ディープラーニングによる振動信号と電流信号の回転機械診断精度の比較・評価

抄録

For comparative experiments between the vibration signal and the current signal, an intelligent fault diagnosis method based on multiclass convolutional neural network (MCNN) has been proposed to identify faults in complex rotor system. Firstly, the both signals were recorded simultaneously under steady-state for four kinds of speed. Secondly, the signal processing technique is chosen to solve the problem of modeling noise instances as true underlying relationship for MCNN. Finally, a one-versus-one and a comprehensive MCNN have been trained with both signal at various conditions individually and collectively, respectively. And the results revealed that the accuracy of the vibration signal is better than the current signal whether it is structure faults or the external bearing faults. Moreover, the fault diagnosis performance is investigated for the wide range of parameters in MCNN.