Fault Diagnosis of Hydraulic Systems with an Improved Transition Matrix Hierarchical Network subject to Multimodal Fusion

Aiming the high accuracy fault diagnosis problem in hydraulic systems for aerospace component testing, computational methodologies are mostly based on typical neural network structures to design specific models that adapt to different types of vibration signals or multimodal sensor data. However, while effective for faults with distinct signatures, these methods often exhibit poor generalization when distinguishing between fault classes that have similar features, thus failing to capture their subtle differences. This causes the model to be insensitive to the severity of the fault, which is precisely the focus of hydraulic system fault diagnosis. Therefore, this paper constructs a multimodal fusion state-attention hierarchical framework, called Transition Matrix Hierarchical Network (TMHN). Specifically, multimodal data are fused at feature level by splicing. The fusion of multimodal information provides a richer perspective for the extraction of state features, thereby improving the accuracy of fault diagnosis. In the hierarchical diagnostic network, the first layer of the network classifies the differently distributed fault types. For uniformly distributed fault types, the classification is done by the pseudo-labeling of the states within the window combined with the state transition matrix in the second layer. This improvement enables the algorithm to focus on time sequence, providing additional basis for distinguishing between faults with varying degrees of uniform distribution. The proposed method achieves an accuracy of 95.03% on our self-built datasets and 98.25% on the public CWRU datasets. Test results on the dataset indicate that our method has a better diagnosis performance and generalization.