Abstract:High order spectrum can effectively remove the Gauss noise, cepstrum analysis is less affected by the signal transmission path, and fractal dimension can reflect the running status of machinery equipment and the irregularity and instability of signals. In order to effectively carry out the machinery fault diagnosis, through the combination of double spectrum, cepstrum and the fractal theory, the fractal dimensions of bicepstrum of mechanical vibration signals were calculated to detect the overflow valves faults. The experimental results show that the fractal dimension of bicepstrum can well distinguish the fault information, which indicates that the combination of bispectrum,cepstrum and fractal theory can effectively characterize the mechanical vibration , thereby machinery faults can be effectively identified.