A Novel Criterion of Degradation Model Selection for Remaining Life Prediction

In stochastic data-driven approaches for remaining useful lifetime (RUL) prediction, the selection of an appropriate degradation model is critically important. However, the majority of existing methods have primarily emphasized goodness-of-fit measure while largely overlooking prediction performance. In this paper, we propose a comprehensive evaluation framework designed to facilitate the selection of an appropriate degradation model, with the objective of enhancing the accuracy of RUL predictions. Unlike conventional goodness-of-fit measure, a novel criterion of model selection is developed by integrating three critical dimensions: goodness of fit, model complexity, and generalization performance. This integration is specifically rooted in the performance of RUL prediction, which subsequently determines the weight assigned to each indicator. Finally, a numerical example is provided to validate the proposed method.