Hidden Markov Model Based Updatable Reliability Assessment for Partially Monitored Systems

Real engineering systems, such as autonomous vehicles (AVs), contain sophisticated architecture and equip some critical components with sensors to monitor their condition. Since not all of the components are monitored, these systems are termed partially monitored systems. Traditional reliability evaluation approaches based on system reliability models may fail for some partially monitored systems due to a lack of clear knowledge about their internal structure. Previous research on conditional monitoring signals has yielded achievements in individual component reliability evaluation as well as reliability-related tasks such as diagnostics. This paper suggests an innovative reliability assessment method for partially monitored systems to fill the research gap of integrating condition monitoring signals into system reliability evaluation, particularly in the absence of system reliability models. The real-time sensor data, along with historical failure event records, are utilized to update the system's reliability. The failure distribution-related parameters are updated based on Hidden Markov models (HMMs). Functional kernel regression is used to express the relationship between system reliability and each component-level failure mode. To show the efficiency of the suggested method, a numerical experiment was built. The updated system reliability is valuable for risk management and maintenance strategy selection.