导　　师： 薛家祥

学科专业： 0804

授予学位： 博士

作　　者： ;

机构地区： 华南理工大学

摘　　要： 铝合金作为代替传统钢铁材料的新型轻质材料在制造业中所占比重越来越大，而焊接则是其工业生产过程中一个必不可少的环节。同时为了提高经济效益和产品质量，厚度为3.0mm以下高强板材越来越多地被采用，尤其是汽车行业。这些薄板的应用，使薄板加工焊接技术成为研究的热点。铝合金薄板在焊接时对电流信号的输入控制敏感性较高，如输入电流不合适将极易引起焊接过程不稳定，焊缝缺陷等问题。所以，在焊接过程中对电流的精确控制是其焊接技术研究的重点。本文针对如何提高铝合金薄板焊接质量及其智能化进行了相关研究。 论文以铝合金薄板脉冲MIG焊接电源为研究对象，对其控制系统的关键技术进行了深入探讨。在基于DSP+ARM双核控制的焊接电源软硬件设计基础上，基于Simulink建立了系统模型，并进行了仿真分析，对加速设计过程，提高电源的性能及后续的研究奠定了基础。 针对焊接过程中的干扰问题，进行了数字滤波研究。首先给出了基于小波软阈值和小波包的滤波器设计思路，并给出了相应的试验结果，试验表明小波滤波能有效地滤除电流信号中的干扰成分；其次研究了一种加权卡尔曼滤波器的抗干扰方法，通过遗传算法对BP神经网络的初始权值和阈值进行全局粗精度预学习，并应用神经网络在线实时整定自适应卡尔曼滤波器的加权因子，保证滤波效果并给出相应的仿真试验，为焊接电源数字化滤波器的设计提供依据。 为了保证焊接电流输出调节的精细，弥补焊接电源离散化分段调节的缺陷，满足对焊接电流的精确调节，提高铝合金薄板脉冲MIG焊电源的智能化，研究基于最小二乘法的参数一元化调节和基于大步距标定的参数自学习算法，通过智能化算法自动生成匹配的焊接参数，达到“连续调节”目的，并利用局部牛顿插值算法，实现参数的自学习，让焊接电源专家数据库的属性由“只读”变成灵活的“自学习”。 针对弧长在外界干扰下会发生变化的问题，实现对弧长的实时控制，保证铝合金薄板MIG焊过程的稳定性和焊接质量。建立电弧弧长数学模型，分析铝焊丝电弧的调节机理与特性，并提出变频弧长控制方法调节电弧。 考虑到铝合金薄板焊接过程控制的复杂性和送丝与焊接工艺参数匹配的困难，研究了送丝速度的智能预测控制策略，建立基于多元回归分析法和基于神经网络法的送丝速度预测模型并进行仿真，结果表明：神经网络对送丝速度的预测效果更好，为提高焊接电源智能水平提供依据。 为了保证在焊接过程中对电流进行更为精细、准确的控制，设计基于模糊PID参数自整定控制算法和一种改进的变论域模糊PID参数自整定控制算法，实现PID参数的在线自整定，在铝合金薄板脉冲MIG焊过程控制中均取得了比传统PID更好的效果。 从电信号的角度进行了焊接性能定量评定研究。利用概率密度分布函数从统计学的角度进行焊接质量稳定性定量评价，研究基于U-I图工作点的周期重复率评判焊接过程稳定性的方法。同时，为了避免焊缝质量评定的主观随意性，利用模糊数学对焊缝质量进行模糊综合定量评价。并利用多信息融合的思想，融合各指标建立神经网络综合定量评定模型，实现对焊接性能的定量评定。 在自行搭建的焊接试验平台上对自主研发的铝合金薄板脉冲MIG焊接电源进行性能测试和单、双脉冲MIG焊工艺性能试验，验证了各章节提出的论点。 Aluminum alloys as light materials have been becoming more and more popular in themanufacture industry and taking place of traditional steel materials, besides, welding is anindispensable procedure in industrial production process. In order to promote benefits andimprove quality of products synchronously, high strength aluminum plates with thicknessbelow3.0mm are commonly used currently, especially in the automotive industry. Theapplication of these sheets, making the sheet-processing and corresponding weldingtechnology become hot spots of research. During welding process, sheets made of aluminumalloys are highly sensitive to the controlling of current signal input, if inappropriate currentinput is loaded, it will probably result in welding instability, weld defects or other problems.Therefore, the precise control of current during welding process is the key point of its weldingtechnology research. This paper is aiming at investigating how to improve the quality andproperty of welding joint of thin aluminum alloy plate and corresponding intelligencetechnology. Specially, research on pulsed MIG welding power source and its key technologies ofcontrol system is our target. Based on software and hardware design of DSP+ARM dual-corecontrolled welding power source, system model was established based on Simulink andsimulation analysis was conducted, which lay the foundation for improvement of powersupply performance and subsequent research. Regarding interference during welding, digital filter study is needed. Firstly,experimental results are deduced from the design of wavelet soft threshold and wavelet packetfilter, which illustrates that wavelet filtering is productive when applied to filtering outinterference within the current signal. Secondly, one of anti-interference methods of weightedKalman filter was explored, and the global rough precision study /(via genetic algorithm/) onBP neural network's initial weights and threshold was also conducted. Besides above, neuralnetwork was used to do on-line real-time setting of self-adjusting Kalman filter’s weightedfactor. In this way, filtering effects are guaranteed and corresponding simulation tests are given, which provide guidance to the design of digital welding power source filter. In order to ensure the accuracy of welding current output regulation, thus make up fordeficiency introduced from multilevel control of discretized welding power source, and torealize the accurate regulation of welding current, then further promote intelligent level ofaluminum alloy sheet pulse MIG welding power source, we studied the parameters unifiedadjustment which based on least squares and the parameters self-learning algorithm whichbased on long-step calibration, to achieve the aim of “continuous adjustment” via generatingmatching welding parameters automatically through intelligent algorithm. Additionally,changing the 'read only' property of welding power expert database to flexible 'self learning'by utilizing local Newton interpolation algorithm to realize parameters’ self-learning. The arc length changes under external interference. For the sake of real-time controllingarc length and then ensure the stability of aluminium alloy plate MIG welding and weldingquality, mathematical model of electric arc length was built up to analyze the adjustmentmechanism and characteristics of weld wire arc. Besides, we proposed that arc length controlmethod can be used to adjusting electric arc. It is complex to control aluminum alloy thin plate welding process and match theparameters of wire feeding and welding process, so developing a strategy of predicting andcontrolling wire feeding speed intelligently is necessary, a wire feeding speed predictionmodel based on multidimensional regression analysis and neural network was built up andused to do simulation. The results showed that neural network is more effective when it comesto prediction of wire feeding speed, which provides some reference for improvement ofintelligent level of welding power source. To ensure a more elaborate and accurate control of current during welding process, wedesigned a control algorithm based on Fuzzy self-tuning PID parameters and anotherimproved control algorithm that based on variable universe Fuzzy self-tuning PID parameters,therewith achieved on-line self-tuning of PID parameters, which contributes to obtainingbetter effects on control of aluminum alloy plate pulsed MIG welding process comparing withtraditional PID methods. From the electrical signal perspective, quantitative evaluation of welding performancewas studied as well. Using probability density distribution function statistically, toquantificationally assess the stability of welding quality, and develop methods of evaluatingthe stability of welding process based on periodic repetition rate of working points in U-Ifigure. Meanwhile, in order to avoid the subjective randomness of assessment on weldsquality, we utilized fuzzy mathematics to evaluate welds quality which adopts the FuzzyComprehensive Evaluation Model. Moreover, quantitative evaluation of welds quality alsoneeds fusion of multi-information to set up a comprehensive and quantitative assessmentmodel for evaluating welding performance based on neural network. Performance tests, single and double pulsed MIG welding process performance tests onwelding power source /(self-developed/) of aluminum alloy plate pulsed MIG welding wereconducted on welding tests platform that is self-designed, in addition, the contentions andarguments proposed in every chapter were verified.

关 键 词： 铝合金薄板 脉冲 焊 神经网络 智能控制策略 定量评定 多信息融合

分 类 号： [TG409]

领　　域： [金属学及工艺]