Predictive Modeling for Casting Impurity Defects Using CNN-Random Forest Fusion

In the abstract, The present study focuses on the thorough evaluation of an automated learning model’s performance throughout different fault classes. The model is developed and tested using a dataset containing various impurity faults found in casting operations. Precision, recall, F1-Score, assistance, Support Proportion, or overall accuracy are used to assess each defect class, which includes Cold Shut, Hot Tear and Cracking, Misrun, Sand Inclusion, Dross, Blowholes or Porosity. The model regularly achieves high precision, and recall, with F1-Score values of above 97% across all fault classes. The confusion matrix also provides an extensive description of true positive, erroneous positive, false negative, and actual negative predictions, demonstrating the model’s ability to classify defects. The total precision of 99% demonstrates the model’s ability to effectively identify impurity flaws. The abstract design emphasises the suggested model’s promising performance, making it an important tool for assessing quality in casting operations.