Fault Diagnosis Technology For CFRP-steel Sucker Rod Pumping System

Sucker rod pumping system is the major way to extract the oil in the oil industry,and CFRP sucker rod plays a more and more important role in oil production because of its high strength,low density,corrosion resistance and other advantages.Therefore,how to accurately know the working status of the pumping system and effectively predict and diagnose fault has been a significant part in improving the efficiency and productivity of the oilfield,and so as CFRP sucker rod pumping system.The indicator diagram contains a wealth of information,do it's important data for pumping unit fault diagnosis.Through the test and analysis of the indicator diagram,the operating status of sucker rod pumping system can be obtained.So fault diagnosis technology based on indicator diagram is still the main method of at present and for a long time to come.In this paper,a method for fault classification pattern recognition and diagnosis is proposed,which takes indicator diagram as the research subject and is based on support vector machine.Firstly,the dynamic simulation model is established by force analysis of CFRP-steel sucker rod pumping system.According to the determining of boundary condition,continuous condition and initial condition,the fault diagnosis model is also established.In order to obtain the force-displacement data of the suspension point as the upper boundary,threshold segmentation,refinement,dilation,edge coordinate extraction and translation scaling changes are carried out on the indicator diagram,and finally the force-displacement data are recovered.Then through the fault diagnosis model,the ground indicator diagram is converted into the pump indicator diagram,and the pump indicator diagram is normalized and meshed.The effective stroke feature,scale feature and invariant moment feature of the pump indicator diagram are extracted as the feature vector of the indicator diagram for pattern recognition input space.Finally,the support vector machine is used as a machine for fault diagnosis pattern recognition of pump indicator diagram,and the normal indicator diagram and other four types of fault indicator diagrams are used as training samples to verify the influence of different pump indicator image characteristics,kernel functions and different multi-class implementation methods on pattern recognition results using cross validation.In this paper,a method for fault classification pattern recognition and diagnosis considering effective stroke,scale feature and invariant moment is proposed,the results show that the effective stroke feature,the scale feature and the low-order moment in the invariant moment feature can classify different fault types very well,and the Gaussian kernel in the kernel function has the highest recognition accuracy and less time consuming,which is suitable for pattern recognition of fault types.