CNN and Logistic Regression Synergy: Advanced Methods for Casting Defect Prediction

The abstract specifically spotlights the performance of a casting defect result by a modeling Classification. It shows that the model has achieved a very good result for all defect categories, from Blowholes to Swell and Pinhole. It gives: 98.31% for Blowholes, 98.27% for Swell, and also a couple of Pinholes so overall F1-Scores will be again finally evaluated on the Class level as follows: Those metrics show the model very well at finding where these defects are and categorizing them correctly, it is consistent which means reliable. The model is very effective in terms of support and accuracy. It was able to process 3,550 samples for Blowholes, reached an accuracy rate of 99%, as well as dealing with others such as Swells (3460) and Pinholes (35100), getting another wearing a percentage similar level. This observation proves that the model can classify a high percentage of defects correctly, corresponding to a good recall rate, and is consistent across different categories of defectivities which means it also has precision. In summary, the defect detection and classification model is well able to accomplish casting quality control as indicated by its performance metrics. Due to its very high levels of precision, recall, and accuracy, the system generates stable results which are indispensable for measuring quality standards in production plants as well as reducing defect rates.