Recursive Support Vector Censored Regression for Nonparametric Lifetime Prediction Using Degradation Paths and Failure Times in Accelerated Life Tests

Accelerated life tests (ALTs) are widely used in many applications for the time-effective assessment of products. Statistical analysis of the performance degradation paths of the products combined with actual failure times have the competence to provide more accurate lifetime estimates. We propose two nonparametric methods for analyzing the ALT data where the degradation paths and failure times are combined to estimate the lifetime of the product: the scale-accelerated degradation path model and recursive support vector regression. The proposed methods represent a novel approach in that functional forms describing the degradation paths need not be specified and the stress-life relationship need not even be assumed in order to extrapolate the lifetimes. A real-life example of an accelerated degradation test in which multi-stresses are employed to expedite a secondary rechargeable battery’s failure during test intervals is presented to illustrate the proposed methods. The results demonstrate the efficiency of the proposed methods in predicting the lifetimes from accelerated test data.