Authors: S.Gayathri, Dr.S.Siva Ranjani, Dr.B.Lalitha
Abstract: This project presents an AI-based thermographic weld joint inspection system designed to automatically detect defects in weld joints using deep learning models, specifically Convolutional Neural Networks (CNN) and the YOLO (You Only Look Once) object detection algorithm. By leveraging thermographic imaging, which captures the thermal profile of welded joints, this system aims to identify inconsistencies and anomalies indicative of defects such as cracks, porosity, and lack of fusion. The proposed approach utilizes CNN for image classification to determine whether a weld is defective or not, while YOLO is employed for precise localization and detection of defects within the thermographic images. The dataset comprises labeled thermographic images of weld joints, preprocessed and augmented to enhance model performance. The CNN model is trained to distinguish between defective and non-defective welds, achieving high classification accuracy. Simultaneously, YOLO is trained to detect multiple types of defects in real-time with high precision and recall. The combination of CNN and YOLO ensures both robust classification and efficient object detection. Evaluation metrics such as accuracy, F1-score, mean Average Precision (mAP), and Intersection over Union (IoU) are used to assess model performance. Experimental results demonstrate the effectiveness of deep learning in automating weld inspection, reducing human error, and increasing inspection speed. The system is scalable and adaptable to various welding processes and materials. Deployment of this AI solution can significantly improve quality assurance in manufacturing.
