Automated Machinery Health Monitoring Using Stress Wave Analysis ArtificialIntelligence

摘要

This paper describes the current state of development of a prototype mechanicaldiagnostic system being developed for the U.S. Army, for application to helicopter drive train components. The system will detect structure borne, high frequency acoustic data, and process it with feature extraction and polynomial network artificial intelligence software. Data for network training and evaluation has been acquired from both healthy and discrepant components, operated over a full range of loads, in a test cell. Stress Wave Analysis is a high frequency acoustic sensing and signal conditioning technology, which provides an analog signal that is a time history of friction and shock events in a machine. This Stress Wave Pulse Train (SWPT) is independent of background levels of vibration and audible noise. The SWPT is digitized and used to compute a set features that characterize the friction signature. Fault Detection Networks of polynomial equations are used to automatically classify SWPT features as being representative of either healthy or discrepant mechanical components. The application of these techniques for automatic classification of friction signatures advances current technology to achieve real time diagnostic capability at all flight power levels.