关键词: CNN, 变异系数, 感应电机, 故障诊断

Abstract: To solve the problems such as excessive reliance on prior knowledge and manual experience when diagnosing rotor broken bar faults through traditional motor signal feature analysis,an induction motor fault diagnosis method integrating convolutional neural network ( CNN) and coefficient of variation is proposed. The time-domain graph of the sum of squares of the three-phase currents of the induction motor is utilized to establish a convolutional neural network for fault diagnosis using image recognition. For faults with indistinct image distinctions,the coefficient of variation is fused for fault diagnosis. After testing,this method can accurately identify the normal state,short-circuit loop faults,the breakage of one conductor and the breakage of two conductors. The square and time-domain graphs of the three-phase current reflect the fault characteristics of the induction motor under different working conditions. Thus, image recognition can be used for fault diagnosis,but there is also the problem of indistinct image distinction. By integrating the advantages of CNN and coefficient of variation,different faults of the motor can be accurately identified.

Key words: convolutional neural network,coefficient of variation,induction motor,fault diagnosis