Abstract: Based on the main problems of multiple fault identification of coal mine scraper conveyor chain, an intelligent identification method of the chain fault of coal mine scraper conveyor based on improved RT-DETR(Real-Time DEtection TRansformer) was proposed. In the construction of the data set of chain, an image preprocessing method based on the method of HSV three-channel was uesd to achieve the purpose of noise reduction and enhancement of image quality. Improved RT-DETR algorithm used the MobileNetV4 as the backbone feature network, enhanced the feature extraction efficiency of the algorithm. Throuth the replacement of the normal convolution in the hybrid encoder with the more effective Ghost convolution, reduced the number of algorithm parameters and improved the speed of algorithm; Throuth using The CSPStage feature fusion module and Inner-GIoU loss function, enhanced the feature fusion capability and the accuracy of algorithm.In order to verify the effect of algorithm improvement, the results of the ablation experiments showed that: compared with the original RT-DETR algorithm, the improved RT-DETR algorithm improved the recognition accuracy by 1.6%, the FPS value by 15.5 frames/s, the model size by 36%, and the number of covariates by 35.9%. A comparative experiment was conducted using the improved RT-DETR algorithm with YOLOv8m-ghost, YOLOv8m-RT-DETR, and YOLOv10s for multi-fault detection. The results demonstrated that the enhanced RT-DETR algorithm outperformed all other algorithms across all metrics, achieving 97.6% accuracy in detecting chain breakage and wear faults in scraper conveyors. With a processing speed of 67.2 frames/s, it effectively handles both unloaded and partially loaded conditions, meeting the demands for efficient and accurate online fault detection in coal mine scraper conveyor systems.