Real Time Defect Detection in Radiographic Images of Aluminum Components using Deep Learning

Despite radiographic inspection of aluminum goods being essential for ensuring structural integrity, conventional defect detection techniques have several drawbacks that make the issue ineffective. Conventional manual inspection is subjective and prone to mistakes; automation solutions are dependent on manually created features and cannot handle noisy environments with low-contrast images that have different problem forms. It suggests using convolutional neural networks (CNNs) as the foundation of a deep learning-based methodology to overcome these difficulties by achieving real-time fault identification. It uses the lightweight CNN architecture, end-to-end learning, and data augmentation to automatically discover and classify flaws in radiography images. In comparison with existing systems, the model achieves competitive results of $\mathbf{9 5. 8 \%}$ accuracy, $\mathbf{9 4. 1 \%}$ precision, 93.3% recall, and 93.7% F1-score. Furthermore, under high Gaussian noise, it achieves above 80% accuracy, demonstrating good resistance against noise. The proposed system industrial relevance for accurate, dependable, and fast non-destructive inspection is confirmed by the results.