吕小青^{1, 2}，王 旭¹，徐连勇^{1, 2}，荆洪阳^{1, 2}
AuthorsHTML:吕小青^{1, 2}，王 旭¹，徐连勇^{1, 2}，荆洪阳^{1, 2}
AuthorsListE:Lü Xiaoqing ^{1, 2}，Wang Xu ¹，Xu Lianyong ^{1, 2}，Jing Hongyang ^{1, 2}
AuthorsHTMLE:Lü Xiaoqing ^{1, 2}，Wang Xu ¹，Xu Lianyong ^{1, 2}，Jing Hongyang ^{1, 2}
Unit:1. 天津大学材料科学与工程学院，天津 300350；
2. 天津市现代连接技术重点实验室，天津 300350
Unit_EngLish:1. School of Materials Science and Engineering，Tianjin University，Tianjin 300350，China；
2. Tianjin Key Laboratory of Advanced Joining Technology，Tianjin 300350，China
Abstract_Chinese:以焊缝高宽比和深宽比作为优化目标，结合径向基函数神经网络和带精英策略的非支配排序的多目标遗传算法NSGA-Ⅱ，实现了多目标优化．建立了以焊接电压、送丝速度、焊接速度作为自变量，预测焊缝熔宽、余高和熔深的5种模型，即误差反向传播神经网络、遗传算法优化的误差反向传播神经网络、克里金插值法、径向基函数神经网络和二阶多项式回归模型．对比分析表明，径向基函数神经网络具有较高的预测精度和稳定性，最为合适．最后，利用NSGA-Ⅱ算法实现了以盖面焊和填充焊为应用场景的工艺参数多目标优化，试验证明了该优化方法的有效性．
Abstract_English:This paper used the combination of radial-based function neural network(RBFNN)and multi-objective genetic algorithm(NSGA-Ⅱ)to realize the multi-objective optimization of the weld reinforcement-width ratio and the penetration-width ratio. With welding voltage，wire feeding speed，and welding speed as independent variables，five models—error backpropagation neural network(BPNN)，BPNN optimized by genetic algorithm，Kriging method，second-order polynomial regression model，and RBFNN—were developed to predict the geometry of welding beads(penetration depth，weld bead width，and weld reinforcement). Comparative analysis shows that RBFNN was selected as the most suitable model due to its higher prediction accuracy and stability. Finally，NSGA-Ⅱ was used to achieve multi-objective optimization for welding filling and cosmetic welding. The verification experiment proved the availability of the multi-objective optimization strategy.
Keyword_Chinese:焊接工艺参数；焊缝形貌；多目标优化；神经网络；多目标遗传算法
Keywords_English:welding process parameter；geometry of welding bead；multi-objective optimization；neural network；multi-objective genetic algorithm

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6524