基于双模态特征变换PMSCNN－BiGRU－SA的机组故障诊断模型OA

To accurately identify potential faults in the operation of hydropower units,a novel fault di-agnosis model based on dual-modal feature transformation,parallel multi-scale convolutional neural network(PMSCNN),bidirectional gated recurrent units(BiGRU),and a self-attention mechanism(SA)was proposed.One-dimensional time-series signals were first converted into dual-modal images through an improved Markov transition field(IMTF)and synchrosqueezed wavelet transform(SWT),to capture temporal correlations inherent in the fault data.These representations were then fed into the PMSCNN－BiGRU－SA network to fuse,extract,and refine fault features,and a Softmax layer was fi-nally applied for fault classification.The effectiveness,advancement,and universality of the proposed model were verified using measured data from the SK power plant and the XJTU－SY bearing dataset,and a comparative analysis was conducted with other methods.The results indicate that the proposed model achieves accuracies of 100.00%and 98.75%in the two engineering applications,which is sig-nificantly better than the compared methods,providing a novel technical solution for fault diagnosis in hydropower units.