RL-based collaborative warranty and condition-based random inspection for stochastically degrading mission-critical systems

Authors:
Preprint
Publisher website
Google Scholar
Abstract
This study tackles the challenges of managing the reliability of stochastically degrading mission-critical systems by proposing an integrated lifecycle framework that combines in-warranty collaborative warranty models with out-of-warranty maintenance strategies. At its core lies the condition-based random collaborative warranty with heterogeneous preventive maintenance (CBRCW-HPM), a user-manufacturer collaborative governance mechanism that categorizes system degradation into light, moderate, and failure states. It formulates state transition probabilities along with cost- and time-related metrics and adopts the reinforcement learning–based online Dyna algorithm to provide accurate warranty cost estimation, thereby enabling adaptability to non-unique initial and terminal states. Warranty pricing is innovatively formulated using the Cobb–Douglas production function, treating warranties as independent value-added services. Complementing this, the bivariate condition-based random inspection (BCBRI) strategy leverages in-warranty operational performance data to classify out-of-warranty systems into high- and low-frequency failure types and optimizes long-term cost rates via the renewal reward theorem through scenario-tailored maintenance actions. Numerical results validate the proposed approaches, and case studies further verify that CBRCW-HPM outperforms fixed-interval warranty models, while BCBRI surpasses fixed-interval inspection schemes in adaptability and cost-effectiveness. This integrated solution effectively meets the core requirements of data-driven reliability management for mission-critical systems in the advanced digital technology era.

This publication has 32 references indexed in Scilit: