Multiobjective Evolution Enhanced Collaborative Health Monitoring and Prognostics: A Case Study of Bearing Life Test With Three-Axis Acceleration Signals

This is the authors' abstract. We don't add key points for this paper.

Machine learning has brought data-driven methods for equipment health monitoring and prognostics to a new level. In fact, the weak cross-domain generalizability and "Black-box" nature remain unsolved. Hence, we propose a novel framework, multi-objective evolution enhanced collaborative health monitoring and prognostics (MEeCHMP), in which multiple relevant objectives are optimized via evolutionary approaches to construct health indicators for monitoring and prognosis of equipment health conditions. Specifically, we quantify prior degradation knowledge and domain discrepancy into an unsupervised multi-criteria decision-making problem and combine various degradation features so that the Pareto dominance should be maintained. Furthermore, the explainable prognostics model is developed to solve a supervised multi-criteria decision-making problem leading to symbolic life expressions, which are accurate and interpretable. Finally, MEeCHMP is applied to bearing life testes involving three-axis acceleration signals to demonstrate its performance in a real case.