Fully Adaptive Particle Filtering Algorithm for Damage Diagnosis and Prognosis

Application of Relative Entropy and Gradient Boosting Decision Tree to Fault Prognosis in Electronic Circuits

Fault prognosis of electronic circuits is the premise of guaranteeing normal operation of a system and carrying out on-condition maintenance. In this work, the remaining useful life (RUL) of electronic elements was estimated by selecting fault features based on variance, measuring fault severity based on relative entropy distance, and conducting fault prognosis based on the gradient boosting decision tree (GBDT) model. At first, the corresponding voltages of amplitude-frequency response, under conditions of changing full-band element parameters, were extracted, and then the frequency bands with large change amplitude were further selected based on variance. Afterwards, using relative entropy distance, the degradation of element parameters was measured, and then the RUL of electronic elements was diagnosed through regression analysis by GBDT. By comparing the data with those arising from the use of other distance-measuring methods, the relative entropy distance shows a larger change range and less apt to suffer interference from noise, which is favorable to subsequent regression prediction. The regression analysis through GBDT is easy to understand and conveniently applied in engineering practice. The application of the method proposed in the study in two examples of electronic circuits indicates that the prediction accuracy of the method for RUL of electronic elements is higher than that of the other distance-measuring methods, and its application in engineering practice is convenient.