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

The main findings of our work on the hybrid CNN-SVM model's forging defect prediction and classification are succinctly summarized in the abstract. The model used consistently achieved high Precision, Recall, and F1-Score values, ranging from 91.67% to 94.19%, across multiple defect classes, including "Unfulfilled section on underfill," "Scale pits," "Col shut," "Flakes," "Surface cracking," "Die shift," "Remaining stress in forging," and "of Incomplete forging penetration." These measurements demonstrate the model's capacity to precisely and efficiently recognize and classify various forging faults. The Support values, which span from 775.00 to 945.00 for each defect class, highlight the resilience of the data in the collection and the model's capacity to manage a variety of forging failure cases. Additionally, the Support Ratio values offer information. These measurements demonstrate the model's capacity to precisely and efficiently recognize and classify various forging faults. The Support values, which span from 775.00 to 945.00 for each defect class, highlight the resilience exhibited by the dataset and the model's capacity to manage a variety of forging failure cases. Additionally, the Support Percent values reveal information on the proportional abundance of defect classes in the whole set of data, with each class making a proportionate contribution to the composition of the entire dataset. Our algorithm performs exceptionally well in precisely predicting forging faults across the board, with a median accuracy of 98%. The model's efficacy and consistency are further confirmed by the 93.3% averages of the macro, weighted, and micro average measures. The results above demonstrate the validity and applicability of one's hybrid model. and its capacity to improve quality control procedures and transform fault prediction in industrial welding operations.