<p>Real-time fault detection has limitations when it comes to laser beam weld application. The need of specialized equipments for such applications increase processing costs, while traditional fault detection models also fail due to inaccurate results it creates which leads to the defects undetected. in-order to solve such issues the present research utilizes advanced weld image processing techniques to enhance quality and improve the accuracy of laser beam weld defect detection. This study introduces an integrated approach called Focused Super Resolution with Self Attention Single candidate and Generative Adversarial Networks (FSRSSGAN) for improved weld defect detection. This method combines Focused Super-Resolution (SR), Self-Attention (SA), Generative Adversarial Networks (GAN), and Inception V4. The SA mechanism captures long-range dependencies, while the Single Candidate Optimizer (SCO) fine-tunes model parameters, ensuring more precise detection of subtle defects. The proposed method enhances feature extraction and reduces information loss with residual dense blocks, and generates high-resolution images. The proposed method achieves 99.02% accuracy, 98% precision, 97% recall, and 99% F1-score. It enhances image clarity and precision and thereby it improves the accuracy of weld defect detection. The proposed FSRSSGAN-based framework for enhancing weld defection system demonstrates exceptional performance, improving image clarity and defect boundary precision through a modified loss function that emphasizes edge sharpness and artifact removal. The Self-Attention Mechanism (SAM) further enhances focus on key characteristics by reducing the impact of fuzzy edges, and focused SR enhances the clarity, resulting in an impressive precision and recall, particularly for small fault sizes.</p> Graphical Abstract <p></p>