A Study on The Reliability Analysis of The Integrated Dual-output Converter Using The Bayesian Networks

This paper focuses on the reliability studies of the output voltages in a multi-port DC-DC converter called the integrated dual-output converter (IDOC). The Markov process and Bayesian networks (BNs) are used to perform reliability analyses. Bayesian networks (BNs), developed based on machine learning techniques, are suitable for unhiding the dependencies along with estimating the conditional probabilities for reliability studies. The BNLearn and the pgmpy are libraries developed in the Python environment that include the main parts of obtaining a BN: structure learning, parameter learning, and inference. A dataset including the reliability model of each device is created using state sampling and Monte Carlo simulation, considering the open-circuit fault. Next, the hill-climbing (HC) and Bayesian information criterion (BIC) methods are applied to the dataset as searching and scoring functions, respectively, to determine the structure of the BN. Finally, using parameter learning and inference, conditional and joint probability distributions (CPDs and JPDs) are obtained, yielding the reliability of output voltages. Theoretical calculations through the Markov process validate the accuracy of the obtained BN.