Abstract:For mechanical equipment under the condition of variable speed, the majority signal which produced is nonstationary signal. And when using the spectral method to analyze the characteristics of signal, it will change with the time, and not to highlight the important signal features, which lead in the difficulties of fault diagnosis and identification. In order to improve this shortcoming, proposed the timefrequency analysis of order spectrum method, this method combines shorttime FFT and frequency order of the speed to get the characteristics of nonstationary signals, and also combined with the method of principal component analysis to reduce the dimensions of the extracted timefrequency order spectrum in the BP neural network, which have a fault diagnosis of gearrotor experimental platform in the nonstationary operation conditions. The results show that: The signal characteristic is not changed by the variable speed, which can be effectively identify the fault of mechanical equipment in the nonstationary operation conditions, and the identification accuracy can be increased from 93.8% to above 989%; the training speed increased from 196 seconds to 139 seconds, increased by 29%, which can achieve the fast fault diagnosis.