Research On Intelligent Fault Diagnosis Method Of Fracturing Equipment Pumping System

In the energy production process of petrochemical,fracturing equipment is one of the important extraction equipments.The pumping system is the core operating part of the fracturing equipment operation.Its main function is to inject the fracturing fluid into the oil layer underground by using high pressure,which can increase the recovery rate of the crude oil.Fracturing equipment generally works in the field,the pumping system usually works under big load with multiple working conditions,which is easy to cause faults while running for a long time,if the fault cannot be detected immediately,it may cause accident with huge economic loss.Therefore,it is of great significance to conduct the fault diagnosis of the pumping system,which will improve the safety and economy of the fracturing operation.The structure of the pumping system is complex,there are many internal vibration sources,and the job site is full of noise.It is difficult to extract the intrinsic characteristics of the fault accurately by using traditional fault diagnosis methods,and it is hard to identify the early fault of the pumping system quickly.Based on these problems,this paper doing the research on the intelligent fault diagnosis method of the pumping system.The main research contents can be summarized as follows:(1)The specific structure and working principle of the pumping system about fracturing equipment are analyzed,and the main causes of common faults in pumping system are studied.Some problems of traditional signal processing methods in pumping system fault diagnosis fields are analyzed.(2)Aiming at the problem that the impact features of bearing early fault signals which is not obvious,and the fault characteristic frequency is easily covered by the natural frequency of vibration,a fault feature extraction method based on robust local mean transient-extracting transform(RLMD-TET)is proposed.Soft sifting stopping criterion and histogram statistical theory are used to optimize local mean decomposition,which greatly reduced the meaningless components generated in the modal decomposition process of rolling bearing fault signals.Meanwhile,the collective delay operator is used to redistribute the energy of the correlated components,which increased the transient impulse characteristics in reconstructed signal.The results of the envelope analysis show that this method can effectively strengthen the fault features of the bearing,which has strong ability in diagnosis early faults.(3)Aiming at the problem that the lowly time-frequency resolution and the uneven distribution of time-frequency energy in extracting the features of the fault signal about the hydraulic end in pumping system,A pumping system fault diagnosis method based on improved multisynchrosqueezing transform(IMSST)is proposed.By adjusting the rounding conditions of time-frequency coefficients,the problem of energy coefficient redistribution disorder in signal processing of MSST and other methods is effectively alleviated,and the crossinterference between time-frequency bands is reduced while efficiently squeezing discrete timefrequency energy.The extraction results proved that this method can accurately capture the vibration frequency change features during the operation of the pumping system,and the diagnosis accuracy of the hydraulic end fault have been increased.(4)Aiming at the problem that the fault features of pumping systems are complicated in frequency,which is difficult to identify the fault features.An intelligent fault diagnosis method based on deep ridgelet convolutional auto-encoder network(DRCAN)is proposed.Based on the deep autoencoder,the ridgelet function is used to improve the sigmoid function,and the ridgelet convolutional autoencoder is constructed by combining the idea of one-dimensional convolutional neural network(1D-CNN)which is weight sharing,and the deep ridgelet convolutional autoencoder network model is formed by stacking the ridgelet convolutional autoencoder.The model is trained and tested with fault samples which obtained by IMSST and other time-frequency analysis methods.The experimental results show that this method can efficiently learn the potential characteristics inside data,which can improve the correct rate of pumping system fault identification.