【作者】 邓耀华；

【导师】 刘桂雄；

【作者基本信息】 华南理工大学 ， 机械电子工程， 2012， 博士

【摘要】 柔性件是一种常见加工材料，在纺织制造、航天航空制造等行业有广泛应用，但由于其刚性差、加工过程工件容易产生较大的拉伸或挤压变形，这使得加工控制变得比较困难。论文以提高柔性件轨迹加工准确度、快速性和可靠性为目标，研究包括柔性件轨迹加工过程变形决策知识提取、加工轨迹变形补偿非线性建模、加工轨迹在线测量反馈控制以及核心算法的嵌入式开发等理论与方法，这对促进先进制造科学、智能测控科学的发展与应用，具有重要的学术价值和实际意义。研究工作得到广东省教育部产学研结合计划项目(2011B090400468)、广州市科技计划重点项目(2007Z2-D3161)、广东省自然科学基金项目(05001838)资助。论文从分析柔性件轨迹加工控制过程及加工评价指标入手，得到要提高柔性件轨迹加工精度及系统性能，加工过程变形补偿控制、加工轨迹在线测量反馈以及加工控制模型自适应性和智能化是研究的关键技术。从柔性件轨迹加工过程MIMO建模方法、柔性件轨迹加工视觉测量方法、智能控制系统软硬件协同设计与硬件加速方法等三方面讨论与柔性件轨迹加工变形补偿控制技术相关研究领域的国内外研究进展，确定论文的研究内容。论文的主要工作包括：㈠开展柔性件轨迹加工过程变形力学分析研究，指出影响柔性件轨迹加工变形的因素相当复杂，必须进行加工变形决策知识提取。讨论柔性件轨迹加工简化力学模型的有限元仿真及求解，得出柔性件轨迹加工变形与作用力、作用点位置、柔性件材料结构参数等因素之间关系；指出作用力变化又与进给深度、进给偏角、图元类型、图元夹角、加工步长、插补方法、插补速度、加工方向角、柔性件夹紧方式以及柔性件夹紧位置等等因素相关，若将众多的加工变形影响因素作为后续预测模型的输入将会形成极其复杂系统结构，必须对柔性件轨迹加工变形影响因素进行提取。㈡提出基于粗糙集(RS)及信息熵约简方法的柔性件轨迹加工变形决策知识提取方法。以柔性件轨迹加工变形影响因素作为条件属性A，加工轨迹变形程度作为决策属性D构成加工变形决策表DDT;研究基于信息熵表示属性重要度的DDT约简算法，将互信息I (P;D)变化程度作为条件属性对决策属性重要性的评价指标，当I (P;D)变化越大则条件属性a对于决策属性D就越重要，具有较强的理解性、客观性、操作性，制订出变形决策表DDT属性约简算法计算流程图，实现柔性件轨迹加工变形决策知识的提取。应用例表明，本文以信息熵约简方法提取的加工变形决策重要度最高影响因素，与Pawlak约简方法、遗传约简算法提取的影响因素一致，但预测误差分别减少32.58%、21.45%；同时由于信息熵约简方法可通过灵活地阈值C I设置满足不同建模精度要求需要，更容易获得一个准确度高的变形影响因素精简集。㈢提出柔性件轨迹加工变形补偿预测ATS-FNN建模方法，集中了模糊聚类、模糊神经网络建模方法优点。该方法由自适应模糊聚类AFCM、Takagi-Sugeno(T-S)型模糊神经网络(TS-FNN)建模方法有效结合，TS-FNN具有学习能力强、逼近非线性函数映射能力好的特点，TS-FNN前件网络引入模糊聚类方法AFCM完成输入空间模糊等级划分、隶属度函数提取，规则适应度计算，实现TS-FNN模型前件网络结构辨识；TS-FNN后件网络比标准T-S模糊神经网络模型增加了隐含层，进一步提高模型的全局逼近性能。实验结果表明，ATS-FNN模型的建模时间比标准的TS-FNN(STS-FNN)减少52.34%，ATS-FNN模型预测误差MSE较STS-FNN减小36.50%、33.34%；经ATS-FNN模型预补偿后的加工轨迹夹角误差、直线度误差分别比STS-FNN模型预补偿、无补偿加工减少40.44%、52.55%及28.76%、44.45%；ATS-FNN模型的图元最小加工时间比STS-FNN减少46.09%，比无补偿加工增加6.65%。㈣提出一种基于机器视觉测量加工误差反馈的ATS-FNN模型，设计以双32位MicroBlaze处理器为核心、TS-FNN、小波变换等专用IP核为辅助的柔性件轨迹加工变形补偿硬件控制器。通过机器视觉测量加工轨迹几何尺寸，轨迹加工偏差经PID调节后对ATS-FNN模型预补偿值修正，有望解决柔性件轨迹加工精度受工件厚度、进给速度、加工轨迹图案变化而受影响等问题；硬件控制器中双处理器基于消息邮箱Mailbox通信机制的协同工作，加速图像处理任务处理；专用IP核以快速链接总线FSL协处理器接入MicroBlaze处理器的多核数据通信方式，较好解决IP核与主处理器之间总线和内存数据传输滞后限制问题。实验结果表明，带视觉测量反馈环节的引入使得加工误差即使加工条件改变而仅产生较小波动；ATS-FNN多核控制器采用双处理器协同工作方式有助于加速控制器的计算速度。㈤提出加工图像小波变换的FIR滤波器加速分解/重构设计方法。利用8抽头转置FIR滤波器设计Daubechies(4)的分解、重构计算IP核，该IP核的小波两级分解总耗时Tw mrt比PC计算时间Tw mrtpc仅增加5.561%；为了加速TS-FNN计算，引入多级流水线设计思想，将TS-FNN前件、后件网络硬件实现电路的组合逻辑延时路径系统分割，在各个分级之间插人寄存器暂存中间数据，获得更短时序路径，实现TS-FNN前后件网络并行计算，采用流水线设计的IP核性能指标得到明显提高，以8位浮点运算为例，流水线设计的IP核运行频率F_ipcorepl比非流水线设计运行频率F_ipcorenpl提高17.85%。㈥结合柔性件轨迹加工变形补偿技术的应用，介绍带反馈ATS-FNN控制器的绗缝加工系统、基于开环ATS-FNN控制器的电脑弯刀机加工系统研制。根据实际绗缝加工提取加工变形影响因素，基于ATS-FNN控制器设计了绗缝加工系统的硬件结构，开发了花模打版、控制软件。应用结果表明，基于ATS-FNN控制器的绗缝加工系统的加工轨迹夹角误差f、直线度误差f l分别比基于PC+NC控制减少32.9%、36.1%，较好地解决绗缝轨迹加工误差随着柔性件厚度增加而增大问题；基于开环ATS-FNN控制器的电脑弯刀机加工系统的技术参数已经达到送料精度±0.015mm、最大折弯角度130°、最大弯曲半径200mm，这表明加工变形补偿控制基础理论在电脑弯刀机加工系统应用已取得较好应用效果。更多还原

【Abstract】 Flexible workpiece is common processing material in textile manufacturing, aerospacemanufacturing and other industries are widely used, but because of its poor rigidity, in theprocess, the workpiece is easy to have a greater tensile or compressive deformation, whichmakes process control becomes more difficult. To improve the flexible workpiece of the pathprocessing accuracy, speed and reliability targets, the study includes: the decision-makingknowledge extraction of workpiece’s deformation, nonlinear modeling of distortioncompensation for processing path, feedback control of processing path through onlinemeasurement and embedded core algorithm development theory and methods, which promotethe advanced manufacturing science, intelligent measurement and control the developmentand application of science, with important academic value and practical significance. Thework was supported by combination project of Guangdong province and the Educationministry of China (2011B090400468), Guangzhou municipal science and technology keyproject (2007Z2-D3161) and Guangdong provincial natural science foundation (05001838)funding.Based on the analysis and processing of flexible workpiece path processing controlevaluation, to improve the machining accuracy and system performance of flexible workpiecepath processing, distortion compensation control of processing, feedback control ofprocessing path through online measurement, adaptability and intelligent of process controlmodel are the key technologies for studying. And discussing the domestic and internationalresearch of flexible workpiece path processing distortion compensation control technologythrough the following three aspects, which is MIMO modeling of flexible workpiece pathprocessing, flexible workpiece path processing measurements by machine vision, intelligentcontrol systems hardware and software co-design with hardware acceleration, then to todetermine the study content of the paper, the main thesis includes:⑴Carried out the mechanical analysis of flexible workpiece path processing deformation,pointed out the factor of flexible workpiece path processing deformation are very complicated,deformation decision-making knowledge extraction of flexible workpiece path processing isnecessary. After discussing the flexible workpiece path processing simulation of simplifiedmechanical model and finite element solution,gained relationship of deformation of flexibleworkpiece path processing, and force, role of position, structural parameters of flexible piecesand other factors;Pointed out the change of forces is effected by depth of feed, feed angle,primitive type and angle,processing step,interpolation means, interpolation speed, processing direction angle, clamping means and clamping location of flexible workpiece, etc., If a largenumber of factors that affecting the deformation of flexible workpiece path processing aredesignated as input of the following prediction model, it will form an extremely complexsystem architecture, it need to extract the deformation factors for flexible workpiece pathprocessing.⑵Proposed the method of deformation decision knowledge extraction for flexibleworkpiece path processing based on RS and entropy reduction. Regarded factors of flexibleworkpiece path processing deformation as the condition attributes A, level of processingpath deformation as decision attribute D to build decision table of processing deformationwhich expressed as DDT; Research is based on information entropy of DDT attributesignificance reduction algorithm, Changes in the level of mutual information I (P;D)as acondition attribute importance to the decision attribute evaluation, the greater the mutualinformation I (P;D),the condition attribute for decision attribute is more important,it hasstrong understanding, objectivity, operational. To achieve decisions knowledge extraction forflexible workpiece path processing deformation based on DDT reduction algorithm flowchartof deformation decision table. Application examples show that, the highest impact factors ofdeformation for flexible workpiece path processing which extracted through entropyreduction method are same as the result by Pawlak reduction methods and genetic reductionalgorithm, However, the prediction error decreased32.58%,21.45%. Meanwhile, the entropyreduction method can be flexibly configured thresholdC Ito meet the requirements of thedifferent needs of modeling accuracy, easier access to a high accuracy deformationstreamlined set of factors.⑶One distortion-compensated prediction ATS-FNN modeling method for flexibleworkpiece path processing is proposed, it focuses on advantages of adaptive fuzzyclustering AFCM method and fuzzy neural network modeling TS-FNN, the method iseffective integrated by the fuzzy clustering AFCM, fuzzy neural network TS-FNN modelingmethod,TS-FNN with learning ability, nonlinear function approximation and good mappingability, with ffuzzy clustering method AFCM, to gain input space of TS-FNN antecedentnetwork, the membership function extraction, and calculation of rule fitness. TS-FNNconsequent network adds hidden layer relative to the standard TS model, and further improveuniversal approximation properties of the model. Experimental results show that, buildingtime of ATS-FNN model is smaller than the STS-FNN model reduced52.34%, ATS-FNNmodel prediction error MSE than the STS-FNN decreased36.50%,33.34%; The processing path angle errors, straightness errors which is pre-compensated by ATS-FNN model, thanpre-compensation by the STS-FNN models were, processing with non-compensation toreduce40.44%,52.55%and28.76%,44.45%; Pixel minimum processing time ATS-FNNmodel is better than STS-FNN decreased46.09%, compared with6.65%increase innon-compensation processing.⑷Proposed feedback ATS-FNN model which path error is measured by machine vision.To design flexible workpiece path processing distortion compensation hardware controllerwith dual32-bit MicroBlaze processor core, TS-FNN, wavelet transform IP core for thespecial auxiliary. Processing path geometry is measured by the machine vision, the processingtrack error is adjusted by the PID regulator, and then correct the pre-compensation value ofATS-FNN model, so precision is expected to solve which caused by the thickness of theworkpiece, the feed rate, machining trajectory pattern changes and other issues.Dual-processor of the hardware controller based on mechanism of message mailboxcommunication to work together, so it can speed up image processing tasks. Multi-core datacommunication of dedicated IP core links to MicroBlaze processor by FSL bus can solve theproblem of bus and memory limitations of data transfer delay between main processor and IPcore. Experimental results show that, with the introduction of error feedback, making theprocessing even if processing conditions changes the error produce only small fluctuations;ATS-FNN controller with dual-core processors work together to help accelerate thecalculation speed controller.⑸Proposed the design which can accelerate decomposition/reconstruction of wavelettransform for image processing by FIR filter. The decomposition of Daubechies (4) andreconfigurable computing IP core are designed by using the8-tap transpose FIR filter. Thetotal time consuming of wavelet two-level decomposition in this IP coreTw mrtis onlyincreased5.561%compared with PC computing timeTwmrtpc. In order to accelerate theTS-FNN calculation, multi-stage pipeline design is introduced, it achieved the systempartitioning of combinational delay logic circuit between hardware for the TS-FNNantecedent and consequent network, registers are inserted between the various classificationsto store intermediate data temporarily, as well as timing path was shorter, it achieved parallelcomputing between TS-FNN antecedent and consequent network. IP core performanceindicators have been significantly improved using pipelined design, taking8-bit floating-pointoperations as sample, the IP core operating frequencyFipcoreplwith pipeline design is improved17.85%compared toFipcorenplwithout pipeline design.⑹Combined with the deformation compensation application of flexible pieces, itintroduces ATS-FNN controller with feedback process system in quilting and thedevelopment of computer bending machine processing system which based on open-loopATS-FNN controller. According to processing deformation factors in actual quilting process,it developed a pattern mold making and control software module, which based on hardwarestructure of quilting process system designed on ATS-FNN controller. And the applicationresult shows that, the processing path angle errorf and the straightness errorf lofquilting process system based on ATS-FNN controller respectively reduced32.9%and36.1%compared to those system which based on PC+NC, it solved the error increase problem inquilted path processing meanwhile the thickness of the flexible pieces increased. Thetechnical parameters of bending machine processing system which based on open-loopATS-FNN controller have reached the feeding accuracy±0.015mm, the maximum bendingangle of130°and the maximum bending radius of200mm,which shows that bending machineprocessing system using the basic theory of deformation compensation processing control hasachieved good application results.更多还原