A desired weld feature such as geometry can be produced using multiple sets of welding variables, i.e., different combinations of arc current, voltage, welding speed, and wire feed rate. At present, there is no systematic methodology that can determine, in a realistic time frame, these multiple paths based on scientific principles. Here we show that the various combinations of welding variables necessary to achieve a target gas metal arc (GMA) fillet weld geometry can be systematically and quickly computed by a real-number-based genetic algorithm and a neural network that has been trained with the results of a heat transfer and fluid flow model. The neural network is computationally efficient and, because of its origin, its input and the output obey the equations of conservation of mass, momentum, and energy. A genetic algorithm is used to determine a population of solutions by minimizing an objective function that represents the difference between the calculated and the desired values of the penetration, throat, and leg length. The model proposed here is different from traditional reverse models, since they cannot provide a choice of solutions and usually do not confirm to any phenomenological laws. The computational methodology provided a choice among various sets of current, voltage, welding speed, and wire feed rate for achieving a given fillet weld geometry specified by a set of leg length, penetration, and throat. The computed results were adequately verified by comparing with experimental results. The results provide hope that other weld attributes can also be tailored based on scientific principles in the future.
|Original language||English (US)|
|Journal||Welding Journal (Miami, Fla)|
|State||Published - Jan 2007|
All Science Journal Classification (ASJC) codes
- Mechanics of Materials
- Mechanical Engineering
- Metals and Alloys