Quality assurance has become vital for modern industrial systems, because even a single failure in delivering substandard products can drag down brand image and customer satisfaction as well as cause an economic loss. It is very difficult to measure quality in mass production with high precisionand manual inspec-tion work often involves such errors, discrimination errors and low efficiency. With the rapid development of intelligent manufacturing, computer-vision-based automated inspection has become an effective approach for reliable and large-scale quality assessment. Here, we examined acasting component dataset called submersible pump impellers to enable efficient pumping machine. The study concentrated on defect identification ofthe casting components. We used defect-free as well defective images to train deep learning models, taking advantage of OpenCV ViT (Vision Transformer) and YOLO-based architectures. The exper-imental results show that the Vision Transformer significantly outperforms all other modelswith impressive 99.9% accuracy. Because of this, it is an extremely effective model for automatically detecting cast defects. This study shows how AI-based estimate techniques can increase industrial productivity, lower human error, and improve quality control in large-scale manufacturing