To address the problem that single sensor data are easily affected by their own quality and environment, which makes it difficult to monitor the overall operating condition of gearboxes, a SA-DACNN(self attention-dynamic adaptive convolutional neural network) gearbox fault diagnosis method based on multi-sensor data fusion is proposed. Firstly, the method treats the collected sensor signals from different locations as multi-channel signals and uses the multi-channel signals as network inputs simultaneously. Then, a multi-channel feature fusion module is designed, which solves the feature-level multi-channel data fusion problem by adaptively weighting the information of different channels to ensure that the important information of different channels can be effectively fused. Finally, before the fully connected layer, a self-attentive module with residual connections is used to help the network automatically learn global information and enhance the feature learning ability of the original vibration signals. Experiments are conducted on two gearbox datasets, and the results show that the proposed method has a high fault diagnosis accuracy and can meet the task of multi-sensor data fusion fault diagnosis.