An Integrated Approach Based on Improved CEEMDAN and LSTM Deep Learning Neural Network for Fault Diagnosis of Reciprocating Pump

The reciprocating pump plays an important role in the petrochemical industry procedure, it is crucial ensuring systematic safety and stability. Since useful feature information of vibration signal from tends to be overwhelmed by background ingredients, tough realize recognition on typical modes. Aiming at extraction mechanical fault features mode recognition, this paper proposes Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise LSTM (Long Short-Term Memory) deep neural network algorithm. Firstly, IMF components are obtained decomposing signals CEEMDAN algorithm, which key parameter β k improved redefined, for optimizing SNRs (Signal Ratio) (Intrinsic Function) components. Then corresponding singular spectral entropy calculated vector constructed. classification modal based developed data dividing-training final process. study shows that proposed method can effectively extract pump, testing modes could accurately recognized model.