The roller bearings are very important to the gearing system of a helicopter

so it’s necessary to monitor and diagnose their conditions and faults. Because condition monitoring and fault diagnosis are similar to Hidden Markov Model（HMM） in nature

four-state continuous Gaussian mixture HMM（Hidden Markov Model） is adopted to monitor and diagnose the roller bearings conditions and faults

which is trained through the features of Cepstrum Coefficient based on Short time Fourier transform extracted from vibration signals. The result shows that this proposal method can be used to diagnose rapidly with high correctness through small training samples.