Multi-Channel Gramian Angular Field Combined with Attention-Enhanced ResNet for Hydraulic Support Solenoid Valve Fault Diagnosis

Abstract

Solenoid valves are the most critical components in hydraulic supports. However, the solenoid valves suffer severely from multiple faults due to the harsh underground working conditions and intense dynamic impact loading, causing a large number of roof uncontrolled collapse accidents. Existing support diagnosis focuses only on the control system, while the solenoid valves, which deserve more attention, have been largely overlooked. This paper proposes a novel support solenoid valves fault identification approach by integrating multi-channel gramian angular field with an attention-enhanced ResNet architecture. A scaled support test rig that replicates the operating principle of a ZY3000/09/20 shield support is firstly constructed, and the triaxial vibration responses under typical valve faults are then obtained via reproducing the faults on the test rig. The multi-channel gramian angular field is introduced to investigate the temporal dependency of the vibration signals by mapping the one-dimensional time-series data into two-dimensional images. To effectively identify fault type, the ResNet50 backbone is enhanced by the convolutional block attention module, which adaptively re-weights channel- and spatial-wise feature maps to amplify fault-relevant signatures. Ablation experiments demonstrate that both the multi-channel GAF encoding and the CBAM attention module contribute meaningfully to the final diagnostic performance.

Conflict of Interest Statement

The authors declare no conflicts of interest.