Vibration monitoring is one of the

Vibration monitoring is one of the most promising methods in a mechanical fault diagnosis [8], [9]. In [3] and [4], some effective parameters, including number, amplitude, and interval of vibration bursts, are extracted from the vibration of the OLTC using wavelet transform. Later, Li et al. [4] proposed a fault diagnosis method based on Markov chain. Meanwhile, to overcome the nonstationary property of OLTC vibration signals, phase-space reconstruction is introduced in a mechanical fault diagnostic of OLTC [6], [7]. So far, the existing methods are almost on the basis of statistical analysis. However, fault samples are often not easy to obtain. When fault signal samples are not enough, the accuracy rate of diagnostics may reduce. What’s more, how to distinguishdifferent degrees of failure is significant in the maintenance. Considering the complexity of OLTC’s inner structure, solving this problem is even harder.