A4 Conference proceedings

Penetration and Quality Control With Artificial Neural Network Welding System

Publication Details
Authors: Penttilä Sakari, Kah Paul, Ratava Juho, Pirinen Markku
Publication year: 2017
Language: English
Title of parent publication: Proceedings of the Twenty-seventh (2017) International Ocean and Polar Engineering Conference
Journal acronym: ISOPE
Volume number: IV
Start page: 54
End page: 61
Number of pages: 8
ISBN: 978-1-880653-97-5
ISSN: 1098-6189
eISSN: 1555-1792
JUFO-Level of this publication: 1
Open Access: Not an Open Access publication


In this paper an Artificial Neural Network (ANN) controlled intelligent GMAW system is created and experimentally verified. A three-layer neural network with 2 inputs, 14 nodes in each layer and 2 outputs is used to control the weld parameters. The objective is to reach quality level B weld with full penetration when joining 5 mm thick S355 steel plates with a butt weld. In the experiments, the input information for the neural network were extracted from the thermal distribution of an IR sensor and seam information (root gap, volume, shape) from a laser sensor. The output parameters of the neural network were planned to be wire feed and arc voltage. In the experimental part, the neural network control system is trained and verified using welding tests. It can be concluded that neural network weld parameter control suits well the butt weld case and full penetration and quality level B weld can be achieved.

Last updated on 2018-19-10 at 07:55