The Research Of Transformer Feature And Condition Monitoring Based On Vibtrion Anlysis Method

As the vibration based monitoring method has the advantages of defecting transformer early mechanical faults and convenient installation without affecting the running of transformers, this method becomes a hot research point of transformer on line monitoring. The windings and core are the most important sources of transformer vibration. Meanwhile, transformer vibration is also affected by many other factors. Recently, many researches of winding vibration and core vibration have been done; however, there still leaves something to be desired on account of the complexity of transformer vibration.Based on the experimental data, this paper analyses the effects of various factors and the difference between vibrations of different measuring points. Some multi-points vibration analysis methods have been presented, including Harmonic Ratio Analysis (HRA), Frequency Complexity Analysis (FCA) and Vibration Distribution Analysis (VDA). HRA researches winding status by analyzing the power ratio of100Hz harmonic; FCA estimates transformer condition through analyzing the complexity of vibration frequencies; VDA diagnoses transformer fault by analyzing the change of vibration distribution on tank. And all these methods are proved effective by experimental data.At last an electrical vibration model for transformer condition monitoring is established on the basis of the Partial Least Squares Regression (PLSR) method, using F test to verify the rationality of the model assumption as well as using T test to verify the significance of model coefficients. The model’s coefficients are calculated based on two inputs, load current and supply voltage, and one output——the vibration on transformer tank, all inputs and output are measured from an110kV transformer under steady-state condition, then some tests and analyses are performed subsequently. Finally, the assessment of transformer condition can be made by comparing the measured vibration and model estimated value. The test results and actual application show that the presented model is accurate and effective.