【作者】 王凤琴；

【导师】 赵军；

【作者基本信息】 燕山大学 ， 材料加工工程， 2003， 博士

【摘要】 板材成形是金属塑性加工领域的一大分支，在汽车、航空、仪表等工业领域占有重要的地位，其技术水平在某种程度上反映一个国家或地区的工业现代化水平。板材成形自动化由于不具备实时监测、识别、和预测的能力，只能按照预先设定好的加工程序和工艺参数完成成形过程。当被加工对象的材质及工况条件有变化或波动时，不能对工艺参数自动地进行相应的调整。板材成形智能化，由实时监测、实时识别、实时预测和实时控制四个基本要素构成，通过控制科学、计算机科学与板材成形理论的有机结合，根据被加工对象的特点，利用易于监测的物理量，实时确定材料性能参数及最优的工艺参数，并自动以最优工艺参数完成板材成形过程。所以板材成形智能化过程是冲压成形过程自动化及柔性加工系统等新技术的更高级阶段，不但可以改变冲压生产工艺的面貌，而且还将促进冲压设备的变革，同时也会引起板材成形理论的进步与分析精度的提高，在降低板材级别，消除模具与设备调整的技术难度，缩短调模试模时间，提高成品率和生产率等方面都具有十分明显的意义。在工程实践中，工件多为以盒形件为典型代表的非轴对称复杂形状。因此，研究盒形件拉深过程智能化控制技术很有学术价值和实际意义。 本文在圆锥形零件拉深成形智能化控制的研究成果基础上，分析了盒形件拉深智能化需要解决的关键技术，并对其中的过程监测和识别模型等方面的相关问题展开了系统研究。 在盒形件拉深智能化控制所要求的四个基本要素中，参数识别模型和最优工艺参数预测模型的建立都是基于对盒形件成形规律的认识程度基础上的。通过假设盒形件圆角区剪应力零线的变形性质同相应的轴对称件相同，根据盒形件变形特点，推导出盒形拉深件法兰区和悬空侧壁区的理论解析，为智能拉深中的参数识别和破裂、起皱预测等问题的研究提供了理论依据。 利用理论解析和有限元仿真模拟等研究手段，分析了对盒形件拉深成形有影响的主要因素，从而确定盒形件智能拉深过程中的参数识别模型。网络拓扑结构选择前向神经网络结构，Levenberg-Marquarat算法作为网络优化算法，并利用Matlab语言进行编程计算。对于样本数据的采集问题，研究了数值模拟替代部分实验获取样本的可行性。此外，研究了样本数据和隐层节点数目对网络模型效率、精度和泛化能力的影响规律，并提出定压边和阶段识别方案作为盒形件智能拉深参数识别策略，使得多种材料样本数据达到了1‰网络平方和误差。对于泛化结果的处理问题，采用去除奇异点和平均输出的混合方法提高泛化识别精度。 实时监测和实时控制两个基本要素主要受信号采集发展水平限制，本文分析了燕山大学工学博士学位论文智能拉深实验系统中原基于控制软件Genie的信号采集系统存在的不足，采用美国M公司的虚拟仪器控制软件LabviEW、6062E数据采集卡及相关模块，建立了便携式数据采集系统。在该系统上开发了信号采集和传感器标定等程序，获得了令人满意的信号监测与控制的速度和精度。利用控制软件LabvIEw提供的Matlab scriPt节点、开发了信号控制程序和参数识别模型接口程序，解决了智能拉深过程中的参数实时识别问题。更多还原

【Abstract】 As a branch of plastic working field, sheet metal forming possesses the important place in the industry field of automobile, aviation, instrument and so on. Its state of the art reflects industrial modern level of a country or region in some degree. Due to lack of the skill in real-time monitoring, identifying and predicting, the automation of sheet metal forming can only finish forming process in the light of pre-established work program and process parameters. When material quality and operating condition of manufactured object change or fluctuate, process parameters can’t be automatically regulated. The intellectualization of sheet metal forming, which includes 4 basic elements, real-time monitoring, identification, prediction and control, is the crossing subject of control-science and sheet metal forming theory. According to the characteristics of the initial piece, utilizing physical quantities easy to be measured, material properties and friction coefficient can be determined in real-time, and then the forming process can be completed automatically with the optimal processing parameters. As a result, the intellectualization of sheet metal forming process is the higher level of new technologies such as press forming automation and flexible process system, by which not only can the feature of manufacturing technique be changed, but also the transformation of press equipment can be forwarded. It brings about the progress of sheet metal forming theory and improvement of analysis precision at the same time, it has the important significance for degrading sheet metal level, eliminating technology difficulty between die and equipment adjustment, shortening die setting time, improving productivity and the rate of finished products, and so on. Most workpieces are non-axial symmetry complex shape represented by rectangular box in engineering practice, so it has much academic value and significance to study intelligent control technology of deep drawing for a rectangular box.Based on the research achievements of intelligent deep drawing for an axis-symmetric part, the key technologies of intelligent deep drawing for a rectangular box are discussed, and the relevant issues to the process monitoring and identification model have been studied in this paper.Among the four basic elements required by intelligent deep drawing for a rectangular box, the establishment of the identification model of parameters and the prediction model of optimal technological parameter is dependent upon the level of understanding the forming law for a rectangular box. With the deformation characteristics of rectangular boxin deep drawing, the theoretic analysis of stress on deformation area has been obtained on the basis of the hypothesis that deformation law of shear stress zero-line is the same as the corresponding axial symmetric workpieces, which provides the theoretic basis for the identification of parameters and the prediction of fracture and wrinkling.The dominant factors influenced on deep drawing forming of a rectangular box have been analyzed by using theoretic analysis and finite element method (FEM) simulation, and then the identification model of parameters for a rectangular box is determined. The Levenberg-Marquarat is chosen as optimal algorithm of neural network whose topology structure is feedforward network, and Matlab language is to program. For the problem of sample data acquisition, the feasibility has been studied that FEM replaces some experiments to get sample data. Furthermore, the influence of sample data and hidden nodes on network convergence efficiency, precision as well as generalization has been studied. The strategy of stage identification with constant blank holder force is put forward as identification scheme for a rectangular box, by which the sum-squared error is stepped downward to l‰ The hybrid method of removing singular points and mean technique is employed to improve generalization precision.Real-time monitoring and control depend on the development of data acquisition (DAQ) technol更多还原