A Data-driven Method Based On Performance Degradation Under The Complicated Stress For The Life Prediction

The lifetime prediction method based on the performance degradation is an important way to solve the lifetime prediction problem of the highly-reliable and long lifetime degradation failure products,and this method is drawing increasing attention in the reliability field.With the development of the science and technology,the functions and structures of products become multiple with complex failure mechanisms and various environment.These factors all make the traditional model-driven method,which based on the performance degradation analysis and modeling,difficult to use in the real problems,which can be reflected in:(1)The experiment data acquisition needs higher requirements for the experimental facilities.Products faced with many kinds of stress in use and it is difficult to control every kind of the stress in the process of an experiment to meet the specified stress level,therefore,the environmental stress which we gathered from experiments are always a time-variable.The traditional analytical modeling method of the degradation experiment data based on constant-stress acceleration,step-stress acceleration,and progressive-stress acceleration cannot be applied to the degradation experiment products with a time-variable stress.(2)The analytical modeling method is difficult to apply to the products with complex failure mechanism.The product's performance degradation process is determined by the failure mechanism,the analytical modeling method based on physics of failure is difficult to use when the failure mechanisms are too complex.Thus we need to research on the accumulation effect of the time-varying characteristic of the stress on the performance degradation.Therefore,focusing on the performance degradation products under complex stress,this thesis comes up with a lifetime prediction method based on data-driven as a feasible method of lifetime prediction for products of which the lifetime cannot be predicted by the model-driven method in actual engineering.The main contents of this thesis are as following:(1)The data-driven lifetime prediction method under complex stress is proposed.The data-driven lifetime prediction method under complex stress is proposed by introducing the traditional performance degradation theories,four steps are included in it: determine the degradation rate based on numerical differentiation,analyze the load spectrum of complex stress based on cluster,analyze the stress and degradation rate based on probability statistics,and predict the lifetime of the product based on cumulative degradation.The theory basis and analytical process of these four steps are descripted specifically.(2)The effectiveness of the method is verified by predicting the lithium ion battery's remain lifetime.We choose the lithium ion battery as object of our research,and acquired the time-variable temperature stress and capacity degradation value by cyclic life test.The data are analyzed and processed by the data-driven method.We thus acquire the remain capacity of the lithium ion battery and predicted the remain lifetime,finally the overlap verifying process of the effectiveness estimation is exhibited.(3)The range of the application of the data-driven method is verified by the simulation experiment data.By the simulation experiment,we produced three kind of the stress value under complex distribution,and by the relative degradation data of the simulation,we verified the veracity of the data-driven method in an overlapping style by statistical analysis of the prediction error.