Enhanced Forging Process Improvement: Utilizing CNN-SVM Hybrid Models for Precision and Efficiency

The abstract presents a detailed investigation into the classification and assessment of different surface defects observed in forging procedures, employing a hybrid CNN-SVM model within a machine learning structure. The primary objective of the research is to evaluate the model's efficiency in identifying and classifying nine specific types of forging defects: unfilled section or underfill, scale pits, cold shut, flakes, surface cracking, die shift, residual stress in forging, and incomplete forging penetration. Key performance indicators such as precision, recall, and F1-score are employed to gauge the model's accuracy across these defect categories. The forging flaw classification model has 95.94% accuracy on 10,131 pictures. The outcomes reveal consistently high-performance levels, with precision values varying from 92.04% to 98.10% and recall values from 92.56% to 96.67%. The model attains an overall macro-average F1-score of 95.93%, indicating its strong capability to manage diverse defect types with precision. Moreover, the study emphasizes the significance of support proportion and weighted averages in interpreting the model's performance concerning the dataset's size and class distribution. The support proportion, which ranges from 0.12 to 0.13 across defect types, mirrors the dataset's balanced representation of various forging flaws, ensuring thorough model training and assessment. In conclusion, the results underscore the effectiveness of the CNN-SVM hybrid model in automated defect identification during forging processes, highlighting its potential to enhance quality control procedures by accurately recognizing and categorizing defects. This investigation provides valuable insights into advancing machine learning applications in industrial environments, to enhance manufacturing efficiency and product quality through advanced defect recognition technologies.