Reliability assessment of multi-state phased-mission systems by fusing observation data from multiple phases of operation

Multi-state phased-mission systems (MS-PMSs) are multi-state systems that intend to complete multiple, consecutive, and non-overlapping phases of operation with different functional requirements. Even though many computationally efficient tools have been developed to facilitate reliability modeling of MS-PMSs in the past few decades, reliability assessment of an MS-PMS by observation data is still a difficulty because of the state and phase dependencies. This paper devotes to addressing this challenge by putting forth a new method that can assess reliability of an MS-PMS by fusing observation data collected from multiple phases of operation. A dynamic Bayesian network (DBN) model is constructed to characterize the state dependence between an MS-PMS and its units as well as the phase dependence of each unit across consecutive phases. By putting observation data into the DBN model, the joint probability distributions of the nodes of the MS-PMS at any time slice can be updated, and with which the unknown parameters associated with the state transitions of all the units can be furtherly estimated by a tailored Expectation-Maximization (EM) algorithm. The reliability function of the MS-PMS can be, then, assessed by these estimates of all the units. Two illustrative examples, including a numerical case and a data collection robot are exemplified to demonstrate the effectiveness of the proposed method.