It is always challenging and meaningful to further enhance the feature extraction capability of the convolutional neural network (CNN) and understand the internal working principle of CNN. In order to ensure that CNN focuses on more key information in the one-dimensional signal and reduces the attention to other information, a novel CNN network architecture based on the fusion of multi-sensor information and coordinated attention (CA) is proposed and applied to fault diagnosis of rolling bearings. Further, we studied the interpretability of the proposed network and intuitively revealed the feature learning process and final expression form of the proposed network. First, a coordinated attention (CA) module suitable for one-dimensional vibration signals is proposed, and a lightweight coordinated attention convolution neural network model ACNN is established. Secondly, a kurtosis-weighted fusion strategy is designed to enhance the shock feature of the sensor channels. Then, based on ACNN and weighting fusion strategy, a fusion framework for multi-sensor information (Multi-sensor ACNN) is constructed. Finally, based on the Multi-sensor ACNN, a new rolling bearing fault diagnosis method is proposed.