely impossible. An alternative solution is to use neural nets coupled with appropriate feature extractors. Neural nets have been used extensively and successfully at ORINCON to solve detection and classification problems in underwater acoustics. Neural networks assume no underlying model for the events of interest. Rather, the neural netork "learns" to detect and classify faults by examination of features from training data which have known fault conditions. In Phase I, ORINCON will develop a prototype system that uses three feature extractors coupled with a multilayer perceptron neural net to solve the problem. The three features selected are well suited for characterization of the expected fault classes. We will demonstrate the prototype system processing real data supplied by Westland Helicopters.