Research of fault diagnosis and prediction based on Bayesian networks and time series model for meta-action

Abstract

Nowadays, electromechanical products have complex structures and a higher incidence of failures. However, traditional fault prediction methods lack accuracy, causing maintenance delays. To address this, a precise fault diagnosis and prediction model based on BN (Bayesian Network) and time series, grounded in MA (Meta-Action) theory, is proposed in this research. Firstly, the FMA (Function-Motion-Action) method decomposes electromechanical products into simpler MA and MAU (Meta-action Unit), reflecting the quality characteristics of the entire product. Next, the BN model is constructed with MA and fault phenomenon as nodes, enabling the diagnosis of faulty MA. Meanwhile, the vibration signal is saved as the standard time series and the database is established. Thirdly, to compare the similarity between different time series, a time series model is constructed. A distance dissimilarity matrix is established by calculating the ED (Euclidean distance) between abnormal and standard time series. The DTW (Dynamic Time Warping) algorithm is applied to compute the warping cost between signals and identify the optimal warping path. Early warnings are issued based on the cumulative costs along each optimal path, focusing on the fault mode and faulty MA with the lowest cumulative cost. Finally, experimental analysis is conducted on the CNC (Computer Numerical Control) rotary table of a specific domestic machining center model, A comparative study using multiple methods showed that the fault prediction accuracy of the proposed method reached 90.51%, demonstrating the model’s ability to effectively predict various fault types. Besides, the macro-mean recall reached 0.896, highlighting the model’s strong performance in multi-fault classification with unbalanced data and reduced sensitivity to sample size. These results confirmed the superiority of the method proposed in this research. Furthermore, the method is applicable to other electromechanical products with structures similar to CNC machine tools.