Research on Fault Diagnosis of Axial Piston Pumps Based on Multi-Source Signal Feature Fusion and BO-Transformer-BiLSTM

The plunger pump is a core power component in hydraulic systems, and its performance is directly related to the safety and stability of the hydraulic system. To address typical fault issues encountered in practical engineering, this paper proposes a novel diagnostic method based on Bayesian Optimization (BO) integrated with Transformer and BiLSTM networks. Firstly, vibration signals and pressure signals during the operation of the plunger pump are collected, and the time-domain features, frequency-domain features of these two types of signals, as well as the wavelet packet energy coefficients obtained through wavelet decomposition are extracted.Secondly, a weighted fusion strategy based on feature importance is adopted to fuse multi-source signal features into feature vectors, thereby constructing a feature dataset.Finally, key hyperparameters such as the number of Transformer attention heads and the number of BiLSTM hidden units are dynamically optimized through Bayesian Optimization (BO). By leveraging the complementary advantages of Transformer in capturing long-range feature dependencies and BiLSTM in extracting local temporal information, fault classification and identification are realized. The results show that the model achieves an identification accuracy of over 98% for faults such as plunger pair wear, valve plate wear, and excessive swash plate assembly clearance, and maintains an accuracy of over 95% even in small-sample scenarios. Compared with single-feature methods, traditional CNN, and BiLSTM, the proposed method significantly improves diagnostic accuracy and robustness, and can meet the needs of fault early warning and diagnosis of plunger pumps in practical engineering.