基于信号特征提取的电阻点焊质量在线评判

以焊点接头强度作为焊点质量评判的指标,通过对点焊过程焊接电流、动态电阻、电极位移信号的同步采集和特征分析,提取若干特征参量监测点焊过程,依据特征参量与焊点接头抗剪强度间的相关分析结果,选取来自不同监测信号的7个特征参量建立了表征点焊过程的特征模式,并将此转化为计算机可以识别的模式矩阵,同时以焊接电流参数为模式分类的依据,建立不同模式矩阵类别和焊点接头抗剪强度之间的映射,将模式矩阵作为Hopfield神经网络的记忆样本存储于网络,利用网络联想记忆的功能实现对未知样本点焊过程的模式识别,进而实现点焊质量的评判。网络测试结果表明,利用Hopfield网络进行焊点质量在线评判可以得到满意的效果。     

基于贝叶斯图像模式识别技术的点焊质量评估

提出一种将点焊过程动态电极位移信号转化为二值图像的方法.基于图像特征分析,从焊点样本电极位移二值图像中提取出15个隐含特征.针对一系列对应不同焊接质量焊点样本电极位移二值图像特征,利用主成分分析消除图像特征间的互相关性,建立了基于最小风险贝叶斯图像识别技术的焊点质量分类器.分类器有效性测试结果表明,电极位移信号二值图像尽可能多的保留了焊点质量信息,特征提取算法简单、高效、易于实现;同时在小数据样本情况下,贝叶斯图像识别技术能够快速、准确地评判焊点质量,有较好的应用前景.     

基于贝叶斯图像模式识别技术的点焊质量评估

提出一种将点焊过程动态电极位移信号转化为二值图像的方法.基于图像特征分析,从焊点样本电极位移二值图像中提取出15个隐含特征.针对一系列对应不同焊接质量焊点样本电极位移二值图像特征,利用主成分分析消除图像特征间的互相关性,建立了基于最小风险贝叶斯图像识别技术的焊点质量分类器.分类器有效性测试结果表明,电极位移信号二值图像尽可能多的保留了焊点质量信息,特征提取算法简单、高效、易于实现;同时在小数据样本情况下,贝叶斯图像识别技术能够快速、准确地评判焊点质量,有较好的应用前景.     

基于支持向量机(SVM)的点焊质量在线监测

电阻点焊过程动态信号蕴含大量直接或间接反映焊点质量的动态信息，通过对焊接过程电极位移、动态电阻信号的同步采集和分析，从两种信号中提取了12个特征参量，建立表征点焊过程的数据集，以焊点抗剪强度作为焊点质量评价指标，利用支持向量机（SVM）统计学习方法，将焊点试样动态监测参量与焊点抗剪强度之间低维的复杂非线性映射关系，映射到一个高维的特征空间（Hilbert空间），原试样数据空间的非线性关系相应变化为高维特征空间的线性关系，在不增加计算复杂度的情况下，实现对未知焊点试样抗剪强度的分类及预测。SVM测试结果表明，支持向量机在小样本情况下具有较好的泛化能力，分类、预测速度快，准确率高，能较为满意地完成焊点强度的分类、预测任务，可以作为进一步研究的方法。     

基于支持向量机的点焊质量分类

以点焊质量监控为目的,通过对点焊过程电极位移信号、动态电阻信号的同步采集和信号分析,从2种信号中提取12个有效的特征分量,构造点焊过程的监测数据集,以焊点的抗剪强度为指标,利用支持向量机(SVM)和RBF神经网络对焊点质量进行分类.测试结果表明,2种信号特征向量作为SVM的输入向量,在参数匹配的情况下,使用不同的核函数对喷溅和焊点质量分类均能取得较好的分类结果,分类器性能明显优于RBF神经网络模型.小样本情况下SVM具有较好的泛化能力,正确率仍保持在80%以上,据此认为提取电极位移和动态电阻的特征向量构筑输入向量,建立的SVM动态模型对点焊接头进行质量分类是可行的.     

基于PCA-SVM方法的点焊质量评估

通过对电阻点焊过程电极位移和动态电阻信号的实时采集和时域特征分析,利用电阻信号动态特征刻画熔核形成不同阶段,从同步电极位移信号中提取9个特征参量建立输入样本数据集.以焊点接头抗剪强度作为焊点质量的评价指标,将PCA(主成分分析)方法与传统的SVM(支持向量机)回归分析相结合,利用PCA方法对支持向量机的输入样本数据集进行主成分分析,消除了输入特征参量间的自相关性,实现数据降维后作为支持向量机的输入,建立焊点质量映射模型.交叉有效性预测结果表明,基于PCA-SVM的算法增强了SVM的泛化能力,比传统的SVM算法具有更高的预测精度.     

基于LabVIEW的直流点焊数据采集系统

笔者研制的用PC机控制的电阻点焊机,其数据采集系统不仅采样速度快、存储数据容量大、系统可靠性高,且实现了点焊参数的可视化、焊点质量的在线评估及历史焊点数据的存储.利用LabVIEW语言开生了直流点焊数据采集系统,对焊接电流、电极压力及电极位移信号进行实时采集,为有效地评估焊点质量提供了可靠的保证.     

基于信息融合技术的焊点质量评估

基于点焊焊接过程电极位移、动态电阻信号的同步采集和特征分析,从2种信号中提取若干特征参量,依据特征参量与焊点接头抗剪切力间的相关性分析结果,选取特征参量建立数据集,利用多元线性、非线性、支持向量机统计分析方法实现多信息融合,构建焊接过程监测参量与焊点强度之间的回归映射模型.进而实现对未知焊点样本强度的预测.交叉有效性检验结果表明以相关性显著的特征参量建立的多元线性回归、非线性回归、支持向量机回归预测模型,对于评估焊点质量是有效的,其中支持向量机回归预测有效性最为显著,可作为进一步研究和实现在线质量监测的方法.     

波动工况下点焊质量在线预测及模型解释

基于电阻点焊过程的多传感信号特征,面向多种板材组合建立焊点质量在线预测模型,研究了异常工况波动对四类机器学习回归模型的影响,分析了不同模型和输入变量对含异常工况试验数据集的适应性,并采用Shapley值、t-SNE等方法对波动工况下的模型性能进行解释.结果表明,高斯过程回归模型和电阻 + 力信号具有最佳的熔核直径预测性能,焊接电流、热输入能量和电极位移峰值特征对于波动工况具有良好普适性.此外,异常工况引起的信号特征分布差异会显著影响回归预测模型的泛化性能,应尽量减少训练集与数据集差异以提高焊点质量预测的准确性.     

铝合金点焊位移信号时频域分析与质量判定

采用分布式多传感器同步采集系统实现了铝合金点焊质量实时监测.通过位移信号时频域分析发现,喷溅焊点位移信号在40～80 Hz 范围内,Butterworth带通滤波极差达 0.5 mm 左右;未熔合或未完全熔合焊点的位移信号极差比另外两类焊点低 1.0 mm 左右.利用上述明显特征信息可以实现点焊质量人工判定.由于位移信号信噪比高,特征提取算法简洁,进一步可实现机器在线判读.同时,利用统计分析,样本197个,在合格区内焊点正确判断率达97.6%.     

用Lyapunov指数研究点焊位移信号的混沌特性

文中以混沌理论为基础,阐述了Lyapunov指数作为点焊位移时间序列的混沌判断原理,并对电阻点焊位移信号中进行了分析与研究,计算了4种不同工作状态下点焊位移信号时间序列的Lyapunov指数,研究表明,点焊位移信号中的最大Lyapunov指数均大于0,根据混沌理论可以断定点焊位移信号中存在混沌现象,应该采用混沌的方法研究与分析点焊位移信号,该研究为点焊质量的判断和预测开辟了新的途径.     

基于回归分析的钛合金微电阻点焊焊接工艺优化

为了研究焊接工艺参数对钛合金微电阻点焊焊接质量的影响规律以及获得最优的焊接工艺参数,利用中心组合试验设计对0.4 mm厚的TC2钛合金进行焊接试验,对焊接试样进行准静态拉伸试验,获取了表征焊点质量优劣的熔核直径、拉伸力、拉伸位移和失效能量. 利用灰熵法得到了每个焊接质量指标的权值,将其加权为焊接综合质量指标. 采用逐步回归分析方法得到了焊接工艺参数(焊接电流、焊接时间以及电极压力)与焊接综合质量指标之间的回归方程. 在此基础上分析了焊接工艺参数两两交互作用对焊接质量的影响规律,采用爬山优化算法得到了最优的焊接工艺参数.     

Application of the wavelet packet and its energy spectrum to identify nugget splash during the aluminum alloys spot welding

Nugget splash during aluminum alloys spot welding has a detrimental effect on weld nugget integrity, strength and durability of the welded joints. This investigation is performed to identify nugget splash from voltage signals became these are easily accessible on-line. In the present work, we propose a novel method based on the wavelet packet transform and its energy spectrum for pattern recognition of splash signal. The result demonstrates that this novel method is more accuracy and a useful way of monitoring the spot welding quality.     

Chaos analysis of acoustic emission signals in spot welding process

Bused on chaos time series and fiactal theory, acoustic emission signals were studied in the process of spot welding. According to calculating 8 welding parameters using phase space reconstruction method, the largest Lyapunov exponents were positive values and chaos characteristics were firstly discovered from acoustic emission signals in spot welding. In order to evaluate acoustic emission signal, Hausdorff dimension is put forward to analyze and estimate chaos characteristics. The experiment and calculation results indicate that the Hausdorff dimension of acoustic emission signal is significantly distinguishable in the nuggets with different welding parameters. This research provides a new method for measuring the resistance spot welding quality.     

Non-destructive methodology for parameterization of Burn-Zinc in the resistance spot welding process

Galvanized steels are used in large quantities, mainly in the automotive industry, for their corrosion resistance and low cost. In this industrial sector, these materials are normally welded by the spot welding process. Because galvanized steels have a zinc-based covering layer, welding is normally carried out using a technique called the Burn-Zinc technique, to minimize the harmful effects of the zinc on this material, during the welding cycle. In this context, this work studied a non-destructive method, with the aim of finding suitable removal of this layer of zinc, on the basis of the signals generated by a sensor that records the electrode displacement during the phases of the spot welding process, taking the reading for the thermal expansion of the galvanized steel sheet as a basis. Thus, physical simulation experiments were undertaken, varying the current and the cycle time of the preheating phase. The Burn-Zinc responses generated by the sensor and the visual analyses by attribute were shown as graphs via an operating envelope. The evidence showed that the signal from electrode displacement proved to be an efficient non-destructive method for determining suitable Burn-Zinc parameters in the spot welding process for galvanized steel sheet.     

基于粗糙集的点焊多信息融合的质量判定

研究了基于粗糙集理论的在线铝合金点焊质量判定的新方法,选取焊接过程中电极位移、电极压力两个参数的八个特征量构成知识表示系统,采用自组织特征映射网络(SOM网络)离散连续属性值、基于差别矩阵的属性约简算法对获取的信息进行处理,提取判别规则,进而通过判别规则来完成焊点质量的分类.该方法不仅降低特征空间的维数、减少质量分类的工作量、降低了信息存储量,而且点焊质量判断准确率可以达到97.06%.

Abstract：

The new method of quality judgment about on-line aluminum alloy spot welding was studied, which was based on the rough set theory. The eight characters selected from two parameters (electrode displacement and electrode force in the weld process) constituted the knowledge representation system. The new method dealt with the obtained information by adopting the discretized continuous attribute algorithm with the self-organizing feature map net-work(SOM) and the attribute reduction algorithm based on discem-ibility matrix, picked up the distinguish rule, then completed the classification of spot welding quality by the rule. The method can not only reduce the dimensions of the feature space, the workload of quality classification and the information memory capacity, but also can make the accuracy of the spot weld quality judgment reach 97. 03%.     

Quality assessment for resistance spot welding based on binary image of electrode displacement signal and probabilistic neural network

To improve the efficiency of acquiring monitored features and present a reliable quality assessment method for resistance spot welding, a novel method for converting the electrode displacement signal into binary image is proposed. The probabilistic neural network is adopted to provide a probabilistic viewpoint and a deterministic classification result of weld quality when the image characteristic of displacement signal is selected as monitored features. Test results of the classifier demonstrate that it is feasible and reliable to utilise binary image of the displacement signal to evaluate weld quality. The method reserves weld quality information as much as possible and it avoids complex algorithm for extracting and selecting monitored features. At the same time, when there are small samples, the classifier can identify good or bad weld rapidly and accurately even though the weld is from abnormal welding process, such as expulsion, current shunting and small edge distance condition.