Fault diagnosis and prognosis based on physical knowledge and reliability data: application to MOS Field-Effect Transistor

The reliability data, generally used for the calculation of the Mean Time To Failure, are rarely used for the online calculation of the Remaining Useful Life, as although the features measured in the reliability tests have a clear physical meaning, they are not always measurable online. In this paper, the proposed solution is the using of the physical knowledge of the components to build models linking these features to the variables measurable online. Then, the physical models are used to generate health indices whose evolution can be estimated and predicted online, and the reliability data used for initializing the trend models of the health indices. To guarantee the robustness of the remaining useful life estimation to changes in Condition Monitoring, the Wiener process whose drift parameter is updated online is proposed in this paper to model the trend of the health indices. The updating methods most used in the literature are presented and tested, the results obtained are analyzed and compared to highlight the influence of the model updating on prognosis performance. Experimental results, obtained by an application on MOS Field-Effect Transistor, show the effectiveness of the proposed method.