Research On Performance Evaluation Of Fault Prognosis Method

The prognosis performance evaluation is becoming more and more significant to the development of fault prognosis and health management. Because only while the accuracy, precision and confidence of prognostic results can be quantified and evaluated, the fault prognosis and health management system can be a practical application.Funded by the National Natural Science Foundation of China ?Research on Cumulative Damage Tracking and Prediction of Helicopter Transmission Chain Based on Phase Space Warping Theory? and Pre-Research Project ?Degradation State Recognition and Remaining Useful Life Prediction Technology of ×××?, targeted for the fault prognosis performance evaluation, this dissertation built a comprehensive evaluation metrics framework,include metrics assessment and according applications. And proposed a uncertainty quantification method as a complement. The main contents of the dissertation include:(1)Derived the particle filter algorithm and RUL calculation process from the mathematics, obtained the RUL prediction results of faulted bearing;(2)Given a standardized definition to the expression of prognosis performance evaluation metrics, and classified those metrics by their function;(3)Proposed the progressively performance evaluation metrics assemble which are accomplished and validated in MATLAB;(4)Proposed the universal performance evaluation metrics assemble to improve their general performance, which are accomplished and validated in MATLAB;(5)Considering the prediction results without PDF information, proposed the uncertainty quantification method which is verified by experiment.The performance evaluation metrics and uncertainty quantification method proposed in the dissertation can give a comprehensive assessment of prediction results to direct the improvement of prognosis method, and is also one of the important indicators about the prognosis technical maturity.