Abstract:For the problems such as variable load and large noise interference of the gearbox of CNC machine tool in actual working environment,it is difficult for the traditional neural network to fully extract the fault characteristics in the signal.In view of this,a multimodal ensemble convolutional neural network(MECNN) was proposed for the gearboxes fault diagnosis of CNC machine tools.The multimodal fusion technology was combined with multiple convolutional neural networks,and the fast Fourier transform method was used to convert the time domain signal into a frequency domain signal.The two convolutional neural network were trained by using time domain signals and frequency domain signals,so that the model could extract features from the time domain and frequency domain respectively,then the shallow features were fused.Finally,the fused features were input into the convolutional neural network for deep mining of fault features and fault diagnosis was carried out.The gearbox dataset of Southeast University was used for verification,and the two feature fusion methods were designed and compared.The experimental results show that under noise,the accuracy and robustness of the MECNN model for fault diagnosis are better than those of single time-domain CNN and frequency-domain CNN.