基于ELM的焊缝成型建模研究  

Welding Shape Modeling Based on ELM

作　　者：叶建雄[1] 李志刚 程群 周金兰 郭波 彭星玲[1] YE Jianxiong;LI Zhigang;CHENG Qun;ZHOU Jinlan;GUO Bo;PENG Xingling(Jiangxi Province Key Laboratory of Precision Drive&Control,Nanchang Institute of Technology,Nanchang 330099,China;School of Mechanics&Vehicle Engineering,East China Jiaotong University,Nanchang 330013,China;Leping No.1 Middle School of Jiangxi Province,Jingdezhen 333300,China)

机构地区：[1]南昌工程学院,江西省精密驱动与控制重点实验室,江西南昌330099 [2]华东交通大学机电与车辆工程学院,江西南昌330013 [3]江西省乐平市第一中学,江西景德镇333300

出　　处：《热加工工艺》2020年第17期142-145,共4页Hot Working Technology

基　　金：江西省教育厅科技项目(GJJ180937);国家自然科学基金项目(51665016);江西省精密驱动与控制重点实验室开放基金(PLPDC-KFKT-201625)。

摘　　要：焊缝形状对焊接质量有着重要的影响,但由于焊接是一个剧烈的、充满着干扰的不稳定过程,因而构建精确的焊缝成型模型比较困难。研究中首先将极限学习机(ELM, extreme learning machine)引入到建模中,ELM可以通过计算方式得到其输出矩阵,因而建模速度很快;其次引入M-估计以抑制焊接过程中粗大误差的影响,设计了整合二者的M-ELM算法,最后利用实际TIG焊接数据进行建模,并与多元非线性回归(multi-nonlinear regression,MNR),线性回归(linear regression,LR),BP神经网络所建的模型进行了性能比较。结果表明,M-ELM不但建模速度快,而且模型拟合精度好,达到了预期目标。The welding shape is closely related to welding quality,but it is difficult to construct accurate weld shape model because welding process is unsteady and full of intense interference.Firstly,ELM was introduced into modeling.The output matrix could be obtained by calculating of ELM,so the modeling speed was very fast.Secondly,M-estimation was also introduced to reduce the influence of rough error,and the hybrid algorithm of M-ELM was designed.At last,TIG welding data were used to construct the model,and the model was compared with the models constructed by multiple nonlinear regression(MNR),linear regression(LR)and BP neural network.The results show that the modeling speed of M-ELM is fast,and the fitting accuracy is good,which achieves the expected goal.

关 键 词：焊缝建模 极限学习机 M-估计 

分 类 号：TG409[金属学及工艺—焊接]