Research On Fault Detection Technology Of Beam Pumping Unit Based On Intelligent Video Analysis

Petroleum,also known as the "blood of industry",is an essential strategic reserve resource for China’s socio-economic development.However,China’s dependence on foreign oil is as high as about 70%,and it is facing profound changes unseen in a century.It is even more necessary to adhere to bottom line thinking,ensure domestic oil extraction,and ensure bottom line demand for oil.In China,beam pumping units are mainly used in oil production.Because of poor working environment and conditions,the probability of fault is high.However,traditional manual on-site inspections and video surveillance inspections cannot detect faults in a timely manner,resulting in a decrease in the efficiency of oil production in oil wells.At present,the automatic detection method for pumping unit faults mainly uses sensors in the oil field Internet of Things to collect data.However,sensors generally have issues such as data drift,which reduces the accuracy of measurement data and affects the reliability of the automatic detection function for pumping unit faults in the oil field well site Internet of Things.In recent years,with the acceleration of digital oilfield construction,the coverage of cameras in oilfield well sites has been continuously increasing,but the level of detecting pumping unit faults based on video analysis is still relatively low.In order to enhance the reliability of the oil field Internet of Things pumping unit fault detection function and improve the level of video analysis based detection of pumping unit faults,this paper proposes a pumping unit fault detection method based on intelligent video analysis using oilfield well site cameras,and designs and implements a pumping unit fault detection system based on intelligent video analysis.The main research content of this paper is as follows:(1)To solve the problem of inaccurate horsehead target extraction,HSB is used to optimize the backbone network of SOLOv2 and enhance the fineness of horsehead mask extraction.Through field photography,the horsehead instance segmentation dataset of the pumping unit is produced.Then,the performance of different backbone networks and different instance segmentation models are tested respectively.Experiments have shown that the improved SOLOV2 in this paper has better segmentation accuracy on the horsehead dataset,laying a better foundation for subsequent fault detection of pumping units,and the inference speed can also meet real-time requirement.(2)In order to further detect the faults of the pumping unit,the characteristics of the horsehead movement time series data during the normal state of the pumping unit,belt slipping,well stopping,and horsehead overturning are first analyzed;Then,based on the characteristics of these data,create a dataset that includes the timing data of horsehead movement during normal states and belt slip faults;Finally,Fourier transform was used to extract features,and the performance of fruit fly optimization algorithm,particle swarm optimization algorithm,and grid search algorithm in SVM parameter optimization was compared.The results show that the grid search algorithm is significantly better,with good detection results and practical value.(3)Considering the actual application scenarios of the algorithm in oil fields,the algorithm in this paper is deployed to the edge end for edge-cloud collaboration.Then,the requirements of the edge end and the server end are analyzed,and a pumping unit fault detection system based on intelligent video analysis is designed,implemented,and tested.This proves that the system implemented in this paper has good performance and certain practical value.