Multi-level adaptive risk control: synergistic maintenance-termination optimization with online parameter learning

Catastrophic functional disruptions of degrading systems during task executions usually incur severe safety repercussions and substantial losses, necessitating prompt risk mitigation strategies. It is therefore of paramount significance to develop distinct risk intervention levels following refined degradation evaluations. This study establishes a collaborative self-adaptive optimization problem of routine maintenance and task termination decisions for heterogeneous degrading systems. A Bayesian learning methodology is employed to continuously update the unknown degradation rate utilizing equidistantly sampled data; leveraging such updates constitutes an adaptive multi-level risk control policy employing age-state-dependent thresholds to select between (a) operational continuation, (b) routine maintenance partially restoring degradation at a severity-dependent cost, or (c) terminate the task instantly for overhaul. In order to minimize the total expense encompassing maintenance expenditures, task failure penalties, and termination losses, we identify several key findings into the optimality of the multi-level control policy, showcasing that the policy is representable as a monotonic threshold policy relative to degradation progression. Implementation on shipboard phased-array radar drivers confirms the framework’s operational viability and demonstrates substantial advantages over conventional strategies in risk loss prevention and task success promotion.