The quality of welded joints is a critical factor affecting the reliability and service life of agricultural machinery, while conventional inspection methods remain dependent on operator experience and are difficult to automate in serial production. This study aimed to develop and experimentally evaluate an integrated intelligent system for automated weld defect detection using machine vision and a convolutional neural network. The proposed system combines an industrial camera, controlled lighting, an industrial computer, image preprocessing algorithms, and a ResNet-18-based classifier. A dataset of 3,000 weld seam images obtained under production conditions was divided into training, validation, and test subsets using a stratified 70/15/15 ratio. The images were classified as defective or non-defective, and data augmentation and class-weighting procedures were applied during model training. Evaluation on the independent test subset yielded an accuracy of 0.88, precision of 0.88, recall of 0.93, and F1-score of 0.90. The results demonstrate that the developed system can reliably distinguish defective from normal weld areas while providing automated inspection and digital recording of classification results. The proposed approach can be integrated into agricultural machinery manufacturing processes and provides a basis for scalable, data-driven weld quality control within modern digital production systems.
This work is licensed under a Creative Commons Attribution 4.0 International License.
