人工神经网络在CO2气体保护焊实时监测中的应用

Application of ANN on real-time monitor in CO2 welding

利用人工神经网络的训练结果可以对焊接过程的各类特征参数进行合理评价。利用LVQ神经网络模型,根据实验采集分析得到的CO2气体保护焊不同焊接工艺条件下焊接电流和焊接电压的概率密度分布曲线以及短路过渡时间和燃弧时间等的时间频数分布曲线,在Matlab神经网络工具箱中开发出焊接过程神经网络识别器,可以自动识别焊接过程中各种干扰因素。识别实验结果表明,利用LVQ神经网络构造的干扰因素识别器识别成功率达到92.5%,识别率高。实验验证了该网络识别器的可行性,可以用于焊接质量的实时监测。

The results trained by artifical neural network could be used to evaluate feature parameters in welding.Based on the probability density distributions（PDD） of welding voltage and current,and the class frequency distributions（CFD） of short-ciruiting time and burning-arc time,this paper developed a welding ANN identifier by ANN toolbox in MATLAB by the LVQ neural network model.The welding ANN identifier could recognize and classify the process disturbances during welding.The experimental results showed that general recognition rate of the LVQ identifier was high in identifying the welding process disturbances,wich reached 92.5 percents.The results indicated the feasibility of the LVQ identifier,wich could be used for the welding quality monitor.