A Performance Prediction Method Utilizing Time-Dependent Subsystem Transfers between Family Systems

Abstract

Based on cluster system theory and the Markov process, a performance prediction method utilizing time-dependent subsystem transfers between family systems is proposed in this paper. The family system is divided via the mean clustering method, with the key performance parameters of subsystems utilized as identification parameters. According to the transition quantity of subsystems in the family systems, the transition probability of subsystems between family systems is described via the Markov process. The transition matrix between subsystems is established by dividing multiple intervals of key performance states. The inter-family transfer matrix and the current family system label of the subsystem are updated in real time. Thus, the transition probability of any subsystem and the total number of subsystems to be transferred to the failure-state family system can be judged, and the remaining life can be further determined. Using the real-world monitoring dataset from the FAST Telescope, the effectiveness and accuracy of the method are verified. Due to the representativeness of family systems to subsystems and the powerful transfer-describing ability of Markov processes, the proposed method shows superiority in online prediction and performance evaluation compared to the fault data-based method, such as improvements in rapidity and accuracy. In addition, the proposed method can be used to evaluate overall reliability without reference samples, thus making the prediction method more practical in complex, large systems with small or even zero sample conditions.