A Weighted Residual Useful Life Prediction Method for Weibull Distribution Model under Multiple Stress

Abstract

This paper presents a weighted method of residual useful life (RUL) prediction based on generalized Eyring model and support vector machine (SVM) under multiple stress. In the first step, the Weibull distribution model is developed and the Weibull parameters can be obtained through maximum likelihood estimation (MLE). Secondly, this paper uses the generalized Eyring model and SVM model to establish two RUL prediction model respectively. Thirdly, a weight coefficient is introduced to allocate the two models. By minimizing the sum of error between real lifetime and estimated prediction, the value of weight coefficient is determined and the final RUL prediction model can be established. An accelerated life testing (ALT) case study of oil paper for power transformer is implemented to illustrate the performance of the proposed method under temperature-voltage stress. And the result of the ALT shows that the prediction accuracy of the weighted model is higher compared with generalized Eyring model and SVM model individually.