Mechanical condition diagnosis method of high-voltage reactor using its oil tank operation deflection shape

After the long-term operation, multiple mechanical defects (i.e. loosening of winding and core) of the high-voltage reactor will emerge. Mechanical conditions of these critical components inside the high-voltage reactors differ vibration of the oil tank. Thus, in this paper, array vibration sensors attached to the external surface of the tank are utilized to derive the operation deflection shape, which provides effective-rich information for defects diagnosis. The influence of the applied voltage and mechanical conditions are deeply investigated, and the results indicate that the vibration of the frequency components at 100 Hz is only related to mechanical conditions but not sensitive to the applied voltage. Thereby, nodal lines are extracted from operation deflection shape at the same frequency (100 Hz) using edge extraction, which contains less interference information compared to the traditional greyscale image. The distance between nodal line features, estimated by the Siamese network, is adopted as an effective indicator to classify various defect types and levels of multiple mechanical conditions. Experimental results confirm the accuracy and feasibility of this proposed method in identifying mechanical conditions, and the respective identification accuracy for defects of winding and iron core reaches 100% and 92.9%, which is better than that from greyscale samples.