Research On Intelligent Diagnosis Technology Of Key Components Of Reciprocating Equipment Based On Improved ShuffleNet

Reciprocating equipment in oil field mainly includes reciprocating pump and reciprocating compressor.Its key components such as air valve,crosshead and connecting rod have strong vibration signal impact and are greatly affected by all kinds of noise.At present,the oil field mainly evaluates and diagnoses the operation status and fault of the equipment by analyzing the operation parameters such as vibration,temperature and pressure of the field equipment.This evaluation and diagnosis method based on the operation parameters has poor effect in analyzing the impact signal of the key components of reciprocating equipment,so it is unable to diagnose the fault effectively.Therefore,taking the key components of oilfield reciprocating equipment as the research object,this thesis studies the impact signal noise reduction and lightweight network intelligent diagnosis method.The main contents are as follows:(1)Vibration data acquisition of reciprocating equipmentAiming at the problem of insufficient fault data samples of key components of reciprocating equipment,this thesis uses the reciprocating compressor test-bed of the fault diagnosis laboratory of China University of Petroleum(Beijing).By replacing the connecting rod at different crack positions,replacing the crosshead with different wear degree,adjusting the anchor bolt,adjusting the air valve bolt and other ways to simulate a variety of different fault types of key components of reciprocating equipment,and collect the vibration signals of the test-bed under each fault state.This thesis provides data support for the task of signal noise reduction and fault diagnosis by combining the fault data of the test-bed with the fault data collected on site.(2)Wavelet threshold denoising optimization of shock signal based on firefly algorithmAiming at the problem that the parameter selection of traditional wavelet threshold denoising is too simple and one-sided,and the noise reduction effect of impact signals of key components of field reciprocating equipment is poor,this thesis applies firefly optimization algorithm to wavelet threshold denoising,which can optimize the threshold of each detail component.This thesis also optimizes and explores other relevant parameters of wavelet threshold denoising.Compared with the traditional wavelet threshold denoising,this method has better impact signal denoising effect,and is more suitable for the impact signal denoising of key components of reciprocating equipment.(3)Fault diagnosis of key components of reciprocating equipment based on improved Shuffle Net networkAiming at the problems of complex architecture,large amount of parameters and low calculation efficiency of the original version of Shuffle Net network,this thesis constructs a lightweight Shuffle Net network structure suitable for the key components of oilfield reciprocating equipment in the case of few fault categories.The network architecture idea of the improved version of Shuffle Net is based on Shuffle Net V2,which simplifies the number of units and adjusts some volume layer settings in the network while maintaining the basic network architecture of 0.5x version.The test shows that the improved version of Shuffle Net network still maintains a good level of fault diagnosis and classification on the premise of reducing the amount of operating parameters and calculation time,and is more suitable for the fault diagnosis of key components of reciprocating equipment.(4)Field combination of intelligent diagnosis methodsIn view of the insufficient combination of the current intelligent diagnosis methods with the field,this thesis combines the proposed intelligent diagnosis method with a variety of fault cases of the field gas valve.The test shows that this method still has a good classification effect on the field gas valve fault signal,and the classification accuracy reaches 98.8%,which proves that this method is also applicable to the oil field.