A hybrid Dynamic Bayesian network method for failure prediction of a lock mechanism

• Published in: Probabilistic engineering mechanics Vol. 74; p. 103532
• Main Authors: Pang, Tianyang, Yu, Tianxiang, Song, Bifeng
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.10.2023
• Subjects: DMMHC algorithm; Dynamic Gaussian Bayesian network; Failure mode; Multiple influence factors; The lock mechanism; Dynamic Gaussian Bayesian network; DMMHC algorithm; Failure mode; The lock mechanism; Multiple influence factors
• ISSN: 0266-8920
• DOI: 10.1016/j.probengmech.2023.103532

This paper aims to construct a failure prediction method for a lock mechanism system to increase prediction accuracy. The two major failure modes in lock mechanisms are kinematic accuracy failure and clamping stagnation. One failure mode is affected by another failure mode as a result of multiple influencing factors that are dependent on one another. Besides, the characteristic values of failure modes are challenging to acquire in some particular situations, such as when sensors are not placed. To address these issues, this study aims to propose a hybrid Dynamic Gaussian Bayesian network (DGBN) model for the failure prediction of a lock mechanism, in which correlations between each influencing factor and correlations between each failure mode are taken into account simultaneously. The improved DGBN model is integrated with measurement data and system failure analysis. The presented model relies on the information about influence factors that are more easily measured in practice and can account for multiple hidden variables for which measurement data is missing. Furthermore, a failure prediction framework is developed based on the proposed model. Finally, the proposed prediction method is tested by the application of a lock mechanism. A comparison is made between the improved method and the data-driven method. The results show that the proposed method can predict failures relatively accurately, even when partial measurement data are missing. The prediction error narrowed from 10% to less than 4%. •Build a hybrid DGBN prediction model for dynamic performance evolution.•A multi-source information fusion-based prediction framework is developed.•The dependence between failure modes and influence factors is modeled accurately.•We investigate online missing data cases of influential factors.•Proposed framework is extended to handle missing data.