【作者】 刘鹏飞；

【导师】 单平；

【作者基本信息】 天津大学 ， 材料加工工程， 2008， 博士

【摘要】 电阻点焊作为一种常用的焊接方法,由于其高生产效率和易于实现自动化的特点,而在汽车、航空、航天产品的生产中得到了广泛的应用,尤其在现今汽车行业的高速发展中,点焊的应用更加广泛。随着世界能源日益紧缺,汽车轻量化理念使得具有优良性能的轻金属-铝合金成为了汽车生产的主要材料,但由于铝合金点焊焊接性较差及影响点焊质量的因素较多的缘故,使得对铝合金点焊质量的研究日益成为焊接工作者研究的重点。目前,点焊质量检测存在两方面的不足,在特征提取上,点焊特征提取主要集中在宏观特征和奇异点特征上,而由于点焊过程的影响因素较多,使得点焊特征信号受噪声的干扰而发生了部分特征变异,而传统的提取方法不能很好的把握这种变化。在检测方法上,目前,点焊质量检测模型主要集中在建立数学函数模型上,但由于点焊的缺陷受到多种因素的影响,其与点焊信号特征之间的函数关系不够严格,导致目前的数学检测模型的检测精度较低。鉴于这种情况,目前已经有部分学者开始把模式识别的理论应用到点焊质量检测上,但还需要进行深入研究。因此本文针对目前点焊质量检测中存在的这些不足,本文做了以下方面的研究工作:(1)构建多信息融合的点焊信息采集系统,获得焊点的4个同步信号。构建了多信息融合的智能信息采集系统,采集了电极电压、焊接电流、焊接过程中的声发射和电极位移四个参量数据。该多信息融合的智能信息采集系统可以迅速快捷的完成点焊数据的采集、转换和存储,各智能采集终端通过数字处理模块实现与计算机的通信。基于多信息融合的点焊信息采集系统可以达到信号采集的快速、同步和高效。(2)应用信号分析理论和非线性动力学理论提取信号的特征值。由于在铝合金点焊过程中,各类信号的变化比较复杂且变化幅度较小,传统的特征提取方法就显得不够准确了,因此,本文结合信号分析理论和非线性动力学理论,考虑到信号分形维数较强的抗噪能力,及该指标对数据信号变化特征的总体反映能力,对点焊电极电压、点焊电流、点焊过程中的声发射信号和电极位移信号进行分形维数特征提取;根据电压对点焊形核的能量输入特征,对电压信号进行了双谱分析,提取了其三阶累积量的线谱最大5个谱线的均值作为电压的第二特征;通过对点焊电流的非线性分析发现,点焊电流信号具有混沌特征,而混沌特征能够有效反映信号的微弱变化,数据序列的混沌特征用Lyapunov指数来表征量,鉴于电流信号的变化对熔核尺寸和飞溅产生的影响极大,因此提取电流的Lyapunov指数作为电流的第二特征。点焊过程中的形核和飞溅都会造成声音信号能量的变化,因此,提取声音的能量变化率作为声音信号的第二特征。点焊的形核和飞溅直接影响了电极位移的变化规律,故提取了表征点焊电极位移变化不确定程度的位移信号的Renyi熵作为电极位移信号的第二特征。对每个焊点共获得了对应的8个特征值。(3)应用人工智能理论,模式识别理论及应用数学方面的知识,把模糊灰色信息系统、支持向量机和隐形Markov链引入点焊的质量检测中,并讨论了各模型的检测效果及本文提取的各数据特征对模型检出准确率的影响情况。首次把模糊灰色信息系统引入点焊质量检测领域,根据模糊灰色关联度的概念,构建了点焊的模糊灰色信息系统检测模型,并比较了各信号特征对模型检出准确率的影响,模糊灰色信息系统对点焊缺陷具有较好的的检测效果,当8个特征输入时达到最大检出准确率,熔核尺寸检出率为91%,飞溅状况检出率为92%;构建了点焊信号的支持向量机检测模型,并分析了个特征对模型检测精度的影响,发现支持向量机检测模型在5个或6个特征输入时,达到最佳检测效果,熔核尺寸检出率为86%。飞溅检测效果在5特征输入时达到最佳检出率为88%。首次构建了点焊的隐性Markov链检测模型,并分析了各特征组合对该模型的影响,发现在7特征输入时熔核的检出率达到最佳,为93%,而飞溅的最佳检出率也是在7特征输入时获得的为87%。(4)为达到同时检测焊点尺寸和飞溅两种缺陷的目的。构建熔核尺寸检测和飞溅检测模型阵列。本文首次构建了模糊灰色信息系统检测模型阵列,达到同时检测熔核尺寸和飞溅的目的,本文构建了8特征输入的点焊尺寸灰色信息系统检测模型和6特征输入的点焊飞溅灰色信息系统检测模型阵列,其最佳检测效果较好,检出率达到89%,支持向量机阵列模型的最佳检出率为80%,隐性Markov链检测模型的最佳检出率为81%。比较发现,模糊灰色信息系统检测模型为最佳点焊缺陷检测模型,其他两种方法可以作为辅助检查方法。更多还原

【Abstract】 Because of the characteristic of high productivity and automatic realization, resistance spot welding, as a common welding method, has obtained widespread application in the production of automobile and aviation industry. It has more application especially in the rapid development of automobile industry. The concept of automobile lightweight makes the aluminum alloy, which is one of the light alloy with excellent performance, to be the leading material in the automobile production due to the lack of energy in the world. The study of welding quality in resistance spot welding of aluminum alloy becomes a important study direction as a result of the poor welding performance of aluminum alloy and the complicated influential factors of spot welding quality. At present, it is insufficient in the two aspect of research on quality of spot welding, one is feature extraction, now the emphasis of feature extraction in spot welding is put on the extraction of the macroscopic feature and singularity, due to the process of spot welding is influenced by a large number of factors, the characteristic signal of welding spot is interfered by noise and changed. But the common method can not extract the change of feature of signal precisely. The other deficiency is existed in detection method, now the quality detection model of welding spot is emphasized on the model construction based on relation of feature between signal and nugget quality, but the deficiency of spot welding is influenced by a large number of factors, there is not a one to one correspondence relation between the deficiency of nugget and spot welding signals, as a result the above reason lead to the low percentage of accuracy of detection of detection model of spot welding, though the theory of pattern recognition has been introduced to quality detection of spot welding by some researcher, the application of this method should be researched in profound direction. The below work is done aim to the deficiency existed in present detection method of spot welding:(1) The data acquisition system based on multi-information fusion is constructed, and four sync signals is obtained.The data acquisition is used in dissertation to collect the four variables: electrode voltage, welding current, welding sound and electrode displacement. This system based on multi-information fusion can complete independently data collection, A/D transfer and data local save. Each intelligent terminal communicates with computer by digital management module. The signals can be collected by the data acquisition system based on multi-information fusion quickly efficiently and synchronously. (2) The eigenvalue of signals is extracted using theory of signal analysis and nonlinear dynamics.Because the change of signals is complex and the amplitude narrow in the process of spot welding, the common extraction method will be inaccuracy, based on the theory of signal and nonlinear dynamics, considering the better noise immunity and express ability to change feature of signal as a whole, the fractal dimension from the signals of electrode voltage, welding current, welding sound and electrode displacement are extracted in the paper; according to the input feature of energy of voltage to nucleation of spot welding, the high-order spectrum is used to analysis the voltage signal, and the five max spectral line of line spectrum of three-order cumulant slice of voltage signal. Through analyzing to non-linear feature of welding current, the chaotic characteristic is discovered, and the chaotic characteristic can reflect the weak change of signal, Lyapunov exponent is the characterization value of chaotic characteristics of signal, because of the great effect of current signal to nugget size and spatter, the Lyapunov exponent of current signal is extracted as the second feature. In the process of spot welding, nucleation and spatter can emit sound, so the change rate of energy of sound is extracted as the second feature of sound. The change trace of electrode displacement can be greatly influenced by nucleation and spatter, the Renyi entropy, which can attribute the uncertainty of electrode displacement, is extracted as the second feature of electrode displacement signal. as a result, each spot weld can correspond to 8 features.(3) The fuzzy grey information system, the support vector machine and the hidden Markov chain are introduced to the quality detection of resistance spot welding by using the artificial intelligence theory, pattern recognition theory and application mathematics knowledge. The detection effect of each model and the influence of each data feature which is extracted in this paper on the percent of accuracy of detection are discussed.The fuzzy grey information system is introduced to the field of quality detection of resistance spot welding for the first time. According to the concept of fuzzy-gray relational degree, the fuzzy grey information system model of resistance spot welding is constructed, and effects of every signal characteristic on the percent of accuracy of detection of the model is discussed. The fuzzy grey information system has a better detection effect on defects of resistance spot welding. When the eight eignvalues are inputted, the maximum percent of accuracy of detection is obtained, the percent of accuracy of detection of nugget size is 91% and the percent of accuracy of detection of spatter is 92%. The support vector machine model of resistance spot welding signal is constructed and the effect of each feature on the detection accuracy is analyzed. The optimal detection result of nugget size is 86% when five or six eignvalues are inputted in the support vector machine model. The optimal detection result of spatter is 88% which can be achieved when five eignvalues are inputted. The hidden Markov chain model of resistance spot welding is constructed for the first time and effects of every feature combination on this model is analyzed. The optimal detection result of nugget size is 93% and the optimal detection result of spatter is 87%, both of which can be achieved when seven eignvalues are inputted.(4) In order to inspect nugget size and spatter simultaneously, the model array of nugget size detection and spatter detection is constructed.An detection model array is constructed which is composed of an eight eignvalues fuzzy grey system nugget size detection model and a six eignvalues fuzzy grey system spatter detection model, and the optimal detection effect can be increased to 89%. The optimal percent of accuracy of detection of the support vector machine model array is 80% and the optimal percent of accuracy of detection of the hidden Markov chain model is 81%. Through comparison, it has been found that the fuzzy grey information system inspection model is the best resistance spot welding defects detection model, another two methods can be used as assistant detection methods.更多还原