Residual stresses and distortion in welded joints undermine the durability of the structure and prevent a correct assembly of the parts. The principal objective of this study is to find a solution to minimize the residual stresses and distortion induced by submerged arc welding process. Accordingly, first, a thermal simulation of the process was undertaken by the finite-element method, and the results were used to provide a mechanical solution. The mechanical solution determined residual stresses and distortion that were found to be consistent with experimental results. Next, drawing on the design of experiment method based on cooling time between first pass and second pass and the first and second pass welding speed, a set of training data was formed for the developed artificial neural network. The trained neural network was then used as input for the optimization algorithm. Single- and multi-objective Genetic Algorithm and single and multi-objective Harmony Search methods were used for optimization process. Results illustrate that artificial neural network and multi-objective optimization algorithms are excellent methods for optimizing the residual stresses and distortion caused by welding process. As it was proved in this study, the single-objective optimization of the welding process is effective in reducing both the residual stress and distortion. The double-objective optimization also contributed to reduce both residual stress and distortion with 4% (for residual stresses) and 26.56% (for distortion) in multi-objective Harmony Search which was the better algorithm based on the solution time. Given the contradiction of the residual stresses and distortion in the welding process, the double-objective algorithm was found to be less successful in minimizing the two target functions relative to the case with the two optimized separately.

中文翻译：

---

基于遗传算法和和谐搜索的埋弧焊过程残余应力和变形的多目标优化

焊接接头中的残余应力和变形会破坏结构的耐用性并妨碍零件的正确组装。本研究的主要目标是找到一种解决方案，以最大限度地减少埋弧焊工艺引起的残余应力和变形。因此，首先，通过有限元方法对该过程进行了热模拟，并将结果用于提供机械解决方案。机械解决方案确定了与实验结果一致的残余应力和变形。接下来，借鉴基于第一道次和第二道次冷却时间和第一道次和第二道次焊接速度的实验方法的设计，为开发的人工神经网络形成了一组训练数据。然后将经过训练的神经网络用作优化算法的输入。单目标和多目标遗传算法和单目标和多目标和谐搜索方法用于优化过程。结果表明，人工神经网络和多目标优化算法是优化焊接过程引起的残余应力和变形的极好方法。正如本研究所证明的那样，焊接工艺的单目标优化在减少残余应力和变形方面是有效的。双目标优化还有助于在多目标和谐搜索中将残余应力和变形降低 4%（残余应力）和 26.56%（变形），这是基于求解时间的更好算法。