Welding defect detection: coping with artifacts in the production line
- PDF / 2,789,151 Bytes
- 11 Pages / 595.224 x 790.955 pts Page_size
- 43 Downloads / 159 Views
ORIGINAL ARTICLE
Welding defect detection: coping with artifacts in the production line Paolo Tripicchio1
· Gerardo Camacho-Gonzalez1 · Salvatore D’Avella1
Received: 21 August 2020 / Accepted: 22 September 2020 © Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Visual quality inspection for defect detection is one of the main processes in modern industrial production facilities. In the last decades, artificial intelligence solutions took the place of classic computer vision techniques in the production lines and specifically in tasks that, for their complexity, were usually demanded to human workers yet obtaining similar or greater performance of their counterparts. This work exploits a Deep Neural Network for a smart monitoring system capable of performing accurate quality checks to detect welding defects in fuel injectors during the production stage. The contribution focuses on a novel approach to cope with unforeseen changes in production quality introduced by the alteration of a particular machine or process. Results suggest that pre-filtering could avoid the retraining of custom-designed networks. Moreover, the introduction of a weighting strategy on the confusion matrix allows obtaining good performance estimations even in the case of small and unbalanced datasets. Concerning a specific demanding case of an imbalanced dataset with very few positive examples, the system displayed a 96.30% accuracy on defect classification. Keywords Automatic optical inspection · Welding · Image classification · Image processing · Imbalanced data
1 Introduction Nowadays, many industries have decided to introduce AI-driven automation systems in several phases of the production line for improving the quantity and the quality of products while reducing the interventions of human workers. Visual quality inspection for defect detection is one of the main processes. Most of the advanced techniques use Deep Neural Networks (DNNs) since they can perform complex computer vision tasks like object detection, object classification, and image segmentation, exploiting an adequate amount of training data. DNNs are experience-based, are scalable,
Paolo Tripicchio
[email protected] Gerardo Camacho-Gonzalez [email protected] Salvatore D’Avella [email protected] 1
Department of Excellence in Robotics & AI, TeCIP Institute, Scuola Superiore Sant’Anna, Pisa, Italy
and have similar performance to human workers. Furthermore, differently from other Artificial Neural Network approaches, these networks do not require to explicitly describe algorithms for feature extraction. Since the development of AlexNet [10], approaches based on Deep Learning have been encouraged. Convolutional Neural Networks (CNNs) have been widely adopted for diverse applications in industries ranging from object detection for pick and place [3] to automating optical quality inspection [5]. Such networks, however, require adequate large amounts of labeled data for training the parameters. Problems solved u
Data Loading...
