Empirical knowledge-driven remaining useful life prediction method under unknown failure pattern

doi:10.1016/j.aei.2025.104291

Empirical knowledge-driven remaining useful life prediction method under unknown failure pattern

2026-01-07

0

PRE

AI

DOI：10.1016/j.aei.2025.104291

Original

Original request for help

Save

Abstract

En 中文

Due to the complex mechanism and harsh environments, the degradation process of machines is generally complex and different since the failure patterns typically vary from each other. In such cases, traditional-model and data-driven-based remaining useful life (RUL) methods often fail to achieve excellent results because of the lack of new failure pattern data. To solve this problem, an empirical knowledge-driven method integrating a statistical model and a deep learning model is presented to tackle the RUL estimation task for machinery under unknown failure patterns. Specifically, the failure patterns can be identified by using the support vector data description method at first. For unknown failure patterns, a Wiener process with varying parameters module is designed to generate and simulate the diversity of degradation processes, and a generalized spatial–temporal regression network is constructed to learn and transfer similar degradation information from known failure patterns to unknown failure patterns. Finally, the RUL prediction can be performed under an unknown failure pattern. By fusing empirical knowledge and synthesizing the benefits of both statistical-model-driven and data-driven methods, this proposed method offers an efficient solution for RUL prediction under unknown failure patterns. Two experiments conducted on the N-CMPASS Challenge 2021 dataset and XJTU-SY bearing dataset demonstrate the generalization and robustness of the proposed method.

Empirical knowledge-driven remaining useful life prediction method under unknown failure pattern

Abstract

Journal

Organization