Research On The Condition Monitoring And Fault Identification System Of Beam Pumping Units

The traditional oil field pumping unit maintenance mainly relies on the staff to monitor the oil production site on a 24-hour shift.The diagnosis efficiency is low and the diagnosis relies on the single data.There is no systematic analysis of multiple data.It is easy to cause the occurrence of failures that endanger the safety of personnel,such as rod corrosion and rod breakage.The oil field production is also under the environment of high load and strong noise,resulting in poor monitoring conditions and high cost of oil field pumping unit working conditions.In view of the need for monitoring the working conditions of beam pumping units in oil fields,the project independently develops a low-power collection system of working conditions information with temperature sensors,noise sensors,vibration sensors and other sensors.After preprocessing the collected data,in order to study the real-time production data changes of beam pumping units under typical working conditions,this paper proposes an improved depth residual shrinkage neural network model for monitoring the working conditions of beam pumping units,by building several neural network models for comparison,it is verified that the improved depth residual shrinkage neural network model has higher diagnostic accuracy and stability.First of all,in view of the problem that the pumping unit works in remote and harsh areas,and equipment data acquisition and equipment maintenance need manual inspection,the multi-sensor data acquisition circuit and hardware selection are designed.The design has low power consumption and low cost,and can detect multiple types of progressive signals of pumping unit working conditions in real time(such as temperature signal,noise signal,vibration frequency signal,bearing offset signal and other fault signals).Upload the data to the system of control room,analyze,mine and fuse the fault signal feature information through neural network technology,and realize the diagnosis of pumping unit working conditions.Secondly,considering the poor coverage of mobile signals and serious signal occlusion at the oil production site,in view of the problem of data transmission in the signal blind area,a low power consumption scheme for wireless communication has been developed,which can effectively make up for the defect of data transmission in the4 G signal blind area.After the gateway device converts the received data into data frames,it connects with the host computer through ethernet and RS-485,and uses the Modscan master station software to poll and read the gateway data according to the MODBUS-RTU protocol.Finally,according to the characteristic data of different fault types measured by the pumping unit working condition monitoring system,the data set of the pumping unit working condition is established,the data set is manually checked and cleared of outliers,and then normalized,and the data are inputed into different different neural network models,and the model establishment and algorithm comparison test are completed in Python.The test results show that the training error of the improved deep residual shrinkage neural network model is lower than that of the convolution neural network model,the residual neural network model and the deep residual shrinkage neural network model,the diagnosis response speed is faster,and the average fault diagnosis accuracy is improved by 3%-8%.