Remaining useful life prediction of equipment using a multiobjective optimization reinforced prognostic approach

Data-driven methods have rapidly advanced equipment degradation monitoring and prognosis. However, traditional deep models rely on weak prior degradation knowledge and may not effectively incorporate degradation damage information. To address this limitation, a Deep Multiobjective Optimization Reinforced Prognostic (MORP) framework is proposed in this paper for equipment health prognosis. Specifically, a priori degradation knowledge and multi-source deep features are combined at both the feature and health indicator (HI) levels. They are then quantified into an unsupervised multi-objective optimization decision. Preceding this step, a multi-degradation criterion and HI generalizability are formulated as a multi-objective function, with the aim of enhancing the generalizability, monotonicity, tendency, and robustness of HIs. Comprehensive Health Indicators (CHIs) are then constructed while retaining the advantages of the Pareto frontier, using a reinforcement learning-guided swarm intelligence optimization method. To address anomalies within CHIs, a HI burr correction method featuring an interpolation-extrapolation term is introduced. Additionally, the prediction of remaining useful life is accomplished through a supervised prognostic scheme. Finally, the proposed methodology is applied to equipment datasets to validate its performance.