盒形件智能化拉深过程中材料性能参数的识别

智能化拉深过程中材料性能参数实时识别是关键技术之一。盒形件拉深难以用精确的力学模型来描述,文章引入基于LM算法的神经网络模型对材料参数进行识别,并仔细研究了在盒形件拉深过程中的适用性。针对盒形件提出了拉深初期采用恒定压边识别的方案,并采用平均值和去除奇异数据的方法大幅度地减小了识别误差,在该文的样本数据范围内,4种材料性能参数的最大识别误差在2%以内,为实现整个拉深成形过程的智能化控制奠定了基础。     

轴对称曲面件拉深智能化控制技术研究

板材冲压智能化是一项涉及控制科学、计算机科学和板材塑性成形理论等领域的综合性新技术。该技术的研究已有 10几年的历史 ,但主要是集中在V型弯曲的回弹控制方面 ,直到 90年代初才开始对筒形件拉深的智能化控制进行探索性研究。本文研究建立了轴对称壳体零件拉深过程智能化控制系统 ,并以锥形件为例 ,分析了轴对称曲面零件拉深过程中的共同特点 ,建立了完整的力学模型 ,实现了轴对称曲面件拉深的智能化控制     

DDCAD系统成形工艺分析模块（一）—任意形状轴对称拉深件的变形分析

介绍了ＤＤＣＡＤ工艺设计子系统成形工艺分析模块中的拉深件的应变分析，为拉深工艺设计提供了重要的依据。     

轴对称曲面件智能化拉深成形过程的解析定量描述

成形过程的定量描述是板材成形智能化控制中在线识别材料参数和工况参数以及预测最佳工艺参数的理论依据,识别和预测精度取决于定量描述的准确程度。为了实现轴对称曲面件拉深过程的智能化控制,分析了轴对称曲面拉深件的共性特征,建立了完整的力学模型,在直线假设、面积不变假设和似直梁弯曲假设条件下,给出了拉深过程中拉深力－行程曲线的解析定量描述。用三种板材以锥形件拉深为例进行了实验验证。     

锥形件智能化拉深系统中材料参数和摩擦系数的在线识别

以拉深过程的解析模拟为基础 ,采用非线性最小二乘曲线拟合原理实现了锥形件拉深智能化控制中材料参数和摩擦系数的在线识别。进而给出了材料参数已知时 ,摩擦系数识别非线性问题的一种线性化算法。实验验证结果表明 ,上述识别方法是正确的     

帽形件智能化弯曲参数实时识别及实时预测的研究

在板材成形智能化控制过程中,材料参数实时识别和最优工艺参数的实时预测是最重要的两个要素,实时识别时间的长短与实时预测精度的高低,直接影响到板材成形智能化控制的成败。文章利用神经网络技术实现了帽形件弯曲成形智能化控制过程中材料参数识别与最优工艺参数的预测,其材料参数及工艺参数均在数值模拟和试验提供的数据范围内,识别和预测模型的收敛精度均达到了0.1%,识别和泛化精度较高,满足帽形件弯曲智能化控制的要求。     

圆筒形及矩形件的拉深模CAD系统

拉深件的生产遍布于军用、民用的各个生产部门,其用量之大也是不言而喻的。因而世界上工业发达国家的大企业集中了较多的人力、物力,从事拉深模CAD的开发。到七十年代中期,国外许多公司已经成功地将计算机技术应用到汽车复盖件生产领域。如美国的通用汽车公司、日本的三菱汽车公司、英国的剑桥大学工业部,美国的康乃尔大学、苏联的国家科学院综合技术研究所等。匈牙利米什科尔茨重工业大学机械系所开发的拉深模CAD/CAM系统,可用于处理一些轴对称件(包括带凸缘件、台阶形工件等)和矩形件。国内,     

基于UG的汽车覆盖件拉深筋参数化设计

应用UG的二次开发技术,对拉深模常用形状结构的拉深筋进行分析研究,从其设计特点、过程出发,提出了从建模到编辑修改的参数化设计方案,开发了拉深筋参数化设计软件模块,实现了拉深筋的参数化、智能化设计,提高了拉深筋设计的质量和效率。     

汽车覆盖件双动拉深模参数化设计中的概念设计

针对目前国内模具CAD结构设计效率低下而专用模具CAD软件适用面过窄的情况，以模具零部件间的尺寸链表为核心，用变参的方法开发汽车覆盖件双动拉深模具参数化设计系统，并完成了其中的概念设计部分，不仅降低了常规模具CAD系统的开发难度，而且极大地提高了模具设计效率。     

筒形件无压边圈拉深工艺参数优化设计

通过对筒形件极限拉深系数的影响因素进行分析，确定合适的凸凹模圆角半径是提高拉深件质量的重要途径，将塑性理论和优化算法相结合，以筒形件达到最小的极限拉深系数为优化目标，对筒形件无压边圈拉深工艺参数优化设计进行了研究。并且通过具体算例证明了所采用的优化模型及算法是有效的。     

板材拉深成形智能化控制过程中摩擦系数的识别

提出在板材拉深成形过程中确定摩擦系数的一种新方法。在阐述解析法描述摩擦系数的基础上,利用人工神经网络来实现对摩擦系数的识别,以便根据摩擦系数的波动,随时调整控制参数,以最佳的工艺参数来完成板材拉深成形的智能化控制过程。     

Numerical simulation of various cross sectional workpieces using conventional deep drawing and hydroforming technologies

This study focuses on the determination of optimum sheet metal forming process and process parameters for various cross sectional workpieces by comparing the numerical results of high-pressure sheet metal forming, hydro-mechanical deep drawing (DD) and conventional DD simulations. Within the range of each cross section, depth, characteristic dimensions ratio and fillet radius have been altered systematically. Steel of types St14 and DC04 have been used as the specimen material in the numerical analyses and the experimental verification throughout the study. All numerical simulations have been carried out by using a dynamic–explicit commercial finite element code and an elasto-plastic material model. During the analyses each workpiece was simulated by the three competing processes. The results of analyses, such as sheet thickness distribution, necking, forming of radii etc., are used for assessing the success of each forming process alternative. The analyses revealed that depending on the workpiece geometry and dimensional properties certain processes are preferable for obtaining more satisfactory products. Working windows for each process have been established based on the analyzed parameters of the circular, elliptic, rectangular and square cross sectional product geometries. This data is expected to be useful for selecting the appropriate production process for a given workpiece geometry and understand the limits of each sheet metal forming processes.     

基于LM算法的神经网络系统辨识

介绍了电流变传动系统，并采用基于Levenberg-Marquardt(LM)算法的BP神经网络对其进行系统辨识，LM算法是梯度下降法与高斯-牛顿法的结合，就训练次数与精确度而言，它明显优于共轭梯度法及变学习率的BP算法，适用于系统辨识，仿真结果表明LM算法可大大在提高学习速度，缩短训练时间，且辨识效果很好。     

A comparative estimation of the forming load in the deep drawing process

The deep drawing process is one of the important sheet-metal forming processes. Using this operation, many parts are manufactured in various industries. In this paper, different methods of analysis such as analytical, numerical and experimental techniques are employed to estimate the required drawing force for a typical component. With this regard, the numerical simulations were conducted using the finite-element (FE) method. In these simulations, the effects of the element type on the forming load and the variation of the thickness strain were studied. Moreover, the influences of the friction coefficient on the load–displacement curve of the process and maximum drawing force were quantitatively investigated for both the analytical and FE methods. A die set including a blankholder was designed to carry out the experiments on a 600 kN Instron testing machine. Different analytical relationships suggested by different researchers were also used to calculate the maximum drawing force. The results obtained from these methods together with the numerical results were compared with the experimental findings. Based on this comparison, it was concluded that Siebel’s formula predicts more accurate results, compared with other analytical relationships. It was also found that this formula is more sensitive to the friction coefficient than the finite-element simulations. On the other hand, the shell elements are more suitable than four-node solid elements for the numerical analyses because the relevant FE predictions present much better agreement with the experimental results.     

Modular system to measure and control the force distribution in deep drawing processes to ensure part quality and process reliability

Force distribution is one of the most important variables in deep drawing. Together with the tribological conditions, it determines the quality of the formed component. This paper presents a novel mechatronic system for measuring and controlling the normal force distribution in deep drawing. The concept is based on an arrangement of a force measuring platform between the upper die and the press ram. Furthermore, its modular design with multiple sensors allows it to be applied to any tool shape. Using the location of the resulting total force enables novel approaches in the process evaluation. Experimental investigations with the measuring system in a controlled process show the potential of new methods to increase process reliability.     

筒形件拉深压边力区间的理论确定

本文建立了筒形件拉深的力学模型,运用能量法推导出筒形件拉深过程中抑制起皱的最小压边力与拉深高度的关系,通过压边力诱发的拉深力和材料的承受极限,推导出抑制拉裂的最大压边力与拉深高度的关系。通过案例,给出了筒形件拉深的合适压边力区间。     

自组织竞争神经网络板材拉深材料参数识别

阐述了自组织竞争人工神经网络模型优化算法的特点，建立了自组织竞争人工神经网络识别模型，给出板材拉深材料性能参数输入输出层设计和样本数据的采集与处理的方法。研究了自组织竞争人工神经网络模型仿真实例，得到了较满意的识别结果，开辟了应用自组织竞争人工神经网络模型优化板材材料性能参数识别诊断研究的新领域。     

基于遗传算法和神经网络技术的板料拉深成形参数优化

结合数值模拟和人工神经网络技术．建立了板料拉深成形的加工参数（压边力和冲压速度）与其成形质量之间的映射关系，既保证了精度，又减少了数值模拟次数。在神经网络建模的基础上，利用遗传算法对板料拉深成形的加工参数进行了优化，通过实例可以看出，该方法具有较好的优化结果。     

A modified deep neural network enables identification of foliage under complex background

ABSTRACT

For the sake of enhancing the identification ability of current network and meeting the needs of the high accuracy of distinguishing similar small objects (foliage) in the complex scenes, this paper proposes a modified region-based fully convolutional network which adopts Inception V3 accompanying with residual connection as the main framework. Incorporating deep residual learning module into Inception V3 can not only save the computational cost by factorising convolutions, but also mitigate the vanishing gradients causing the increasing depth of the network. Additionally, this combination can alleviate the degradation problem in the process of extracting features and providing proposals. Experimental results show that the modified approach can identify out different leaves with similar characteristics in one scene, and demonstrate the superiority of our proposed approach over some state-of-the-art deep neural networks, when it comes to recognise foliage in complicated environments.