An Intelligent Diagnosis Method Of Reciprocating Machinery Faults In Variable Conditions Based On Vibration Mechanism Features

As important reciprocating power machinery,reciprocating compressors and piston engines are widely used in agriculture,mining,nuclear power,petrochemical industry,national defense and military industry,ship transportation,urban transportation and other fields.Reciprocating machinery is a key source of power and its normal and smooth operation is vital to the entire system.However,since the working characteristics of reciprocating machinery,it often works in a complex environment with large vibration noise,strong interference noise and variable working conditions.Meanwhile,compared with rotating machinery,reciprocating machinery has compact structure,multiple excitation sources.In addition,vibration signals have typical non-stationary and nonlinear characteristics.Various signals are coupled together,which makes it difficult to extract features.Therefore,it causes great difficulty in monitoring the status of equipment.Whether the reciprocating machinery is running smoothly or not will directly influence the uptime and maintenance cost of the entire set of equipment,which in turn will affect the economic benefits and system reliability.Therefore,the research on reciprocating equipment and fault diagnosis is of great significance.In recent years,the rapid development of deep learning methods has also provided us with methods to solve non-stationary and nonlinear signal feature extraction.Aiming at the problem of intelligent working condition identification and fault diagnosis of reciprocating machinery,this paper mainly conducts the following research work:(1)Propose a signal whole-period interception method based on image similarity comparison,and then propose a vibration signal noise reduction method based on EEMD.After denoising the vibration signal,the time domain,frequency domain and time-frequency domain mechanism characteristics are extracted;Aiming at the displacement signal of the piston rod,an accurate calculation method for the position of the axis of the piston rod is proposed from the perspective of the movement mechanism of the piston rod.Meanwhile,an envelope method of the axis of the piston rod is proposed;After that,the information entropy theory was further used to extract the information entropy characteristics of the aforementioned mechanism characteristics;(2)Aiming at the recognition problem of reciprocating machinery,the three deep learning methods of CNN,LSTM,and attention mechanism are integrated,then the convolutional recurrent neural network model based on the attention mechanism(CRNN-AM)is constructed which the hyperparameters are selected by the Bayesian optimization method;The model was verified using variable working condition data of diesel engine and reciprocating compressor,and CRNN-AM has reached over 90%of the working condition recognition accuracy of the two reciprocating machinery;(3)Aiming at the problem of coupling between reciprocating machine working conditions and faults,a fault feature extraction part was added on the basis of CRNN-AM,and the working conditions and fault features were merged to form a multi-feature extraction CRNN-AM model;The model was validated with abnormal valve clearance failures of diesel engines,piston component failures and cylinder collision failures of reciprocating compressor.The results showed that the accuracy of the model’s identification of several types of failures reached 100%.