Fault Diagnosis Method Of Electrical Equipment In Single Well Of Oil Production Based On The Combination Of Acoustic And Vibration Signal Characteristics

Power supply transformer and large-scale three-phase asynchronous motor are the main electrical equipment of single well oil field,and also the power source of beam pumping unit in oil production.The implementation of state monitoring and fault diagnosis is directly related to the production safety of oil field.The energy transfer process of transformer,motor and pumping unit is complex.It is very important to find out the defect and abnormal state of the equipment in time by studying the characteristics of the accompanying signal and the relationship between the signal and the operation state,so as to arrange the equipment maintenance of the single well.Based on the analysis of the operation and demand of single well electrical equipment,this paper selected acoustic and vibration signals as the research object,extracted features according to the differences of acoustic and vibration signals of different equipment,and further studied the state identification and fault diagnosis methods of transformer and motor.Firstly,based on the analysis of the single well electrical equipment structure and operation fault,the scheme of information collection and status monitoring of single well electrical equipment operation was designed,including hardware design and software design.Then,using complete ensemble empirical mode decomposition with adaptive noise(CEEMDAN)to decompose and denoise to the transformer acoustic signal,a state identification method of gray wolf optimization(GWO)-support vector machine(SVM)was proposed,which combined the characteristics of acoustic texture and vibration entropy.Secondly,for the acoustic signal of single well motor,the background noise database and sparse representation were used to remove the noise.The acoustic signal was bandpass filtered(7khz-20khz),and the low-frequency vibration signal(within 7k Hz)was superimposed to form a more complete representation information of motor state.Overlapping data expansion was conducted after filtering and purification,and the data sample was input into1D-CNN for learning and training.The local response normalization(LRN)and kernel function decorrelation were used to improve the model structure,which reduced the bad impact of positive and negative half cycle condition fluctuation of pumping unit on the accuracy of motor fault diagnosis.Finally,the feasibility of the state identification method of electrical equipment in single well was verified by field test.