Machinery Diagnostic Feature Extraction and Fusion Techniques Using Diverse Sources

Abstract : In order to optimize helicopter operational readiness a Joint Advanced Health and Usage Management System (JAHUMS) for helicopter must be highly reliable, minimize false alarms, and prevent catastrophic failures, while operating in real time. To achieve these goals, a fusion of features extracted from non-commensurate factors such as, vibration with oil temperature, oil pressure, and wear debris signatures was implemented via statistical fusion techniques. This feature fusion of non-commensurate factors provides improved diagnoses capability and reduces false alarms. For example, there may be instances where one analysis factor indicates a fault while another has a contra indication. Clearly, fusion of non- commensurate features will be a very effective way to overcome these conflicts, thereby providing better diagnosis performance and improved flight safety of helicopters. Another advantage of this feature fusion is significant data compression through dimensionless statistical discriminators, which is indispensable to efficient storage utilization and on-line real-time analysis. Therefore, data fusion of non-commensurate sources provides efficient machinery diagnosis and prognosis for both the military and commercial field.