激光一电弧复合焊专题机械制造文摘焊接分册激光-MIG复合焊根部驼峰缺陷预测刘秀航，叶广文，黄宇辉，张艳喜，冯桑，高向东（广东工业大学，广东省焊接工程技术研究中心，广东广州510006)摘要：激光-熔化极惰性气体（Meltinertgas,MIG）复合焊过程中容易出现根部驼峰缺陷，为了实现焊接过程根部驼峰缺陷的同步预测，研究根部驼峰缺陷预测的算法并对其预测结果进行分析。采用高速摄像机进行复合焊接过程的实时视觉传感采集，提取焊接过程的正面熔池和匙孔的时序特征信息，并对这些特征信号进行小波包分解（Waveletpacketdecompo-sition，WPD）与重构。应用激光扫描仪获得背部焊缝余高，以此作为驼峰状态标记的依据。再通过长短期记忆（Longshort-termmemory,LSTM)神经网络对焊接过程中根部驼峰状态进行预测。结果表明，WPD-LSTM算法对根部驼峰预测的准确率达到97.85%。相比其它算法，基于焊接过程正面视觉传感时序特征信息的WPD-LSTM算法预测准确率更高，且预测结果具有较高的连续性，有利于实现对焊接过程根部驼峰缺陷的同步检测与控制。关键词：根部驼峰；视觉检测；长短期记忆神经网络；小波包分解中图分类号：TG409Roothumpdefectpredictionforlaser-MIGhybridweldingLiuXiuhang,YeGuangwen,HuangYuhui,ZhangYanxi,FengSang,GaoXiangdong(GuangdongProvincialWeldingEngineeringTechnologyResearchCenter,GuangdongUniversityofTechnology,Guangzhou510006,Guangdong,China)Abstract:Theroothumpdefectiseasytoappearinthelaser-MIGcompositeweldingprocess.Inordertoreal-izethesimultaneouspredictionoftheroothumpdefectintheweldingprocess,thispaperstudiesthealgorithmofroothumpdefectpredictionandanalyzesthepredictionresultsofdifferentalgorithm.Thereal-timevisualsensinginformationofcompositeweldingprocessiscarriedoutbyahigh-speedcamera,thetimeseriescharac-teristicinformationofthefrontweldpoolandthekeyholeintheweldingprocessisextracted,andthecharacter-isticssignalsaredecomposedandreconstructedbywaveletpacketdecomposition(WPD).Then,theresidualheightofthebackweldisobtainedbyalaserscanner,whichisusedasthebasisformarkingthehumpstatus.Longshort-termmemory(LSTM)neuralnetworkwasusedtopredictthestatusofroothumpintheweldingprocess.ExperimentalresultsshowthattheaccuracyofWPD-LSTMalgorithmfor...
