Abstract: Aiming at the uncertainty of a single network model in the fault diagnosis of rolling bearings, and taking into account the advantages of non-contact measurement of production signals, a multi-reputation convolutional neural network (CNN) model fusion method for rolling bearing production and academic fault diagnosis is proposed. Employ multi-channel transmitter signal to train each CNN, and then uses the blending model fusion method to merge the multiple CNN models to achieve more accurate and reliable fault diagnosis. The effectiveness of the proposed method is experimentally verified with the transmitter data of the semi-consumable indoor rolling bearing test bench. Compared with other methods like the single CNN model, support vector machine (SVM), random forest method (RF), multi-layer perceptron (MLP), this method can avoid the complex manual feature extraction process, and render the higher diagnosis accuracy through model fusion furthermore, and to a certain extent, it can solve the problem that it is difficult to select the location of the transmitter in the acoustic diagnosis.