MCFPred: A Novel Multichannel Signal Adaptive Fusion Framework for Fault Diagnosis in Hydraulic Systems

Fault diagnosis plays a critical role in industrial equipment health monitoring, while effective multichannel sensor data processing is a key to efficient and accurate fault diagnosis. This article proposes a novel multichannel signal adaptive fusion framework, namely MCFPred, for fault diagnosis in hydraulic systems. Specifically, MCFPred adopts a data acquisition module to capture multidimensional equipment operational signals via multichannel sensors first, and then employs an adaptive weighted nonsymmetric projection for a time series fusion (AWNPTF) module to adaptively fuse multichannel time series data to generate high-quality 2-D feature maps. Finally, a deep residual network (ResNet) [attention-integrated residual network (AIRNet)] is devised and utilized for feature extraction and classification. In MCFPred, AWNPTF helps realize adaptive weighted multichannel time series data fusion with improved computational efficiency, while AIRNet incorporates channel attention mechanism (CAM) and spatial attention mechanism (SAM) to improve fault feature extraction and diagnostic accuracy. Extensive experiments with publicly available hydraulic system datasets are conducted to verify the effectiveness of MCFPred. Results convincingly demonstrate that MCFPred outperforms those existing methods in different cases with an average fault diagnosis accuracy of over 99% been achieved. Both stability and generalization of MCFPred are further validated with another bearing fault dataset.