Optimized Weld Defect Detection: A Comparative Analysis of CNN and SVM Approaches

This research presents a comprehensive evaluation of a machine learning model for slag inclusion recognition in welding applications. The model demonstrates exceptional performance in accurately classifying various stages of slag inclusions. The model achieves high precision values, ranging from 81.47% to 93.98%, indicating its ability to correctly identify positive examples. The recall values, ranging from 86.80% to 88.66%, demonstrate the model's sensitivity in capturing relevant cases. The Fl-score, a balanced measure of precision and recall, consistently achieves an average of 88.2 %, further validating the model's effectiveness. Support percentage values provide insights into the distribution of slag inclusion levels within the dataset, highlighting the model's ability to handle different class sizes. The consistent performance across various evaluation metrics, including macro, weighted, and micro averages, emphasizes the model's reliability and generalizability. The findings of this research demonstrate the potential of machine learning to significantly improve welding quality control processes. By accurately identifying and classifying slag inclusions, the model can contribute to enhanced product quality and operational efficiency.