The Use of Artificial Neural Networks Tomodel Distortion Caused by Welding

Abstract

Electric arc welding is the preferred joining process for many metal-using industries, from small sheet metal products to very large ships. Where thin-walled structures are built, the welding process causes the structure to distort, and this can lead to serious problems in fit up and assembly that require extensive rework. The mechanisms that lead to distortion have multiple causes, and thus modelling distortion in order to predict likely effects has in the past proved to be difficult. This paper outlines a novel approach using artificial neural networks, taking advantage of the ability to deal with multiple inputs. The results are promising, and further research is planned, alongside full-scale industrial trials of the software that has been developed.

Key Words:

• Distortion
• prediction
• design for production
• welding
• artificial neural networks

Additional information

Notes on contributors

G. Bruce

George Bruce is Professor of Ship Repair and Conversion in the School of Marine Science and Technology at the University of Newcastle. He had a 30-year industrial career at Swan Hunter Shipbuilders, British Ship-builders, and Appledore consultants, where he became technical director. He has advised many UK and overseas marine companies, including BAE Systems Marine, Devonport Management Ltd., Vosper Thornycroft, and A & P Group. His current research is focused on the prediction and management of uncertainty in manufacturing production, in particular in large made-to-order products including ships and offshore structures. He has published several papers on prediction of weld distortion and is completing a project to model distortion in plate welds using artificial neural networks. He has also published work on design for producibility, which is an important component of this research. He joined the School of Marine Science and Technology in 1996, as a lecturer, and was promoted to Senior Lecturer in 2001 and then to a Professor in 2004. Since 1996 he has been awarded six teaching company programs, three European Community projects, and three LINK-funded projects within a program managed by the Shipbuilders' Association (SSA). He is a member of the SSA Management Council and has hosted several workshops and research dissemination events for the association. He was chairman of a committee of the International Ship Structures Congress (ISSC) 2000–2003.

M. Lightfoot

Martyn Lightfoot is a research associate in the School of Marine Science and Technology at New-castle University. He worked in an offshore fabrication yard for three years in the Welding Engineering Department, developing weld procedures and specification. His current research is on the prediction and management of weld-induced distortion. He has published several papers on prediction of weld distortion.