Application of CNN and KNN Algorithms for Casting Defect Classification

The study abstract outlines the utilization of advanced machine learning to identify and categorize casting defects such as Blowholes, Pinholes, and Swell with high precision, recall, and F1-scores. The precision and recall rates for Blowholes are reported at 98.30%, while Pinholes and Swell exhibit slightly higher rates at 98.31% and 98.27% respectively, indicating the model’s consistent performance in accurately detecting defects. The distribution of the dataset reveals 3520 instances of Blowholes, 3550 of Pinholes, and 3460 of Swell, ensuring a balanced representation that is essential for effective model training. With 10,530 photos in the dataset, our model produced the overall accuracy of 98.29%. This balanced distribution supports the model’s capacity to generalize well across different defect types. With an overall accuracy of 99%, the model demonstrates its proficiency in distinguishing and classifying casting defects, highlighting its significance in enhancing quality control in manufacturing processes. This high level of accuracy is crucial for minimizing defects and ensuring product consistency. In conclusion, the research underscores the importance of advanced machine learning techniques in industrial quality assurance. By precisely identifying and categorizing casting defects, this approach contributes to enhancing product quality and reliability in manufacturing. Future studies could focus on expanding the dataset to include additional defect types or optimizing the model for broader applications in industrial settings.