Skip to main content

Analysis weld pool on-line adaptive weld process control hazardous environments & manufacturing applications using neural networks

Article Category

Article available in the folowing languages:

Real time adaptive welding using neural networks

Welding is a widely used process for joining individual pieces of metal. The metal pieces are fused together by using an oxy-acetylene flame or an electric arc. The current project with the aid of neural networks aims to solve the many problems and difficulties encountered when welding is automated and performed by robots.

Industrial Technologies

Automatic welding systems, though very useful, are faced with many problems associated with alignment and cast variations between the metal parts that are going to be welded together. Variations in groove form, dimensions, position and even parent metal variations result in serious weld defects. Especially when attempting to join parts of the same grade but of different cast, all welding systems up to now, develop penetration defects. The resulting repair costs as well as the costs of welding procedures themselves are dramatically raised. A consortium comprised of seven partners from five European countries, among them industry laboratories and Universities, and four industrial end users out of which two are SME’s, has carried out an in depth analysis of the problems associated with automatic welding and have developed state of the art, innovative solutions. Project partners have focused on the most dramatic problem encountered when using automatic welding; the major penetration defects. Realising that a skilled welder overcomes the problem by easily modifying the welding parameters like the welding speed or torch oscillation in real time based on his perception of the weld pool, they have concluded that automatic welding is only possible if the machining and matching tolerances are accurate. In simple words the welding machine must adjust itself to the given situation encountered. This can only be achieved with the aid of neural networks. It is the objective of the NEUROWELD project to develop a fully automatic, auto-adaptive, in real time system that makes use of neural networks and allows variations in groove form, dimensions and position and parent metals to be taken into account. Then torch position, trajectory and welding parameters can be automatically modified to avoid the production of weld defects. The variation of the parameters will be determined from a prior series of optimisation tests. Project partners currently hold the secret know-how of such a controller and have also developed a power source for high speed welding applications, which has already been extensively used. The benefits of the NEUROWELD project are numerous and much needed. Reduction of weld defects gives rise to improved security and reduced repair costs. The costs of welding procedures themselves are also expected to reduce drastically.

Discover other articles in the same domain of application