Reliability assessment of deep-sea human occupied vehicle propulsion systems under long-term memory degradation and random shock competing failures

Deep-sea HOV propulsion systems are affected by degradation, random shocks, electromechanical dependence, and maintenance, whose interactions complicate reliability assessment. This study develops a competing-failure reliability framework incorporating long-term memory degradation, shock effects on degradation level and rate, a failure threshold that decreases with degradation, time-varying dependence, and imperfect maintenance. A Memory Structured Path Particle Inference (MSPPI) method is proposed for dynamic estimation of degradation states and system reliability while retaining history and event information. The framework is evaluated through Monte Carlo analysis, an external run-to-failure benchmark, controlled cases with prescribed shocks and maintenance, and long-duration thruster data. The external benchmark shows lower survival-probability prediction error under natural degradation conditions, while the controlled analysis supports predictive performance under shock-maintenance conditions. MSPPI becomes more advantageous with stronger degradation persistence or multi-step prediction. Additional analyses show that long-term memory and time-varying dependence have greater effects on reliability under relevant operating conditions. Thruster data further support prediction of subsequent vibration-based mechanical condition observations. The framework provides a unified approach for modeling and assessing competing failures in deep-sea HOV propulsion systems.