基于电弧声信号的MIG焊熔滴过渡类型识别

发布时间：2018-04-16 04:02

本文选题：电弧声 + 复合传感　；参考：《南昌航空大学》2017年硕士论文

【摘要】：在熔化极气体保护焊中,如何在线监测或控制焊接质量是一项最为重要的研究课题。在实际生产中,焊接质量决定着产品的最终质量,而焊丝的熔滴过渡方式不仅决定了弧焊焊丝熔化时的平稳性还严重影响着焊缝的成形、熔深以及材料的消耗、冶金等方面,与焊接质量密切相关。本文针对MIG焊薄板平铺过程中的电弧声信号,着重研究基于电弧声信号识别不同熔滴过渡类型。搭建MIG焊不同熔滴过渡模式下电弧声信号和电信号同步采集与分析系统试验平台,包括机器人焊接系统、复合传感系统、熔滴过渡模式高速摄影系统及系统软件。根据MIG焊方法特点,研制了合适的电信号传感器系统和电弧声传感器系统,能够有效地采集焊接过程中的电弧声信号、电信号以及焊丝熔滴过渡状态的图像信号。对短路过渡模式下的电弧声信号与电信号进行自相关函数分析,并且对电弧声信号与电流信号和电弧声信号与电压信号进行互相关函数分析,结果显示:电弧声与电压、电流信号具有相似的周期性,电弧能量与电弧声密切相关。针对短路过渡、射滴过渡和射流过渡的电弧声信号进行功率谱分析,由功率谱波形图可以发现不同熔滴过渡模式下,电弧声信号的频率分布有明显差异,并且有一定的规律性,即短路过渡过程低频成分较多,射滴过渡和射流过渡高频成分较多。对短路过渡、射滴过渡和射流过渡的电弧声进行小波包分析。小波包分解时小波基函数选择db14,分解层数设置为4。提取电弧声信号小波包4层分解后频带能量特征值。电弧声信号S_(4,0)、S_(4,2)、S_(4,3)频带能量分布百分比差异明显,可作为识别熔滴过渡类型的特征向量。提取不同熔滴过渡状态的电弧声信号峰度系数值,分析发现,短路过渡、射滴过渡和射流过渡的峰度Ku存在差异性,可以作为熔滴过渡类型的特征向量。鉴于此,识别熔滴过渡模式的四维联合特征向量就构造完成了。基于MATLAB软件平台设计了电弧声信号MIG焊熔滴过渡类型模式识别网络模型,网络选择广义回归神经网络和概率神经网络。结果显示:GRNN网络熔滴过渡类型识别率为96.7%,PNN网络熔滴过渡类型识别率为93.3%。通过构造的电弧声四维联合特征向量能够有效识别熔滴过渡类型,识别精确度较高,达到了预期实验目标。
[Abstract]:How to monitor or control welding quality online is one of the most important research topics in gas shielded electrode welding.In actual production, the welding quality determines the final quality of the product, and the droplet transfer mode of the welding wire not only determines the stability of the arc welding wire melting, but also seriously affects the weld formation, penetration depth, material consumption, metallurgy and so on.Closely related to welding quality.In this paper, the arc sound signal in the process of MIG welding sheet tile is studied, and the recognition of different droplet transfer types based on the arc sound signal is studied.A test platform for synchronous acquisition and analysis of arc acoustic signals and electrical signals in different droplet transfer modes of MIG welding was built, including robot welding system, composite sensing system, high-speed photography system and system software.According to the characteristics of MIG welding method, a suitable electric signal sensor system and an arc sound sensor system are developed, which can effectively collect the arc sound signal, the electric signal and the image signal of the welding wire droplet transfer state.The autocorrelation function analysis of arc sound signal and electric signal in short-circuit transition mode is carried out, and the cross-correlation function analysis between arc sound signal and current signal and arc sound signal and voltage signal is carried out. The results show that: arc sound and voltage,The current signal has similar periodicity and the arc energy is closely related to the arc sound.Based on the power spectrum analysis of arc acoustic signals with short-circuit transfer, droplet transfer and jet transfer, it can be found that the frequency distribution of arc acoustic signals is obviously different under different droplet transfer modes and has certain regularity.That is to say, the low frequency components of short circuit transition are more, and the high frequency components of droplet transfer and jet transfer are more.The arc sound of short circuit transfer, droplet transfer and jet transfer is analyzed by wavelet packet.Wavelet packet decomposition when the wavelet basis function selection db14, decomposition layer set to 4.The band energy eigenvalues of the arc acoustic signal are extracted after the decomposition of the wavelet packet 4 layers.The percentage difference of energy distribution in the frequency band of the arc sound signal S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / S / SThe kurtosis coefficients of arc acoustic signals in different droplet transfer states are extracted. It is found that the kurtosis Ku of short-circuit transfer, droplet transfer and jet transition is different, and can be used as the characteristic vector of droplet transfer type.In view of this, the four dimensional joint Eigenvectors for identifying droplet transfer patterns are constructed.Based on the MATLAB software platform, the network model of droplet transfer pattern recognition for MIG welding with arc sound signal is designed, and the generalized regression neural network and probabilistic neural network are selected.The results show that the recognition rate of droplet transfer type of the WGRNN network is 96.7% and the recognition rate of droplet transfer type of PNN network is 93.3%.The arc acoustic four-dimensional joint eigenvector can effectively identify the droplet transfer type, and the recognition accuracy is high, and the expected experimental goal is achieved.
【学位授予单位】：南昌航空大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TG444.74

【参考文献】