Individual Residual Life Prediction Method of Electronic Components Based on Model Fusion

Electronic components are usually the weak parts of a system and are generally required to fulfill critical tasks. If a seriously degraded electronic component is not replaced in time, its failure may result in the failure of the whole system. Therefore, it is significant to accurately predict the residual life of electronic components in the system. Due to factors such as complex failure mechanisms, little prior information, and insufficient monitoring data, the degradation model of an electronic component is usually uncertain, which may cause an inaccurate prediction of the residual life. However, the existing studies did not consider the predictive ability of a model and required much prior information or degradation data. Therefore, a residual life prediction method for an individual electronic component based on model fusion is proposed in this paper. Alternative models are fused based on a novel model evaluation index that incorporates the fitting ability, complexity, and prediction accuracy of a model. Then the residual life of an electronic component is predicted based on the fused model. The proposed residual life prediction method is focused on the case that a single performance parameter can characterize the performance of an electronic component. Firstly, the alternative degradation model is established for typical and common degradation patterns of electronic components. Secondly, the alternative residual life distribution model is derived based on the alternative degradation model. Thirdly, the maximum likelihood estimation method is adopted to estimate the unknown parameters in the alternative model using the available degradation data collected up to the current time. Fourthly, the probability that an alternative model is the true one is constructed according to the proposed novel model evaluation index. Then the alternative residual life distribution models are fused via the law of total probability to obtain the fused model of the residual life distribution. Fifthly, the expected residual life is taken as the point estimate of the residual life and calculated based on the fused residual life distribution model. Three real cases are carried out to verify the proposed residual life prediction method. Through comparison analysis, it is shown that the proposed method is superior to the methods that predict the residual life based on the best model selected according to a traditional model evaluation index, such as the Akaike information criterion (AIC) and the value of the likelihood function, or the proposed index. Moreover, it is verified that the proposed method can ensure good prediction accuracy, even if the regularity of the degradation data is poor and the amount of the data is small. Some specific conclusions are summarized as follows. (1) The proposed novel model evaluation index is comprehensive since it simultaneously considers the fitting ability, complexity, and prediction accuracy of a model. (2) The strategy of model fusion can reduce the risk of abandoning a model with better prediction results and make a fused model have the characteristics of all the alternative models. The drawbacks of using a single model can be overcome to a certain extent. (3) The proposed residual life prediction method for electronic components is independent of much prior information or degradation data and has a high prediction accuracy and powerful practicability. © 2023 Chinese Machine Press. All rights reserved.