Research on reliability of centrifugal compressor unit based on dynamic Bayesian network of fault tree mapping

For the first time, the fault tree mapping dynamic Bayesian network (DBN) method is applied to the reliability research study of centrifugal compressor units, and its usability and reliability are evaluated dynamically. First, the fault data using the K-means method eliminates abnormal data and builds a fault tree model. The structure is mapped to the DBN structure to complete the structure learning, and the logic gate is mapped to the conditional probability to achieve the parameter learning. Then the model is solved bidirectionally. The change rule of system node reliability with time is solved forward, and the posterior probability of the system node is solved reversely to complete fault diagnosis. Finally, Monte Carlo simulation analysis and the Markov process verify the dynamic reliability and steady-state availability. The results show that the Monte Carlo simulation method is almost consistent with the reliability prediction curve of the DBN model. Indicating that the accuracy of the DBN model is reliable and the computational efficiency is improved by about 81434 times. The steady-state availability calculated using the DBN model is approximately 0.99963, which is close to reality compared to the Markov process. This method can better describe centrifugal compressor units’ dynamic reliability and maintainability and provide decision support for regular maintenance of essential parts and enterprise procurement.