基于CNN-MOGWO的钢/铝异种金属激光焊焊接工艺参数多目标优化  

Multi-objective optimization of laser welding parameters for steel/aluminum based on CNN-MOGWO

作　　者：张鑫然 郭永环[1] 范希营[1] 张岚峰 闫勃 ZHANG Xinran;GUO Yonghuan;FAN Xiying;ZHANG Lanfeng;YAN Bo(Jiangsu Normal University,Xuzhou 221116,Jiangsu,China)

机构地区：[1]江苏师范大学,江苏徐州221116

出　　处：《焊接》2025年第2期17-23,共7页Welding & Joining

基　　金：国家自然科学基金资助项目(51905238);江苏师范大学科研创新计划资助项目(2022XKT0373)。

摘　　要：【目的】在汽车轻量化的发展趋势下,钢/铝混合结构在汽车中应用越来越多,但钢与铝的物理性能差距极大导致难以熔合,所以需要合理地选择焊接工艺参数以提高焊接接头性能。【方法】基于正交试验设计以及Visual environment模拟所得结果,建立了以激光功率、焊接速度、光斑直径为输入,残余应力和z轴方向的母材变形量为输出的卷积神经网络(Convolutionalneuralnetwork,CNN)回归模型。将建立的2个回归模型作为目标函数,使用多目标灰狼优化算法(Multi-objective greywolfoptimization,MOGWO)探寻最优工艺参数。【结果】CNN回归模型预测值与仿真值的误差分别为1.76%和3.30%,均在4%以下;MOGWO寻得激光功率为706.61 W,焊接速度为15.00 mm/s,光斑直径为0.91 mm时,取得最优结果。相较于优化前残余应力降低了7.80%,z轴方向变形量减少了24.17%。【结论】CNN模型能够精确描述焊接工艺参数与性能间的非线性关系,且其与MOGWO算法的结合能够寻得最优的工艺参数,提高钢/铝异种金属激光焊焊接接头性能。[Objective]In the strategic context of automotive lightweighting,steel/aluminum hybrid structures are increasingly being employed in vehicles,yet the significant differences in physical properties between steel and aluminum make it extremely challenging to fuse them seamlessly.Therefore,it is crucial to judiciously select welding process parameters to enhance the performance of welded joints.[Methods]A CNN regression model is eatablished based on orthogonal experimental design and simulation results obtained using Visual environment.The model takes laser power,welding speed,and laser spot diameter as inputs,with residual stress and the amount of deformation in the z-direction of the base material as outputs.Two such regression models are then utilized as objective functions within a MOGWO algorithm to explore the optimal welding parameters.[Results]The errors between the predicted values from the CNN regression model and the simulated values are 1.76%and 3.30%,both falling below 4%.The MOGWO algorithm yields the optimal result when the laser power is 706.61 W,the welding speed is 15.00 mm/s,and the spot diameter is 0.91 mm.Compared to before optimization,the residual stress is reduced by 7.80%,and the deformation in the z-direction decreases by 24.17%.[Conclusion]The CNN model can accurately describe the nonlinear relationship between welding parameters and performance,and its integration with the MOGWO algorithm enables the identification of optimal welding parameters,thereby enhancing the performance of steel/aluminum dissimilar metal laser welded joints.

关 键 词：卷积神经网络 多目标灰狼优化算法 激光焊 异种金属焊接 工艺参数优化 

分 类 号：TG456.7[金属学及工艺—焊接]