Optimal Degradation-Aware Control Using Process-Controlled Sparse Bayesian Learning

Abstract

Efficient production planning hinges on reducing costs and maintaining output quality, with machine degradation management as a key factor. The traditional approaches to control this degradation face two main challenges: high costs associated with physical modeling and a lack of physical interpretability in machine learning methods. Addressing these issues, our study presents an innovative solution focused on controlling the degradation, a common cause of machine failure. We propose a method that integrates machine degradation as a virtual state within the system model, utilizing relevance vector machine-based identification designed in a way that offers physical interpretability. This integration maximizes the machine’s operational lifespan. Our approach merges a physical machine model with a physically interpretable data-driven degradation model, effectively tackling the challenges in physical degradation modeling and accessibility to the system disturbance model. By embedding degradation into the system’s state-space model, we simplify implementation and address stability issues. The results demonstrate that our method effectively controls degradation and significantly increases the machine’s mean time to failure. This represents a significant advancement in production planning, offering a cost-effective and interpretable method for managing machine degradation.