Fault Analysis And Diagnosis Of Intelligent Sucker Rod Pumping System

In the process of oil and gas production,due to the diversity of oil and gas properties and the complexity of downhole conditions,the suck rod pumping system will occur such as rod breaking off,pump bumping,insufficient liquid supply and other types of faults,which affect the normal operation of oilfield production activities.If the faults can’t be found in time,it will cause serious consequences,such as reducing the production efficiency and economic benefits of the oilfield.Aiming at the long-term problems in this oilfield,this paper collects fault data to form indicator diagram,processes and analyzes indicator diagram,diagnoses fault type and degree based on neural network,and finally puts forward relevant control scheme.According to the working principle,mechanical characteristics and structural characteristics of sucker rod pumping system,seven kinds of common faults in oilfield production are selected for analysis,and the causes of faults are identified.Through collecting the data of suspension load and displacement of beam pumping unit under normal and fault conditions,50 groups of indicator diagrams are drawn,normalized and Freeman chain code processed,and the indicator diagram sample database is established.For the diagnosis of fault types,two different artificial neural networks are established,the results are compared and analyzed.For the same sample,the weights and thresholds of BP neural network after training are not fixed,and the training results are unstable.After multiple training,the fault diagnosis accuracy of BP neural network reaches 86.25%;After training,the correct rate of SOFM network fault diagnosis is 83.75%,and the fault degree can be obtained by calculating the Euclidean distance between neuron weight vectors.For the fault of insufficient fluid supply,the daily oil production under different opening and closing time is predicted based on the neural network analysis method,and the optimal inter pumping mode is confirmed;For the breakout of sucker rod,the minimum polished rod load is predicted by neural network method,and the breakout position of sucker rod is calculated,the results show that the error is within 5.7%.In order to reduce the probability of failure,the relevant control scheme is put forward for the failure of sucker rod breaking off,pump bumping up and pump bumping down.The simulation results show that the fluctuation of the suspension load is reduced by 53.8% and71.1% in the up and down strokes respectively,and the rod running speed is successfully reduced to 0.0007 m/s and-0.0010 m/s in the up and down dead centers respectively.Finally,according to the realization process of fault diagnosis,the related software is designed,which mainly realizes the functions of data acquisition,sample data management,neural network training,fault type and degree diagnosis,fault regulation and control.