Research On Methods For Reliability Modeling And Remaining Useful Life Prediction Using Degradation Data

The high-reliability and long-lifetime products have been widely existed in the aerospace, electronics industry, military, etc. Due to the high cost, small batch production and complex failure mechanism, how to estimate their reliability and remaining useful life has become an urgent challenge. With the rapid development of sensor and information tech-nology, the high-reliability and long-lifetime product with degradation failure can monitor the vital performance parameters and obtain degradation data. It can use degradation data to realize reliability modeling and remaining useful life prediction. The degradation data can provide a wealth of the information associated with the product lifetime, and make up for the lack of reliability information. This research on reliability modeling and remaining useful life prediction based on degradation data has important theoretical significance and practical ap-plication values. In this paper, we investigate this problem by using stochastic process theory and sparse Bayesian learning. The main contents could be summarized as follows.(1) This paper presents a degradation modeling and life estimation approach based on the expectation-maximization algorithm and Wiener process to solve the reliability modeling for high-reliability and long-lifetime products. This approach regards the drift parameter as a random variable that follows a normal distribution, which can merge the unit-to-unit variabil-ity into the model, and the corresponding reliability function is derived in terms of the first hitting time concept. In addition, since the resulting likelihood function contains unobserved latent variables, an expectation-maximization algorithm suitable for our problem is developed to estimate the maximum likelihood estimators of the model parameters efficiently. Finally, the effectiveness of the developed approach is demonstrated using numerical simulation and a case study for momentum wheels.(2) This paper investigates a degradation modeling and remaining useful life prediction approach based on the inverse Gaussian process with random effect, which is suitable for the products with the linear and monotonic degradation process. An effective method for parameter estimation is proposed based on Bayesian updating and expectation maximization algorithm. In our method, the stochastic parameters are updated by Bayesian method, and an expectation-maximization algorithm is used to estimate other non-stochastic parameters in the degradation model. This method can trace the latest condition of products by means of updating degradation data continuously, and obtain the explicit expression of remaining useful life distribution. Finally, a numerical example and a practical case study are provided to show that the presented approach can effectively model degradation process for the individual product and obtain better results for remaining useful life prediction.(3) A new approach for remaining useful life prediction based on the mean entropy and sparse Bayesian learning was proposed from the point of view of time series analysis, which is suitable for the products with the nonlinear degradation process. A wavelet threshold denoising method is used to preprocess the monitoring degradation data to effectively reduce the impact of noise on the prediction results. The mean entropy based method is then used to select the optimal embedding dimension for correct time series reconstruction. Relevance vector machine, a sparse Bayesian learning technique, is employed as a novel nonlinear time-series prediction model to predict the future degradation data, and achieve the probability distribution function of the remaining useful life based on the critical level of degradation. Relevance vector machine can provide the probabilistic interpretation of the prediction results, effectively quantify the uncertainty for remaining useful life prediction, and make up for the deficiency of traditional artificial intelligent methods. The experimental results demonstrate the effectiveness of the proposed approach through experimental data collected from Li-ion batteries, which can achieve better results for remaining useful life prediction.