基于等离子体视觉和光谱特征的AH36钢激光焊接熔深熔宽预测研究  

Research on Predicting the Depth and Width of Laser Welding of AH36 Steel Based on Plasma Vision and Spectral Features

作　　者：程晋宜 田雷[1] 沈楠楠[1] 檀财旺 CHEN Jinyi;TIAN Lei;SHEN Nannan;TAN Caiwang(Offshore Oil Engineering(Qingdao)Co.,Ltd.,Qingdao 266500,China;Shandong Provincial Laboratory of Special Welding Technology,Harbin Institute of Technology at Weihai,Weihai 264209,China)

机构地区：[1]海洋石油工程(青岛)有限公司,山东青岛266500 [2]哈尔滨工业大学(威海)山东省特种焊接技术重点实验室,山东威海264209

出　　处：《电焊机》2024年第5期31-38,45,共9页Electric Welding Machine

摘　　要：针对AH36船用钢不同激光功率和焊接速度的激光焊接,通过高速摄像提取焊接过程等离子体的图像,同时采集了光谱信息,最终建立GA-BP(Genetic Algorithm-Back propagation)神经网络以预测焊缝的质量即熔深和熔宽,并与常用的回归分析方法和未经优化的BP神经网络预测效果相比较。结果表明,焊缝熔深和熔宽均随激光功率的增大而增大,随焊速提高而降低;等离子体面积、高度、灰度以及电子密度均随激光功率增大呈增大趋势。当采用仅考虑焊接工艺参数的回归模型进行预测时,熔深预测相对误差18.2%,熔宽预测相对误差12.1%。而采用引入焊接过程等离子体特征的BP神经网络,虽然精度提高,但由于随机性存在导致相对误差变化较大,进一步采用GA算法进行全局求解能力优化后,熔深预测相对误差降低至7.23%,熔宽预测相对误差降低至5.79%,模型稳定性也明显改善。综合考虑工艺参数和等离子体特征建立的神经网络模型能够更准确预测AH36钢激光焊接焊缝形貌,为焊接工艺规划和质量预测提供参考。For laser welding of AH36 steel with different laser powers and welding speeds,high-speed cameras were used to extract plasma images during the welding process,meanwhile spectral information was collected.Finally,a GA-BP(Genetic Algorithm-Back propagation)neural network was established to predict the quality of the weld,namely the depth and width of fusion,and the prediction results were compared with commonly used regression analysis methods.The results showed that the depth and width of the weld seam increase with the increase of laser power,and decrease with the increase of weld‐ing speed;The plasma area,height,grayscale,and electron density all show an increasing trend with increasing laser power.When using a regression model for prediction,the relative error of melt depth prediction is 18.2%,and the relative error of melt width prediction is 12.1%.The use of an un-optimized BP neural network results in significant changes in relative error due to randomness.After optimizing using the GA algorithm and introducing process parameters as input features,the rela‐tive error of melt depth prediction is reduced to 7.23%,and the relative error of melt width prediction is reduced to 5.79%.

关 键 词：AH36钢 激光焊接 等离子体 回归分析 神经网络 

分 类 号：TG456.7[金属学及工艺—焊接]