Development of Integrated Prognostics: Application to Bearing and Bevel Gear Life Prediction

2012-12-10T00:00:00Z (GMT) by Jinghua Ma
Prognosis and health management (PHM) for complex systems have become more and more important when the economic impact of reliability related issues and the cost effective operation of critical assets is rapidly increasing. Current maintenance strategies have progressed from periodical maintenance and break down maintenance, to preventive maintenance, then to condition-based maintenance (CBM). CBM is based on using real-time data to prioritize and optimize maintenance resources. Prognosis as the most important part of CBM is becoming more and more important in these fields such as aeronautics and astronautics. The objective of this dissertation is to develop an integrated machinery prognostic methodology based on particle filtering and validate the developed prognostic methodology using real industrial case studies. In this dissertation, an integrated machinery prognostic methodology based on particle filtering has been developed. In the development of the proposed prognostic methodology in this research, these three fundamental issues have been addressed: (1) defining the state transition function using a data mining approach; (2) integrating a one-dimensional health index (HI) into particle filtering to define the measurement function; (3) developing an l-step ahead RUL estimator incorporating with a measure of the associated error. The developed prognostic methodology has been validated using three industrial case studies. The first case study concerns steel bearing prognosis and remaining useful life prediction. The second case study concerns spiral bevel gear prognosis and RUL prediction. In the last case study, the ground truth data of hybrid ceramic bearings are used to validate the methodology. The specifically contributions of the dissertation are summarized as follows: (1) An integrated particle filtering algorithm was developed in which a one-dimensional HI was integrated into particle filtering to define the observation parameters. (2) Instead of using Paris’ Law, data mining algorithm was used to build the state function. (3) Data mining based approaches were used to build the observation function. (4) An l-step ahead state parameter prediction and RUL estimator was developed. (5) The presented prognostics method has been validated using data from steel bearing, hybrid ceramic bearing and spiral bevel gear case studies.