Gearbox fault diagnosis method based on RIME-VMD and CNN-BiLSTM

Abstract

To address the issue that the selection of penalty factors and modal components in Variational Modal Decomposition (VMD) technology for decomposing gearbox fault signals relies excessively on expert experience, this study proposes a gearbox fault diagnosis method combining VMD optimized by the Rime Optimization Algorithm (RIME) with a convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM). First, the RIME algorithm analyzes gearbox vibration signals, employing minimum envelope entropy as the regularization function to calculate optimal IMF component counts and penalty factors. This algorithm leverages robust parameter search capabilities to obtain more precise frequency characteristics. Second, a convolutional neural network (CNN) extracts time-domain and frequency-domain features from the vibration signals, which are then fused. Subsequently, a bidirectional long short-term memory network (BiLSTM) extracts time-series features to classify different faults. Finally, the proposed model was validated using the dataset from the comprehensive powertrain test bench. Comparative experiments against models such as CNN-LSTM, CNN-BiLSTM, and VMD-CNN-BiLSTM demonstrated that the proposed method achieved the highest fault recognition rate with an average diagnostic accuracy of 99.25%. This approach is feasible and holds significant research value for the intelligent diagnosis and practical application of gearboxes.