Auto-learning process risk optimization considering uncertain degradation pathways: A bayesian-learning-informed termination approach

Safety-critical task systems operating under uncertain degradation pathways demand precise decision paradigm to balance operational continuity against catastrophic failure risks. This study addresses a risk control problem arising in mission-critical systems under degradation evolution uncertainties. To tackle potential failure risks stemming from process uncertainties, we develop a tractable risk control model that incorporates parameter learning into the adaptive termination decision process, constituting an auto-learning control-limit policy. The integrated optimization problem is representable as a finite-horizon MDP framework, which strives to mitigate the aggregate losses originating from (a) task termination and (b) operational anomalies. Theoretical analysis confirms the presence of termination thresholds along with its monotonic characteristic relative to inspection counts and degradation severities, revealing an age-state-dependent threshold structure that adapts to non-steady conditions. We further account for the implication of core degradation/cost parameters on risk alleviation, which facilitates efficient decision-making. Comparative evaluations demonstrate that the optimal policy outperforms alternative strategies over risk loss control.