Application of neural networks to predict UH-60L electrical generator condition using (IMD-HUMS) data

Abstract

In 2003, the US Army began using the Integrated Mechanical Diagnostics Health and Usage Management System (IMD-HUMS), an integrated airborne and ground-based system developed by Goodrich Corporation, to support maintenance of the UH-60L. IMD-HUMS is responsible for collecting, processing, analyzing, and storing an enormous amount of vibratory and flight regiime data obtained from sensors located throughout the aircraft. The purpose of this research is to predict failures of the UH-60L's electrical generators, applying Airtificial Neural Networks (ANN) on the IMD-HUMS-produced data. Artificial NNs are data based vice rule based, thereby possessing the potential capability to operate where analytical solutions are inadequate. They are reputed to be robust and highly tolerant of noisy data. Software tools such as Clementine 10.0, S-Plus 7.0, and Excel are used to establish these predictions. This research has verified that ANNs have a position in machinery condiiton monitoring and diagnostics. However, the limited nature of these results indicates that ANNs will not solve all machinery condition monitoring and diagnostics problems by themselves. They certainly will not completely replace conventional rule-based expert systems. Ultimately, it is anticipated that a symbiotic combination of these two technologies will provide the optimal solution to the machinery condition monitoring and diagnostics problem.