Casting Defect Forecasting with Integrated Convolutional Neural Networks and Random Forest

The abstract offers a thorough analysis of the ground-breaking casting defect prediction system, demonstrating its remarkable efficacy over a wide range of defect types. All defect types, including Blowholes, Pinholes, Swell, Cold Shut, and Hot Tears, routinely perform exceptionally well as measured by the precision, recall, and F1-score metrics listed in Table 1, routinely above the noteworthy
$96.5{{\% }}$
criterion. The present research used a dataset with 4350 photos, and the model obtained an astounding
$96.55{{\% }}$
overall accuracy. These scores highlight the model's exceptional ability to correctly detect and classify casting problems. The model was tested on a large, evenly distributed dataset, as evidenced by support values between 830 and 920 . This large dataset guarantees a thorough analysis of every defect category, confirming the model's ability to handle a range of casting flaws. The metrics for Macro Average, Weighted Average, and Micro Average all constantly registered at
$96.55{{\% }}$
, demonstrating the model's unwavering consistency and dependability in classifying defects throughout the study. Values for the Support Proportion that range from 0.19 to 0.21 highlight the model's capacity to preserve an accurate representation of every defect class and demonstrate its reliable performance in a variety of contexts. At a remarkable
$99{{\% }}$
accuracy rate, the model continuously demonstrates its capacity to provide accurate forecasts. To sum up, the hybrid CNN and Random Forest technology has worked remarkably well, offering a reliable way to anticipate casting defects accurately for a wide range of industrial applications. These results highlight the model's overall consistency and dependability throughout the dataset as well as its ability to accurately forecast particular errors. This approach offers the possibility of improved operational efficiency in manufacturing processes, which is a major contribution to the continuous search for better quality assurance measures in industries. Its ability to improve overall product quality and lessen post-production problems makes it an invaluable tool in the industrial fault detection space. Table 1 Precision visualization metrics Classes Precision Recall F1-Score support Support Proportion Accuracy Blowholes 96.47 96.47 96.47 850 0.20 0.99 Pinholes 96.55 96.55 96.55 870 0.20 0.99 Swell 96.74 96.74 96.74 920 0.21 0.99 Cold Shut 96.59 96.59 96.59 880 0.20 0.99 Hot Tears 96.39 96.39 96.39 830 0.19 0.99 Macro Average 96.55 96.55 96.55 − − − Weighted 96.55 96.55 96.55 − − − Average − Micro Average 96.55 96.55 96.55 − −