【摘要】 变频器（Frequency Converter）作为电动机的调速设备，在使用过程中会出现各种故障情况。针对传统的变频器故障诊断方法存在准确率和实效性较低的问题，该文提出一种基于改进深度残差网络(Deep Residual Network,DRN)的故障分类方法。通过对传统Resnet50模型减少卷积核数目和大小，使模型计算量明显减少，模型训练时间缩短1分54秒，提高了实效性。通过改进深度残差网络的激活函数和池化方式，保证梯度为负时，同样可以被激活，防止了高维信息的丢失，故障分类准确率提高至99.88%。分别利用不同的变频器故障分类方法进行对比实验。实验结果表明，该文改进的深度残差网络故障分类方法准确率最高、实效性最好。更多还原

【Abstract】 As the speed regulating equipment of motor,frequency converter will have various faults in the process of use. Aiming at the problems of low accuracy and effectiveness of traditional inverter fault diagnosis methods,this paper proposes a fault classification method based on improved deep residual network. By reducing the number and size of convolution kernels for the traditional Resnet50 model,the calculation amount of the model is significantly reduced,and the training time of the model is increased by 1 minute 54 seconds,which improves the effectiveness. By improving the activation function and pooling method of the deep residual network,it is ensured that when the gradient is negative,it can also be activated to prevent the loss of high-dimensional information,and the accuracy of fault classification is improved to 99.88%. Different methods of frequency converter fault classification are used for comparative experiments. The experimental results show that the improved deep residual network fault classification method has the highest accuracy and effectiveness.更多还原