The vibration signals of an operated transformer are closely related to the mechanical condition of winding and core.  However, those signals, collected from the transformer tank, are resulted from the nonlinear overlap of winding vibration and core vibration, and transmitted by transformer oil. How to separate the vibration signals of transformer tank is particularly important for the mechanical condition monitoring of winding and core of power transformer with high accuracy. In the paper, Morphological Component Analysis is applied to separate the vibration signals with the proper evaluation index. Then the normalize feature frequency (NFF) for the vibration signal are proposed to assess the mechanical condition of winding and core of transformer. The on-lining monitoring vibration signals of some 500kV transformer are analyzed to verify the effectiveness of the proposed method.
The vibration on-lining monitoring system was installed for a 500kV power transformer, where the vibration signals of transformer tank and load current are collected by acceleration sensors and current transducer with the sampling frequency of 10kHz. Since the applied voltage of power transformer always keeps constant and the load current is small at night, the sparsity of the vibration signals is estimated and the K-SVD dictionary is selected in MCA algorithm to separate the vibration signals of winding and core. Then the NFF of vibration signals are calculated to analyze the mechanical condition of winding and core.
With the estimated sparsity of the vibration signals and the selected number of atoms, the K-SVD dictionary is trained with the data set which is composed of the vibration signals of power transformer. Then the block relaxation iteration algorithm is applied to solve the object function in MCA method for the separation of winding vibration and core vibration. With the separated vibration signals of winding and core, the time-varied curves of NFF of vibration signals of winding and core can better assess their mechanical condition.
The vibration signals of winding and core are separated well by MCA method according to the and the evaluation index of relevant coefficient, signal independent coefficient and ratio of signal to noise. The time-varied curves of NFF of vibration signals of winding and core show that the mechanical condition of winding and core of power transformer are in good order.

condition monitoring,power transformer,,Morphological Component Analysis,vibration signal,feature frequecny