Infrared thermal imaging technology measures the intensity of external radiation emitted by objects, producing thermal images that depict their outlines and surface temperature distribution. This method holds tremendous potential for monitoring the health status of gearboxes, offering a highly promising approach to ensure their optimal performance and longevity. By utilizing the rich features of thermal images, it is possible to distinguish the health status of gearboxes from a new technological perspective. However, the traditional infrared thermal imaging technology method based on Convolutional Neural Network (CNN) has limitations in capturing the overall features and contextual relationships of images. Vision Transformer (ViT) network uses self-attention mechanism to fully explore the correlation within features, which can effectively represent the global information of high-dimensional features. Therefore, this article proposes a ViT network method for gearbox fault diagnosis by using infrared thermal imaging. The effectiveness of this method is verified by using experimental data from an industrial gearbox. Compared with other comparison methods, the proposed method exhibits better diagnostic results.