Compressor Health Prediction Based on Multi-Source Information Fusion and LSTM

Reciprocating compressor is core equipment of petrochemical industry. Accurate state of health monitoring and faults early warning state are very important for the smooth running of a reciprocating compressor. Aiming at the lower accuracy of traditional vibration signal prediction, the LCD-LSTM is used for the accurate and stable vibration signal prediction. The LCD is used to decompose the original vibration signal into many intrinsic scale components(ISCs). And all of the ISCs are predicted by LSTM separately and accumulated as vibration signal predicted result. Meanwhile, in order to overcome the limitation of information in single vibration signal, a multi-source information fusion (MSIF) strategy is designed to build the state of health prediction model with more comprehensive information by considering multiple parameters information related to the main faults. Then a health curve which will reflect and predict the running state of compressor can be obtained. To validate the prediction capacity of the proposed MSIF-LCD-LSTM method, running data of the reciprocating compressor in offshore natural gas exploitation of China National Offshore Oil Corporation is used for modeling and testing. The results of experiments demonstrate that the proposed model has great superiority over other models and good performance of running state detection and faults early warning.