基于机械臂氩弧焊工艺参数预测与优化研究

作者：西安工业大学机电工程学院闫渭丘，田军委，苏宇；内蒙古北方重工业集团有限公司刘雪松，张杰

摘要：针对机械臂氩弧焊中焊接参数与焊缝性能存在的高维非线性耦合问题，设计了4因素（焊接速度、气体流速、焊接电流、摆动幅度）5水平的正交试验，以熔宽和熔深作为关键焊缝性能指标，构建了系统的试验数据集，并通过正交试验对焊接参数进行优化。基于此，对比分析了多种数据驱动预测模型的性能，结果表明：所提出的BP神经网络与信念规则库（BP-BRB）混合模型在测试集上表现出最优的预测精度，其预测结果的均方误差（MSE）总和由传统BP模型的1.534947显著降至1.170058，精度提升约23.78%，有效提升了对复杂焊接过程的建模能力。为进行对比，将3种工艺下的焊缝分别标记为：试样1（PSO-GA融合算法优化）、试样2（正交试验优化）及试样3（优化前基准工艺）。结果表明：融合算法优化后的焊接效果最好，试样1相较于试样2，焊接效率提升了1.5%，试样3的焊缝宽深比提高了61.74%，焊接效果提升显著。

关键词：焊接参数；遗传算法；神经网络；粒子群算法

Research on prediction and optimization of process parameters for robotic arc welding

Abstract: In response to the high-dimensional nonlinear coupling issue between welding parameters and weldperformance in robotic argon arc welding, a systematic experimental dataset was constructed through a 4-factor (welding speed, gas flow rate, welding current, oscillation amplitude) 5-level orthogonal experiment, with melt width and melt depth as key weld performance indicators. The welding parameters were optimized based on the orthogonal experiment. Subsequently, the performance of various data-driven prediction models was compared and analyzed. The results indicated that the proposed hybrid model combining BP neural network and belief rule base (BP-BRB) demonstrated the best prediction accuracy on the test set. The total mean square error (MSE) of its predictions significantly decreased from 1.534947 for the traditional BP model to 1.170058, representing an accuracy improvement of approximately 23.78%, effectively enhancing the modeling capability for complex welding processes. For comparison, welds from three different processes were labeled as follows: Sample 1 (optimized by PSO-GA fusion algorithm), Sample 2 (optimized by orthogonal experiment), and Sample 3 (baseline process before optimization). The results showed that the welding effect after fusion algorithm optimization was the best. Compared to Sample 2, Sample 1 achieved a 1.5% improvement in welding efficiency, while Sample 3 exhibited a 61.74% increase in weld width-to-depth ratio, indicating significant enhancement in welding performance.

Keywords: welding process parameters; genetic algorithm; neural network(s); particle swarm optimization