<p>To address the high construction cost of datasets for object detection, particularly in industrial application scenarios where sufficient sample images cannot be obtained from the Internet due to the specialized nature and diversity of objects and their working environments, this paper proposes a method to automatically generate synthetic datasets and train object detection models on them. First, 3D models of the target devices are created and rendered to ensure that the synthetic images exhibit realistic texture and detail. Next, a simulation environment is constructed and the 3D models are integrated into this environment using global domain randomization techniques. Finally, computer graphics methods are applied to automatically annotate target objects in the synthetic images. This approach effectively reduces the cost of data acquisition while maintaining the detection accuracy of the models. Several mainstream object detection models, including Faster R-CNN, SSD, and YOLO, are trained on synthetic datasets of anti-vibration dampers. Experimental results on real-world images demonstrate that models trained on synthetic data achieve relatively high accuracy. Furthermore, fine-tuning these models with a very small number of real images significantly enhances their performance. In addition, the models exhibit robustness against interference and occlusion.</p>