The Fault Diagnosis Of Oil Pumping Based On Grey Profile Potential Energy And SVM

Over the years, oil has become an essential resource in our life. The demand of oil isincreasing year by year. And oil itself has become a resource that potentially determines lifequality. However, if a pump fault cannot be detected and resolved in time, a huge economicloss may be incurred due to the enormous size of the pumping well system structure, thecomplexity of the working environment, and the expensive cost of the pumps themselves.Hence, diagnosing the pump fault accurately and timely plays a crucial role in the oilfieldproduction. Currently, among various diagnostic procedures, the most applied algorithmsfor pump fault diagnosis are the Intelligent Algorithms. However, the complexity of thesealgorithms presents challenge for the computer systems when a real time fault diagnosis isdesirable. The challenge is more pronounced in our application since the number of pumpsfor an oilfield is often large. It is well known that a fault occurring in a pumping well is arare event. Therefore, in this paper we first seek a simple method that can correctlydistinguish the pump between normal operation and fault. Only when a fault is detected, aprecise diagnosis is implemented. This two-stage method could reduce the computing timedue to less computation complexity and therefore release the computational resources. Themain work is as follows:First the working mode of the pump and the characteristics of a typical fault indicatordiagram are introduced. When recognizing the indicator diagram, grey method is adopted.In the diagram recognition process, instead of compressing the indicator diagram image, thecharacteristic of the grey matrix is merged, which not only reduces the characteristic lossesdue to image compression, but also magnify the fine feature of the image edge. Thecalculation of pixel potential, cumulative potential and higher-order potential in imagepotential is discussed. The image potential energy is introduced into feature recognition ofgreyscale indicator diagram. The values of the eigenmatrix of indicator diagram areassigned by using contours potential based on the feature of indicator diagram. Theimprovement of the original potential linear reference system and the assignment thatcombines indicator diagram characteristics and contour potential make the identification ofindicator diagram not affected by image scaling, rotation, translation and other factors,which improves the accuracy of information extraction and provides foundation for precisefault diagnosis.Second, gray correlation method is used to distinguish the pump between normaloperation and fault based on eigenvectors of the previous indicator diagram characteristicmatrix. Information entropy is introduced during the diagnosis process. The entropy takeseffect in measuring the effectiveness of different characteristic factors in the fault diagnosis,which makes the diagnosis more precise.Third, for the faulty oil pump diagnosed by gray correlation above, in order to eliminate the influence of external environment, the indicator diagrams during a period oftime are superposed. Then the aforementioned feature extraction method is applied. SVM isadopted for diagnosis. Due to the asymmetry property of the fault sample, we implementeda fuzzy SVM procedure that is different from the traditional method where the class label isbinary. In this paper, the calculation of the membership function is based on both distanceand sample density, which results in a better classification rule. Experimental results showthat the proposed method improves the classification accuracy significantly.In this paper the computer workload is dramatically reduced both in informationextraction and in fault diagnosis process. The unnecessary computer resource is released,which solves the computation difficulty in diagnosing fault of pump timely and accurately.The proposed method improves the diagnostic accuracy that achieves desired expectation.