摘要
为了提高熔化极惰性气体保护电弧焊的焊缝成形预测精度和预测稳定性,提出了优化深度置信网络的预测方法。经分析选择了对焊接成形有影响的4个参数作为预测输入,分别为焊接速度、焊接电流、焊接电压、焊丝干伸长;选择可以反映焊缝形状的熔深、熔宽、余高作为预测输出。使用深度置信网络构造焊缝成形预测网络模型,鉴于误差反向传播的参数训练方法容易陷入局部极值,这里提出了分阶段扰动粒子群算法对DBM参数进行优化,从而提高算法预测精度和预测稳定性。经试验验证,优化深度置信网络对焊缝形状参数的预测精度比传统深度置信网络提高了一个数量级,同时预测稳定性也高于深度置信网络,证明了这里算法对焊缝成形预测的有效性。
To improve weld formation prediction accuracy and stability of Metal Inert Gas Arc Welding,prediction method based on optimized Deep Belief Network is proposed.Through analysis,four parameters influencing weld are selected as predicting input,including welding speed,current,voltage and wire extension.The parameters of weld penetration,weld width,weld height reflecting weld shape are selected as predicting output.Weld predicting network model is constructed by deep belief network.Considering that parameters training method of error backward propagation is easy to fall into local optimal,staged disturbed particle swarm algorithm is put forward to optimize the network parameters to improve its predicting accuracy and stability.It is clarified that weld shape parameters prediction accuracy of optimized deep belief network is one order of magnitude higher than traditional deep belief network,and prediction stability is also higher than deep belief network,which proves weld formation prediction validity of the method in this essay predicting accuracy and stability of by improved particle swarm algorithm optimizing deep belief network are highest,which can prove validity of the essay method on weld formation prediction.
作者
张峰
阴伟锋
田坤
张建霞
ZHANG Feng;YIN Wei-feng;TIAN Kun;ZHANG Jian-xia(He'nan Institute of Technology,Department of Intelligent Engineering,He'nan Xinxiang 453003,China;Beijing Institute of Technology,School of Optics and Photonics,Beijing 100081,China)
出处
《机械设计与制造》
北大核心
2022年第11期174-178,共5页
Machinery Design & Manufacture
基金
河南省科技厅科技攻关项目(172102210123、192102110198)
新乡市科技攻关项目(CXGG17012)。
关键词
焊缝成形预测
深度置信网络
改进粒子群算法优化
预测稳定性
Weld Formation Prediction
Deep Belief Network
Improved Particle Swarm Algorithm
Predicting Stability
