基于注意力机制优化的CEEMD-CNN-BiGRU模型的离心泵偏工况诊断OA

To address the variation characteristics of pump-body vibration signals with flow rate during centrifugal pump operation,a CEEMD-CNN-BiGRU diagnostic model optimized by an attention mecha-nism for off-design condition diagnosis was proposed to compare the influence of vibration signals col-lected at different measurement locations on diagnostic accuracy.Firstly,vibration signals of the cen-trifugal pump under different operating conditions were decomposed by complementary ensemble empi-rical mode decomposition(CEEMD)to obtain intrinsic mode functions(IMFs),and the energy mo-ments of the IMFs were calculated as feature indicators.Then,the feature vectors were fed into a paral-lel hybrid convolutional neural network and bidirectional gated recurrent unit(CNN-BiGRU)structure,optimized by an attention mechanism for training.The network simultaneously captured multi-scale spatial features and global temporal features of the vibration dataset and performed feature fusion,thereby achieving high-accuracy off-design condition diagnosis.Finally,other algorithmic models were introduced for comparative analysis to verify the diagnostic performance of the proposed model.The results show that the proposed model performs best at the radial measurement point on the volute,effectively characterizing the features associated with off-design conditions of the centrifugal pump,with a test accuracy of 99.4%.The CEEMD-CNN-BiGRU model optimized by the attention mechanism outperforms other models in diagnostic performance.In particular,CEEMD shows outstan-ding performance in extracting feature vectors associated with off-design conditions,and the introduction of the attention mechanism significantly improves model performance.These results provide a new and effective solution for off-design condition diagnosis of centrifugal pumps.