Fault diagnosis systems have become an important aid to ensure the safe operation of industrial systems and equipments. In the early stages of the fault, the diagnosis system can quickly provide early warning information and reference for the development of the fault mitigation plan. For this purpose, a fault diagnosis method based on optical gray image recognition is proposed. The method constructs the running state image of the system according to the real-time monitoring data, intercepts the gray image by CCD recording, and extracts the gray image from different resolution systems. The grayscale image features of the system are extracted from the approximate histogram. Based on the Euclidean distance between these features and the standard features, the current state is divided into the standard feature classes with the smallest distance. The experimental results show that the method can detect the fault type quickly and correctly, and can provide effective support information for the system related design and maintenance personnel.