Research On The Diagnosis Method Of Transformer Winding Looseness Based On Vibration Nephogram

Power transformer is an important power equipment in the power grid to realize long-distance transmission and power distribution through voltage change.According to data statistics,the winding of power transformer is most prone to malfunction and mechanical faults in windings are irreversible,leading to the decrease of the short-circuit resistance of transformer.It is easy to be damaged when encountering the short-circuit current impact,and it can't work properly,affecting production and domestic electricity leading to economic losses.Based on the mechanical dynamic characteristics of power transformer windings,the vibration analysis method can effectively judge the mechanical results of windings by the vibration signals.This method can detect the fault when the transformer is on-line,avoiding the inconvenience caused by the outage.The traditional vibration analysis method mainly uses single point vibration to analyze the time-frequency signal of vibration signal.Therefore,studying the overall vibration of the transformer tank is an important significance to broaden the vibration analysis method of power transformer.In this paper,the 10kV model transformer oil tank surface vibration array test platform is built to obtain the transformer vibration data with piezoelectric acceleration sensor,data acquisition card and computer analysis platform.There are three working states of transformer,winding normal,winding stud loose and winding pad falling off.The vibration cloud picture of transformer oil tank surface is obtained by using cubic spline interpolation vibration restoration method.Based on the vibration cloud picture of transformer under different working conditions,three image processing methods to extract image texture features presented in this paper,GLCM(gray level co-occurrence matrix),HOG(histogram of oriented gradient)and LBP(local binary Patterns),combine with machine learning and statistical methods to classify them for achieving purpose of diagnosing transformer winding looseness.Through the experimental research,three methods can be used to detect whether the transformer winding is in fault.There are HOG feature combined with SVM(support vector machine),LBP feature combined with SVM,LBP feature combined with chi square test.In order to realize the classification of transformer winding fault types,this paper makes further research.The gray level co-occurrence matrix of the vibration cloud image is extracted,and five commonly used eigenvalues of entropy,energy,correlation,inverse different moment contrast are studied and analyzed.By extracting the gray level co-occurrence matrix of 100Hz frequency component vibration cloud image,and using the difference between its correlation mean value and standard deviation,a method can be used to diagnose whether the transformer winding is loose and classify the types.Through Studying and analyzing the above methods and compare their advantages and disadvantages,this paper proposes a method to classify the LBP features and GLCM features by using statistics,which can distinguish whether the single-phase double winding transformer winding is in normal state and whether the winding fault is bolt loose fault or pad falling off fault.This paper results provide a reference for the diagnosis method of transformer winding looseness based on vibration sensing vibration train,promote the application of image processing and machine learning in transformer fault diagnosis,wide the method of transformer fault diagnosis based on vibration analysis,and enrich the whole vibration analysis method of transformer.It has important reference significance for the application and development of smart grid to realize the unmanned on-line monitoring of power transformer.