Sintering machine trolley is the key equipment for producing sintered ore, in order to avoid the production accidents caused by the missing grate bars,missing wheel lock nuts and wheel falling off on the trolley,and to improve the sintering production efficiency. According to the actual operating conditions of trolleys, a fault detection scheme was formulated at the hardware level,and a sintering machine trolley defective parts detection system based on You Only Look Once version 7 (YOLOv7) and DeepStream was constructed. The YOLOv7 network model was selected to be trained on the defective parts dataset, and a weight file obtained from the training of the YOLOv7 model was deployed on DeepStream6.1 platform for accelerated inference, and Kafka messaging component was used to push the inference results. Experimental results show that the average precision of YOLOv7 for all classes of detection was 0.991,which can be used for the target detection of defective parts. System monitors the operation of sintering machine trolley in real time, and parses Kafka messages to realize fault determination rules. Through real-time fault determination, storage,display and alarm, the system realizes intelligent detection of defective parts of sintering machine trolley,and provides a new solution for maintenance of defective parts of sintering machine trolley.