Transformer Winding Condition Detection Based on Image Segmentation and Wavelet Ridges
The construction of a new power system poses increasing requirements for the safe and reliable operation of electrical equipment and power grids.As one of the essential pieces in a power system,it is important to effectively detect the winding condition of the transformer with high accuracy.Vibrational signals of an operated transformer always carry abundant information about transformer winding and have served as an important indicator for describing winding conditions.However,the vibration monitoring method has the potential blurry fault patterns in the vibration signals caused by the limitations of the time-frequency methods and the noises from sensors.It is still arduous to accurately identify the transformer's winding condition under sudden short-circuit currents.This paper introduces the image segmentation technique to analyze the vibration signals of power transformers with the extraction of wavelet ridges.Specially,the continuous wavelet transform is applied to construct the wavelet coefficient modulus matrix.Here,the complex Morlet wavelet function with bandwidth and center frequency of 4 is selected.With the gray treatment of the wavelet coefficient modulus matrix,the maximum inter-class variance method is selected to perform the image segmentation on the wavelet coefficient modulus matrix for the detailed description of the key regions.The second segmentation is further made with the proper selection of the segment threshold.After the element extraction in each region of the wavelet coefficient modulus matrix with the mode maximum method,the wavelet ridge matrix is constructed through the polynomial fitting of the maximum element coordinates.Finally,the wavelet ridge feature vector angle(WRFVA)index is defined to evaluate the condition of the transformer winding.This method can ensure the accuracy and clarity of the obtained wavelet ridges,and the defined WRFVA index can effectively capture the vibration signal variations to judge the winding condition of the transformer.A simulated signal mainly comprises multiple cosine components with a primary frequency of 100 Hz.The calculated results show that the wavelet time-frequency graph has a high resolution in the energy concentrations,which helps distinguish the target signal from noise.In addition,the vibration signals during the multiple short-circuit impulse tests on a 110 kV three-winding power transformer are analyzed.There is a noticeable increase in vibration amplitude at the 100 Hz component with the increase of short-circuit current,accompanied by the increase of high-frequency components with different degrees.Furthermore,the WRFVA of the vibration signals decreases with the increase of short-circuit impulse currents.The distribution of wavelet ridge in the time-frequency graph can be described,and the winding condition variation with high efficiency can be illustrated.In conclusion,the proposed method based on image segmentation for the wavelet ridge extraction of vibration signal and winding condition assessment reveals several key findings.(1)The grayscale threshold obtained through the maximum between-class variance method facilitates effective image segmentation.Regions corresponding to major frequency components can be accurately identified and extracted,eliminating the background noise.(2)The wavelet ridges extracted through the image segmentation have a high time-frequency resolution,sensitively reflecting the changes in vibration signals and mechanical condition variations of the transformer winding.(3)The defined WRFVA index,which reflects the wavelet ridge distribution in time and frequency domains,exhibits noticeable changes with the degradation of transformer windings.When the variation of WRFVA exceeds 2 degrees under the same short-circuit current,it indicates the presence of slight loosening or deformation in the winding.It is suggested that the winding condition of the transformer needs to be considered at this time.
Power transformerwinding conditionwavelet ridgesmaximum inter class varianceimage segmentation
国家科技期刊平台
NSTL
万方数据
