Fault diagnosis and remaining life prediction of key industrial equipment based on machine learning

Fault diagnosis (FD) has certain practical significance for enterprise operations and social and economic development. This paper focuses on the deep learning-based rotating machinery FD and remaining useful life (RUL) prediction methods to carry out research. The feature extraction module contained in the framework can fully capture the time dependence of the data in the time domain. The lightweight temporal convolutional network-broad learning system-fault diagnosis (LTCN-BLS-FD) framework has achieved the best diagnostic results, and the average values of the four indicators accuracy (Acc), mean precision(MP), mean recall (MR)and F1-score (MF)are all above 0.9773. The temporal convolutional autoencode temporal convolutional network (TCAETCN-Res Net) framework has achieved the best prediction results in the ablation experiment and the comparative experiment, and its root mean square error (RMSE) for RUL prediction is 0.1658, mean absolute error (MAE) is 0.1259, and R2 is 0.6567. The above verifies the effectiveness of TCAETCN-Res Net in RUL prediction. The wide learning classifier with strong nonlinear mapping ability and low complexity accurately identifies various types of faults through the feature fusion vector.