Classification of operation cases in electric arc welding wachine by using deep convolutional neural networks
- PDF / 1,966,208 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 11 Downloads / 236 Views
(0123456789().,-volV) (0123456789().,-volV)
ORIGINAL ARTICLE
Classification of operation cases in electric arc welding wachine by using deep convolutional neural networks Hidir Selcuk Nogay1 • Tahir Cetin Akinci2 Received: 25 November 2019 / Accepted: 10 October 2020 Ó Springer-Verlag London Ltd., part of Springer Nature 2020
Abstract Electric arc welding machines are widely used in industry in metal technology. In parallel with the advancement of technology for the development and automation of electric arc welding machines, it is necessary to conduct scientific studies on the determination of optimal operation cases and control for optimal welding process. In this study, operating zones were classified and determined according to the measured welding current graph during the 5-s operation of the MAG electric arc welding machine. Five deep convolutional neural networks were used for this purpose. Four of these deep learning methods are pre-trained models. We used the concept of ‘‘ransfer learning’’ to use pre-trained models. According to the results we obtained from five different models, we were able to estimate the operating range of the electric arc welding machine, with 93.5% accuracy with the designed model and 95-100% accuracy with pre-trained models. Keywords Welding Deep convolutional neural network Graphical image Pre-trained Alexnet Googlenet Squeezenet Resnet18
1 Introduction The development of the metal industry is based on the ability to produce better quality products at cheaper prices by using new technologies. Welding technology has an important place in the metal industry. The welding process used in this research is the metal active gas (MAG) welder, which has been widely used in metallurgical fields [1] for many years. The increase in production level and the obligation to provide the desired quality accelerate the transition to automatic and robotic welding systems. This requires the control and monitoring of certain welding parameters [2]. The control of the welding parameters can be performed with an online monitoring system, which allows & Tahir Cetin Akinci [email protected] Hidir Selcuk Nogay [email protected] 1
Department of Electrical and Energy, Mustafa Cikrikcioglu Vocational College, Kayseri University, Kayseri, Turkey
2
Department of Electrical Engineering, Istanbul Technical University, Istanbul, Turkey
monitoring of the main welding parameters in various welding methods (MIG/MAG, TIG, REL, etc.) [3]. However, in the event of an error, it is necessary to react timely and appropriately to reject a defective product that does not meet the desired quality without performing costly and time-consuming quality control operations. For this reason, it is necessary to develop sensor systems for process control, which can perform automatic control of the welded product during the welding process. Process-controlled monitoring systems can be implemented with appropriate sensors [4, 5]. However, with the control of the input parameters of the welder, preliminary work
Data Loading...
