Aiming at the problem of variability in multimodal data and missing samples, we propose a multi-modal information fusion method (MIF) based on vibration and infrared image data for effective and speediness evaluation of power transformer fault status. First, the bidirectional gated recurrent unit extracts the feature from the text data of the vibration, frequency image of the vibration, and infrared image of the power transformer, and obtains the feature vector of difference modal. And then, the cross-attention mechanism builds the relationship between the difference modal for obtaining the fusion feature. Finally, the convolution layer and full connected layer output the fault status of the power transformer. The experiment data come from the 10kV power transformer, which contains the vibration signal and infrared images. The experiment result shows that the MIF obtains more reliable diagnostic results, and provides a method basis for fault detection based on the multi-modal data of the power transformer.