2.845

2023影响因子

(CJCR)

  • 中文核心
  • EI
  • 中国科技核心
  • Scopus
  • CSCD
  • 英国科学文摘

留言板

尊敬的读者、作者、审稿人, 关于本刊的投稿、审稿、编辑和出版的任何问题, 您可以本页添加留言。我们将尽快给您答复。谢谢您的支持!

姓名
邮箱
手机号码
标题
留言内容
验证码

机器人化复合材料自动铺层技术综述

郭鹏 杨辰光 李祥利 章艺 李淼

王向阳, 牛盼盼, 杨红颖, 李丽. 基于多相关HMT模型的DT CWT域数字水印算法. 自动化学报, 2021, 47(12): 2857−2869 doi: 10.16383/j.aas.c190075
引用本文: 郭鹏, 杨辰光, 李祥利, 章艺, 李淼. 机器人化复合材料自动铺层技术综述. 自动化学报, 2024, 50(5): 873−897 doi: 10.16383/j.aas.c230149
Wang Xiang-Yang, Niu Pan-Pan, Yang Hong-Ying, Li Li. A blind watermark decoder in DT CWT domain using Weibull distribution-based vector HMT model. Acta Automatica Sinica, 2021, 47(12): 2857−2869 doi: 10.16383/j.aas.c190075
Citation: Guo Peng, Yang Chen-Guang, Li Xiang-Li, Zhang Yi, Li Miao. A review on robotized automated lay-up technology for composite material manufacturing. Acta Automatica Sinica, 2024, 50(5): 873−897 doi: 10.16383/j.aas.c230149

机器人化复合材料自动铺层技术综述

doi: 10.16383/j.aas.c230149
详细信息
    作者简介:

    郭鹏:华南理工大学自动化科学与工程学院硕士研究生. 2019年获华南理工大学学士学位. 主要研究方向为机器人视觉感知与控制. E-mail: auto_guopeng@163.com

    杨辰光:华南理工大学自动化科学与工程学院教授. 2010年获新加坡国立大学博士学位. 主要研究方向为人机交互, 智慧系统设计. 本文通信作者. E-mail: cyang@ieee.org

    李祥利:武汉大学工业科学研究院博士研究生. 2022年获北京工业大学硕士学位. 主要研究方向为机器人化复合材料加工技术. E-mail: lixiangli00@163.com

    章艺:华南理工大学自动化科学与工程学院硕士研究生. 2022年获上海电力大学学士学位. 主要研究方向为人机技能传递, 机械臂自适应控制. E-mail: izyzhangyi@163.com

    李淼:武汉大学工业科学研究院副教授. 2016年获瑞士洛桑联邦理工学院博士学位. 主要研究方向为机器人学习, 多机器人协同. E-mail: miao.li@whu.edu.cn

A Review on Robotized Automated Lay-up Technology for Composite Material Manufacturing

More Information
    Author Bio:

    GUO Peng Master student at the School of Automation Science and Engineering, South China University of Technology. He received his bachelor degree from South China University of Technology in 2019. His research interest covers robot vision perception and control

    YANG Chen-Guang Professor at the School of Automation Science and Engineering, South China University of Technology. He received his Ph.D. degree from National University of Singapore in 2010. His research interest covers human robot interaction and intelligent system design. Corresponding author of this paper

    LI Xiang-Li Ph.D. candidate at the Institute of Technological Sciences, Wuhan University. He received his master degree from Beijing University of Technology in 2022. His main research interest is robotized composite material processing technology

    ZHANG Yi Master student at the School of Automation Science and Engineering, South China University of Technology. She received her bachelor degree from Shanghai University of Electric Power in 2022. Her research interest covers human-robot skill transfer and adaptive control for robot arm

    LI Miao Associate professor at the Institute of Technological Sciences, Wuhan University. He received his Ph.D. degree from Swiss federal Institute of Technology in Lausanne in 2016. His research interest covers robot learning and multi-robot collaboration

  • 摘要: 碳纤维增强复合材料(Carbon fiber-reinforced composite, CFRC)因具有轻质高强、耐腐蚀、耐冲击等优越性能, 在生产生活中的应用已越来越广泛, 然而复材产品的生产制造仍是劳动密集性产业, 主要依靠人工. 机械臂自上世纪50年代进入工业生产中以来, 极大提高了生产效率和质量, 然而目前机械臂在复材产品制造中的应用是少见的, 主要集中在机械臂形式的自动铺丝(Automated fiber placement, AFP)中. 复材产品制造工艺繁琐, 将复合材料铺放在模具上是复材产品制造过程中的一个重要环节, 本文称之为“铺层”, 使用机械臂完成复合材料自动铺层将是未来复材产品制造自动化、智能化发展的一个关键方向. 本文将机械臂进行复合材料自动铺层操作分为两种主要形式: 铺片和铺带(丝), 通过案例调研和分析, 归纳总结现有的设计理念和技术方法, 提出未来发展趋势, 以期对机械臂的应用和研究、复材产品的智能化制造和工业4.0的发展形成参考.
  • 随着计算机技术和互联网技术的飞速发展, 当今社会进入了大数据时代, 数据的重要性也变得越来越高. 而与此相关的人工智能领域, 例如目标检测[1]、图像识别[2]和语音识别[3]等, 对数据的需求也在不断提高. 但是, 这些图像和语音类数据需要标注的数据量巨大, 采用传统的专家标注方法已经不能满足需求.

    近年来, 随着AMT (Amazon mechanical turk)、CrowdFlower和Clickworker等众包平台的出现, 众包技术为获取大量数据标记提供了一种经济、高效的方式. 众包学习也因此成为了一个新兴的研究领域. 在众包学习中, 首先通过在线的平台雇佣多个众包工人, 对每个实例进行标注, 获得该实例的多噪声标记集. 然后通过标记集成算法, 从每个实例的多噪声标记集中推理出一个合理的集成标记. 目前, 已经有研究者在标记集成算法的研究上做了大量工作, 例如: MV (Majority voting)算法[4]、ZC (ZenCrowd)算法[5]、MNLDP (Multiple noisy label distribution propagation)算法[6]、M3V (Max-margin majority voting)算法[7]、QS-LFC (Quality-sensitive learning from crowds)算法[8].

    然而, 众包工人在专业知识水平、打标积极性和对评价指标的理解等方面的差异性, 导致数据标注的质量普遍偏低. 因此标记集成算法得到的集成标记中仍然存在一定比例的噪声, 即实例的集成标记与真实标记不一致. 标记噪声的存在会损害数据集的标记质量, 而具有高标记质量的数据集在相关研究和应用技术中又是至关重要的. 所以, 对标记集成后的数据集通过众包标记噪声纠正算法进行纠正从而提高数据集的标记质量具有重要的研究意义.

    为了降低标记噪声的影响, 提高标记集成后的数据集的标记质量, 目前已经有研究者在众包标记噪声纠正方向做了一些工作, 主要可分为3类:

    1)基于监督学习的众包标记噪声纠正算法. Nicholson等[9]提出了适用于众包领域的标记噪声纠正算法PL (Polishing labels). PL算法通过将数据集分成10个子集, 并在每个子集上训练一个分类器, 由生成的10个分类器对每个实例进行分类投票, 最终将得票最高的标记赋予实例. 而Xu等[10]结合重抽样的思想, 提出了一种基于重抽样的众包标记噪声纠正算法(Resampling-based noise correction, RNC). RNC通过在干净集和噪声集上按照比例多次重抽样训练多个分类器, 再用得到的多个分类器对整个数据集进行纠正, 从而提高数据集的标记质量. 除了以上两种众包标记纠正算法, 部分研究者还将研究的重点聚焦于监督学习领域知识和众包数据集信息的不确定性方法, 其中的代表性算法有AVNC (Adaptive voting noise correction)算法[11]、BMNC (Between-class margin-based noise correction)算法[12]以及CENC (Cross-entropy-based noise correction)算法[13]. AVNC算法提出了众包领域的噪声纠正框架, 利用众包服务提供的标记信息, 估计了众包工人的质量, 并进一步评估了数据集中含有噪声的比例. 然后对实例的噪声等级进行了排序, 最终通过集成模型对被认定为噪声的实例进行纠正. BMNC算法则是利用了众包数据集中实例的多噪声标记集信息, 评估每个实例的集成标记的置信程度, 从而更加精准地过滤出噪声实例, 并通过最终得到的干净集训练分类器对噪声实例进行纠正. 而CENC算法同样利用了众包数据集中实例的多噪声标记集, 计算多噪声标记集的信息熵去评估集成标记的置信程度, 然后在得到的干净集上训练多个分类器对噪声集实例进行预测, 进一步利用交叉熵的思想去衡量噪声集中实例标记的真实分布和预测分布的相似度, 从而实现对噪声实例的纠正, 提高数据集的标记质量.

    2)基于无监督学习的众包标记噪声纠正算法. Nicholson等[9]提出了一种基于聚类的众包标记噪声纠正算法CC (Cluster-based correction). 聚类是无监督学习技术, 在噪声处理领域的优势是不依赖实例的类标记. 这种优势的存在使得该方法在性能上往往优于基于监督学习的算法, 但是CC算法在时间效率方面则较差. CC的核心思想是进行多次聚类算法, 在每次聚类执行时, 对每个簇中的实例计算并赋予相同的权重, 权重反映了实例属于不同类别的可能性. 最终根据每个实例的最大权值对实例进行重新标注, 从而提高数据集的标记质量.

    3)基于半监督学习的众包标记噪声纠正算法. Nicholson等[9]提出的STC (Self-training correction)算法则是受自训练过程的启发, 首先用过滤器将数据集分为干净集和噪声集, 然后在干净集上训练分类器用于对噪声集中的实例进行预测打标, 再选取每类中打标置信度最高的噪声实例, 将预测标记赋予实例并加入干净集中, 重复上述步骤直到纠正的噪声实例达到设定的阈值. 通过以上步骤, 该算法提高了数据集的标记质量和模型精度.

    在上文提到的3类众包标记噪声纠正算法中, 基于监督学习的方法训练分类器估计实例的集成标记质量, 并对噪声实例进行纠正; 基于无监督学习的方法不依赖数据集的集成标记, 通过数据的特征估计实例所属某类的权重; 基于半监督学习的方法是将干净集实例作为已标记实例来训练分类器, 对噪声集实例进行重新标注的方法. 在基于半监督学习的方法中, STC算法因为简单有效, 不需要特定的假设条件而被广泛应用. 但通过对STC算法的分析, 本文发现该算法仍然存在3个方面的不足: 1)当用于训练初始分类器的干净集不能很好表示整个数据空间, 难以反映数据的分布时, STC算法得到分类器的效果较差. 从而较多错误纠正的实例加入干净集, 误差的影响在循环中不断扩大, 导致STC算法效果受损; 2)经过过滤得到的干净集仍然存在噪声实例, 而STC算法是在干净集上训练单个分类器用于纠正噪声集, 因此训练的分类器纠正效果不佳; 3) STC算法是基于半监督学习领域的思想, 未充分利用众包数据集所含的标记信息, 难以取得更好的众包标记噪声纠正效果.

    针对STC算法存在的不足, 本文提出了一种基于自训练的众包标记噪声纠正算法(Self-training-based label noise correction, STLNC). 本文同时使用仿真标准数据集和真实众包数据集进行了实证研究, 结果表明在噪声比和模型质量两个度量指标上, STLNC算法比其他五种最先进的标记噪声纠正算法表现更好.

    在一个众包系统中, N个实例由J个工人打标后, 获得了众包数据集$D = \left\{ {\left( {{{\boldsymbol{x}}_i},{L_i}} \right)} \right\}_{i = 1}^N $, 其中$ {L_i} = \left\{ {{l_{ij}}} \right\}_{j = 1}^J$为每个实例的多噪声标记集, $l_{ij} $表示实例${\boldsymbol{x}}_i $由众包工人${u_j}\left( {j = 1,2, \cdots ,J} \right) $给出的标记. 对于二分类问题, $ l_{ij} $属于{+, 0, −}, 其中的值分别表示众包工人对实例打正标, 未给出标记和打负标. 众包数据集$ D$中的每个实例$ {\boldsymbol{x}}_i $在经过标记集成算法推理得到集成标记${\hat y_i} $后, 获得了带有集成标记的众包数据集$ D' = \left\{ {\left( {{{\boldsymbol{x}}_i},{{\hat y}_i}} \right)} \right\}_{i = 1}^N$. 而数据集$ D'$正是众包标记噪声纠正工作的起点.

    上文分析了STC算法的不足, 为了提高STC算法的性能, 本文提出的STLNC算法利用加权密度峰值聚类算法构建数据集中低密度实例指向高密度实例的空间结构关系, 以便更好地指导噪声集实例的选择. 同时, 将STC中的单个分类器替换为集成分类器, 在纠正阶段利用设计的纠正策略对选择的噪声实例进行纠正, 降低了纠正的误差, 提高了数据集的标记质量和模型精度, 整个算法的框架如图1所示.

    图 1  STLNC算法的框架
    Fig. 1  Framework of STLNC

    图1可以看出, STLNC算法工作的起点是带有集成标记的数据集$D' $, 该数据集的集成标记是通过标记集成算法推理获得的. 在获得数据集$ D'$后, STLNC算法可分为3个阶段: 1)噪声过滤; 2)构建空间结构关系; 3)实例选择与纠正. 在噪声过滤阶段, 采用过滤器对$D' $进行过滤, 得到干净集$D_c'$和噪声集$D_n'$; 在构建空间结构关系阶段, 利用加权密度峰值聚类构建数据集中低密度实例指向高密度实例的空间结构关系; 在实例选择与纠正阶段, 根据获得的干净集和噪声集以及空间结构关系设计实例选择策略, 在每次循环中选择最适合纠正的噪声实例. 再利用干净集$D_c'$训练集成分类器$M $, 对选择的噪声实例通过设计的实例纠正策略进行纠正, 然后加入到$D_c'$中. 循环执行实例选择和纠正步骤, 直到所有噪声实例被选择并纠正, 获得最后一轮训练的集成分类器$M'$. 最后采用集成分类器$M'$通过多数投票(Majority voting)的方式对$D'$所有实例进行纠正, 得到最终纠正后的数据集$D_{{\rm{correct}}}'$.

    在整个STLNC算法流程中, 空间结构关系的构建以及实例选择和纠正策略的设计是本算法的关键, 本文将在后续内容中详细介绍.

    1.2.1   构建空间结构关系

    为了解决精确选择噪声实例的问题, 受到Wu等[14]的启发, STLNC算法利用了加权密度峰值聚类算法构建数据集中低密度实例指向高密度实例的空间结构关系, 从而指导噪声实例选择过程.

    密度峰值聚类的主要思想基于以下两点: 1)聚类中心的密度大于其邻居的密度, 也就是被低密度实例围绕; 2)聚类中心与更高密度的实例之间的距离较远, 即簇间距离较大. 对于众包数据集$D^{'} $中每一个实例${{\boldsymbol{x}}_i} $, 它的局部密度$ {\rho _i}$定义为

    $$ {\rho _i} = \sum\limits_{j,j \ne i} {{I} }\left( {{d_{ij}} < {d_c}} \right) $$ (1)

    其中, ${I}\left( \cdot \right)$为指示函数, $ {d_{ij}} < {d_c}$值取1, 否则为0. ${d_{ij}} $为实例$ {{\boldsymbol{x}}_i}$${{\boldsymbol{x}}_j} $之间的距离, 一般为欧氏距离. $ d_c$为截断距离, 具体定义为

    $$ {d_c} = {d_{C_N^2 \times s}} $$ (2)

    其中, ${d_{C_N^2 \times s}} \in [ {{d_1},{d_2}, \cdots ,{d_{C_N^2}}}]$, 该集合包含每对实例之间的距离, 并按照升序排序. ${d_{C_N^2 \times s}}$代表该集合中第${C_N^2 \times s}$个距离, s为原始密度峰值聚类算法给出的经验参数, 本文取值为0.2.

    在获取到局部密度${\rho _i} $之后, 密度峰值聚类算法需要进一步得到实例$ {{\boldsymbol{x}}_i}$${\delta _i} $值, ${\delta _i} $定义为: 1) 对于非最高密度实例${{\boldsymbol{x}}_i} $, 计算实例${{\boldsymbol{x}}_i} $与较高密度最近邻的距离; 2) 对于最高密度实例${{\boldsymbol{x}}_i} $, 计算实例$ {{\boldsymbol{x}}_i}$和最远实例之间的距离. 具体为

    $$ {\delta _i} = \left\{ {\begin{array}{*{20}{l}} {{{\max }_j}\left( {{d_{ij}}} \right),}&{\forall j,\;{\rho _i} \ge {\rho _j}}\\ {{{\min }_{j:{\rho _i} < {\rho _j}}}\left( {{d_{ij}}} \right),}&{其他} \end{array}} \right. $$ (3)

    在计算出所有实例的局部密度$\rho _i $和距离$ \delta _i$之后, 局部密度$\rho _i $和距离$ \delta _i$均相对较大的实例将作为簇的中心, 剩下的实例被归于密度较高的最近邻所属的簇, 从而得到最终的聚类结果.

    在本文中, 为了精确构建众包数据集实例的空间结构关系, 我们将众包数据集的信息引入密度峰值聚类算法的距离计算中. 对于每个众包数据集中的实例, 实例所带有的多噪声标记集以及众包数据集的工人质量是能够反映实例之间相关关系的重要属性. 因此本文在计算实例$ {{\boldsymbol{x}}_i}$$ {{\boldsymbol{x}}_j}$之间的距离${d_{ij}} = dist\left( {{{\boldsymbol{x}}_i},{{\boldsymbol{x}}_j}} \right) $时, STLNC算法通过使用实例的多噪声标记集衡量实例之间的相关关系, 具体定义如下:

    $$ \begin{split} {d_{ij}} = \;&{dist}\left( {{{\boldsymbol{x}}_i},{{\boldsymbol{x}}_j}} \right)=\\ & {w_{ij}}\sqrt {\sum\limits_{k = 1}^K {{{\left( {{x_{ik}} - {x_{jk}}} \right)}^2}} } \end{split} $$ (4)

    其中, $K $是特征维度, ${w_{ij}} $表示工人对实例$ {\boldsymbol{x}}_i$$ {\boldsymbol{x}}_j$的打标相似度. 为了利用实例的多噪声标记集计算实例${\boldsymbol{x}}_i $$ {\boldsymbol{x}}_j$的打标相似度, 首先需要根据实例的噪声标记集${L_i} = \left\{ {{l_{ij}}} \right\}_{j = 1}^J $, 构建每个实例的标记分布, 对二类数据集, 实例具体的标记分布为${{\boldsymbol{P}}_i} = \left\{ {p_i^ + ,p_i^ - } \right\}$, 其中$ {p_i^ + }$代表的是所有工人对实例${{\boldsymbol{x}}_i} $打标为正类的概率, ${p_i^ - } $代表的是所有工人对实例$ {{\boldsymbol{x}}_i}$打标为负类的概率. 具体为

    $$ \left\{\begin{aligned} &p_i^ + = \frac{{Nu{m_i}( + ) + 1}}{{Nu{m_i}( + ) + Nu{m_i}( - ) + 2}}\\ &p_i^ - = \frac{{Nu{m_i}( - ) + 1}}{{Nu{m_i}( + ) + Nu{m_i}( - ) + 2}} \end{aligned}\right. $$ (5)

    其中, ${Nu{m_i}} $(+)为实例${{\boldsymbol{x}}_i} $多噪声标记集中正类标记的数目, $ {Nu{m_i}( - )}$为实例${{\boldsymbol{x}}_i}$多噪声标记集中负类标记的数目. 为了满足后续标记分布之间关系计算的需要, 式(5)使用了拉普拉斯纠正[15]. 在本算法中, 使用标记分布的好处是能够使用数值显式地表示工人打标的分布, 以便精确估计两个实例之间打标的相似程度. 因此在获得每个实例的标记分布之后, 我们采用KL (Kullback-Leibler)散度[16]来计算两个实例标记分布的近似程度, KL散度越小则说明工人对两个实例打标的情况越接近, 因此$ w_{ij}$也越小, 具体为

    $$ \begin{split} {w_{ij}} =\;& \frac{1}{2}\left( {KL\left( {{{\boldsymbol{P}}_i}||{{\boldsymbol{P}}_j}} \right) + KL\left( {{{\boldsymbol{P}}_j}||{{\boldsymbol{P}}_i}} \right)} \right)=\\ &\frac{1}{2}\left( {\sum\limits_{m \in \{ + , - \} } {{p_i}} (m) \cdot \log \frac{{{p_i}(m)}}{{{p_j}(m)}}}+ \right.\\ &\left. { \sum\limits_{m \in \{ + , - \} } {{p_j}} (m) \cdot \log \frac{{{p_j}(m)}}{{{p_i}(m)}}} \right) \end{split} $$ (6)

    其中, ${{p_i}(m)} $表示实例$ {{\boldsymbol{x}}_i}$标记属于$m $类的概率.

    本文在根据实例之间的打标相似度$ w_{ij}$得到$d_{ij} $后, 继续进行密度峰值聚类过程. 直到计算实例的${\delta _i} $值时, 会为每个实例${{\boldsymbol{x}}_i} $找到具有更高密度的最近邻实例$ {{\boldsymbol{x}}_j}$. 根据Wu等[14]的证明, 通过实例${{\boldsymbol{x}}_i} $指向实例${{\boldsymbol{x}}_j} $可以发现整个数据空间的真实结构. 这种由低密度实例${{\boldsymbol{x}}_i} $指向高密度实例${{\boldsymbol{x}}_j} $的空间结构关系不会因为实例的标记改变而发生变化, 是一种稳定的结构关系, 能够指导我们发现密度更高更具有代表性的实例. 因此, 本文执行加权密度峰值聚类步骤计算所有实例的${\delta _i} $, 从而也获得了整个众包数据集的空间结构关系.

    1.2.2   实例选择与纠正策略

    在通过加权密度峰值聚类算法构建数据集中低密度实例指向高密度实例的空间结构关系之后, 本文结合噪声过滤阶段得到的干净集和噪声集, 设计了两种噪声实例选择策略. 首先, 本文定义了两类实例点, 当实例$ {{\boldsymbol{x}}_i}$指向实例${{\boldsymbol{x}}_j} $, 那么定义$ {{\boldsymbol{x}}_i}$${{\boldsymbol{x}}_j} $的“previous”点, $ {{\boldsymbol{x}}_j}$${{\boldsymbol{x}}_i} $的“next”点.

    1) 实例选择策略A: 在整个数据空间中, 搜寻所有“previous”点为干净集实例, “next”点为噪声集实例的实例对, 将每个实例对中的噪声集实例作为纠正的对象;

    2)实例选择策略B: 在整个数据空间中, 搜寻所有“previous”点为噪声集实例, “next”点为干净集实例的实例对, 将每个实例对中的噪声集实例作为纠正的对象.

    在实例选择与纠正阶段, 首先根据实例选择策略A选择噪声实例, 当没有满足条件的噪声实例时, 再使用实例选择策略B选择实例, 直到噪声集实例全部纠正并加入干净集. 根据第1.2.1节的说明, 本文构建空间结构关系目的是为了发现实例之间的真实结构, 指导算法寻找更高密度更具代表性的实例, 从而训练出更加可靠的分类器. 因此本文先执行了实例选择策略A来寻找具有更高密度的噪声实例, 这些高密度噪声实例更具代表性. 将这些高密度噪声实例先进行选择并纠正能够使扩充后的干净集更好的代表整个数据空间, 使训练得到的分类器更加可靠, 提高标记纠正的效果. 在执行完实例选择策略A直到没有满足条件的噪声实例后, 算法将训练得到一个可靠的分类器. 为了进一步扩充干净集, 本文随后执行了实例选择策略B, 对剩余的低密度噪声实例进行选择, 从而完成了整个实例选择的过程.

    解决噪声实例选择的问题后, 剩下的问题就是如何设计实例纠正策略, 提高纠正的准确性. 在STLNC算法中, 为了提高纠正的准确性, 本文将用于标记纠正的单个分类器替换为由3个异质的分类器构成的集成分类器. 同时根据第1.2.1节得到的每个实例的标记分布${{\boldsymbol{P}}_j} $, 计算每个实例多噪声标记集的信息熵值$Entropy\left( {{{\boldsymbol{P}}_i}} \right) $, 从而估计该实例集成标记的置信度$e_i $. 当$e_i $越小时, 说明工人打标的一致性越高, 则集成标记的置信度越高; 反之则集成标记的置信度越低. $ e_i$的具体计算为

    $$ {e_i} = { Entropy }\left( {{{\boldsymbol{P}}_i}} \right)= - \sum\limits_{m \in \{ + , - \} } {{p_i}} (m)\log {p_i}(m) $$ (7)

    其中, ${p_i}(m) $表示实例$ {{\boldsymbol{x}}_i}$标记属于$ m$类的概率.

    本文结合上述集成分类器和众包数据集标记信息, 设计了STLNC的实例纠正策略:

    1)当实例${{\boldsymbol{x}}_i} $的集成标记置信度$e_i $大于设定的阈值$T $, 则说明${{\boldsymbol{x}}_i} $集成标记的质量较低. 当集成分类器对${{\boldsymbol{x}}_i} $进行纠正时, 采用多数投票(Majority voting)的策略对$ {{\boldsymbol{x}}_i}$进行标注.

    2)当实例$ {{\boldsymbol{x}}_i}$的集成标记置信度$ e_i$小于设定的阈值$T $, 则说明${{\boldsymbol{x}}_i} $集成标记的质量较高. 当集成分类器对$ {{\boldsymbol{x}}_i}$进行纠正时, 采用共识投票(Consensus voting)的策略对$ {{\boldsymbol{x}}_i}$进行标注, 即当三个分类器打标结果相同, 则对${{\boldsymbol{x}}_i} $进行纠正, 否则不纠正${{\boldsymbol{x}}_i} $的标记.

    综上所述, 本文提出的STLNC算法的详细步骤如算法1所示.

    算法1. STLNC算法

    输入. 带集成标记的数据集${D'} = \left\{ {\left( {{{\boldsymbol{x}}_i},{{\hat y}_i}} \right)} \right\}_{i = 1}^N$, 实例集成标记置信度阈值$ T$.

    输出. 纠正后的干净数据集$D_{{\rm{correct}}}'$.

    应用CF过滤器过滤数据集${D'}$, 将${D'}$分为干净集$D_c'$ 和噪声集$D_n'$;

    for $i = 1\;{\rm{to}}\;{\rm{sizeof}}\left( {{D'}} \right)$ do

    根据实例$i $的多噪声标记集$L_i $, 获得$i $的标记分布$ {{\boldsymbol{P}}_i}$;

    end for

    for $i = 1\;{\rm{to}}\;{\rm{sizeof}}\left( {{D'}} \right)$ do

    for $j = 1\;{\rm{to}}\;{\rm{sizeof}}\left( {D'} \right)\;\& \& \;i! = j $ do

    计算实例$ i$$j $之间的$w_{ij} $;

    根据$ w_{ij} $, 计算实例$i $$j $的距离$d_{ij} $;

    end for

    根据$d_{ij} $计算实例$i $的局部密度$ {\rho _i}$;

    根据$d_{ij} $$ {\rho _i}$计算实例$ i$$ {\delta _i}$值;

    end for

    获取带有指向关系的实例对集合 ${D_{{\rm{couple}}}} =\{ \left( previou{s_i}, nex{t_i} \right) \}_{i = 1}^U$;

    count = 1;

    while $count > 0\;\& \& \;{\rm{sizeof}}\left( {D_n'} \right)> 0$ do

    count = 0;

    $D_c'$上训练集成分类器$ M$;

    for $i = 1\;{\rm{to}}\;{\rm{sizeof}}\left( {{D_{{\rm{couple}}}}} \right)$ do

    if $previou{s_i} \in D_c'\;\& \& \;nex{t_i} \in D_n'$

    通过集成分类器$ M$对实例$nex{t_i} $按照实例纠 正策略进行纠正, 并加入干净集$D_c'$;

    count++;

    end if

    end for

    根据新的干净集和噪声集更新$ {D_{{\rm{couple}}}}$;

    end while

    count = 1;

    while $count > 0\;\& \& \;{\rm{sizeof}}\left( {D_n'} \right)> 0$ do

    count = 0;

    $D_c'$上训练集成分类器$M $;

    for $i = 1\;{\rm{to}}\;{\rm{sizeof}}\left( {{D_{couple}}} \right) $ do

    if $previou{s_i} \in D_n'\;\& \& \;nex{t_i} \in D_c'$

    通过集成分类器$ M$对实例$previou{s_i} $按照实例 纠正策略进行纠正, 并加入干净集$D_c'$;

    count++;

    end if

    end for

    根据新的干净集和噪声集更新${D_{{\rm{couple}}}} $;

    end while

    获得最后一轮循环训练的集成分类器$M'$;

    $M'$$D'$中所有实例进行纠正, 得到$D_{{\rm{correct}}}'$;

    return $D_{{\rm{correct}}}'$.

    在本节中$, $ 为了验证提出的STLNC算法的有效性, 本文在仿真标准数据集和真实众包数据集上进行了实验. 将STLNC算法与目前5种最先进的众包标记噪声纠正算法PL[9]、STC[9]、CC[9]、AVNC[11]和CENC[13]在纠正后数据集噪声比以及训练模型的模型质量两个度量指标上进行了比较. 其中, 数据集噪声比(Noise ratio)定义为经过算法纠正后的众包数据集中集成标记与真实标记不同的实例所占整个数据集的百分比. 模型质量(Model quality)定义为在纠正后的众包数据集上训练的目标分类模型的测试精度.

    利用CEKA (Crowd environment and its knowledge analysis)平台[17], 本文实现了STLNC算法和CENC算法, 同时使用了该平台现有的MV、PL、STC、CC和AVNC算法和WEKA (Waikato environment for knowledge analysis)平台[18]上现有的K-means算法、C4.5算法、KNN (K-nearest neighbor)算法和LR (Logistic regression)算法. 6种众包标记噪声纠正算法在实验中设置如下:

    1) PL: PL算法使用的基分类器为C4.5;

    2) STC: STC算法使用的基分类器为C4.5, 使用的过滤器为CF (Classification filter)过滤器[19], 需要被纠正的噪声实例比例设置为0.8;

    3) CC: CC算法使用的聚类方法为K-means聚类算法, 该算法执行次数设置为10次, 而簇中心的个数设置为2到数据集实例个数的一半;

    4) AVNC: AVNC算法使用的基分类器为C4.5, 数据集分割为大小相同子集的个数设置为10个, 训练的分类器个数设置为5;

    5) CENC: CENC算法使用的基分类器为C4.5, 训练的分类器个数设置为10个, 而实例的多噪声标记集的信息熵阈值$ T$设置为0.1;

    6) STLNC: STLNC算法使用的集成分类器由C4.5、KNN $(K = 3)$ 和LR构成, 使用的过滤器为CF过滤器, 截断距离$d_c $s的值设置为0.2, 实例的多噪声标记集信息熵阈值$T $设置为0.1.

    其中, CF过滤器将整个数据集分割为10个大小相同的子集, 而且所使用的基分类器同样为C4.5.

    为了验证STLNC在不同工人质量和数据特征下的有效性, 本文使用了22个二分类标准数据集由仿真实验模拟不同专业水平的工人进行打标, 数据信息具体如表1所示, 其中“#Ins”表示数据集实例的数量, “#Att”表示数据集实例的属性维度, “#Pos”表示标记为正的实例数量, “#Neg”表示标记为负的实例数量. 为了模拟众包过程为每个实例获取多噪声标记集, 对使用的数据集隐藏了实例的真实标记, 同时模拟了9个工人对每个实例打标的过程, 其中每个工人的打标质量为${p_j}\;( j = 1,2, \cdots , 9)$. 这代表着每个工人有$ p_j$的概率给实例打正确标记, $1-p_j $的概率打错误标记. 值得注意的是, 数据集中的正例和负例均有$p_j $概率被正确标记, $ 1-p_j$概率被错误标记.

    表 1  22个仿真标准数据集详细描述
    Table 1  Description of 22 simulated benchmark datasets
    数据集#Ins#Att#Pos#Neg
    biodeg105541356699
    breast-cancer268985201
    breast-w69910241458
    credit-a69016383307
    credit-g100021300700
    diabetes7688268500
    heart-statlog2701412032
    hepatitis1552012332
    horse-colic36822232136
    ionosphere35135225126
    kr-vs-kp31963715271669
    labor57163720
    mushroom81242339164208
    sick3772302313541
    sonar2086111197
    spambase4601578132788
    tic-tac-toe95810332626
    vote43517168267
    climate5402049446
    colic36822136232
    monks4326228204
    steel-plates-faults1941336731268
    下载: 导出CSV 
    | 显示表格

    为了保证在不同工人质量下实验结果的可靠性, 我们设置了两种工人质量的方案:

    1)在第1个系列的实验中, 所有工人的打标质量设置为0.6, 即$ p_j=0.6$;

    2)在第2个系列的实验中, 所有工人的打标质量由均匀分布的区间$\left[ {0.55,0.75} \right] $随机生成, 即${p_j}\in \left[ {0.55,0.75} \right] $.

    在为每个实例获得9个带有噪声的众包标记之后, 本文使用最经典的标记集成算法MV获取每个实例的集成标记, 并将该算法处理后的数据集噪声比和训练模型质量作为基线. 然后, 使用6种不同的众包标记噪声纠正算法对带有集成标记的众包数据集中的噪声实例进行识别并纠正, 目标分类模型将在纠正后的数据集上进行训练. 最后, 本文将评估每个数据集上各种众包标记噪声纠正算法纠正后的噪声比以及训练模型质量. 值得注意的是, 与评估数据集的噪声比不同, 我们在评估训练模型质量的时候采用了10折交叉的验证方法, 特别是在同一个数据集上运行不同算法的时候用了相同的训练集以及测试集.

    表2表3详细展示了在工人质量$p_j=0.6 $的情况下各个算法在不同数据集上的实验结果. 根据表2中的纠正后的数据集中的噪声比以及表3中的模型质量, 本文采用了威尔科克森(Wilcoxon)符号秩检验[20-21]来比较实验用到的每一对标记噪声纠正算法. 表4表5分别展示了每组实验的威尔科克森符号秩检验的比较结果. 其中符号“•”代表该行中的算法明显优于对应列中的算法, 符号“$\circ$”代表该列中的算法明显优于对应行中的算法. 表中, 主对角线以下区域的显著性水平为$ \alpha = 0.05$; 而主对角线以上区域的显著性水平$ \alpha = 0.1$.

    表 2  工人质量0.6时的噪声比对比结果 (%)
    Table 2  Noise ratio comparisons with pj = 0.6 (%)
    数据集MVPLSTCCCAVNCCENCSTLNC
    biodeg28.2529.9528.3419.5318.4821.9015.83
    breast-cancer27.6226.9225.8731.1226.5729.3724.84
    breast-w28.769.0119.3110.309.168.447.30
    credit-a26.6720.0015.9418.8413.0413.3312.90
    credit-g26.6027.4028.4026.6025.3027.5026.40
    diabetes26.6932.2926.5626.9523.7023.9622.79
    heart-statlog25.1919.2623.7022.9624.0725.9318.52
    hepatitis30.3219.3526.4520.6527.7425.1630.97
    horse-colic27.7232.3417.3921.2017.6614.1314.13
    ionosphere27.9216.2421.659.1210.8313.3911.68
    kr-vs-kp27.3821.9610.4519.342.192.852.28
    labor31.5824.5624.5615.7912.2831.587.02
    mushroom26.7112.656.434.300.040.100
    sick27.602.608.8310.311.782.283.37
    sonar26.9231.7329.3324.0425.0024.5218.75
    spambase27.0227.4719.5014.789.1110.568.06
    tic-tac-toe26.2034.1323.0724.4322.4422.2314.61
    vote25.984.6010.3411.263.914.144.14
    climate27.418.5227.4114.078.528.528.52
    colic27.4522.2820.9223.1014.1314.9513.59
    monks26.3925.0011.3421.765.326.712.78
    steel-plates-faults27.5134.989.2218.7000.100.15
    平均值27.4521.9719.7718.6013.6915.0812.21
    下载: 导出CSV 
    | 显示表格
    表 3  工人质量0.6时的模型质量对比结果(%)
    Table 3  Model quality comparisons with pj = 0.6 (%)
    数据集MVPLSTCCCAVNCCENCSTLNC
    biodeg71.3775.9172.1378.7774.3474.2178.29
    breast-cancer67.0071.2869.9869.0870.9268.1369.73
    breast-w92.8592.4790.6893.3894.0092.8595.54
    credit-a82.0384.7884.4984.7883.9183.0483.48
    credit-g62.2068.3067.4069.7067.3063.9070.40
    diabetes71.7470.5671.5070.8071.7272.1274.00
    heart-statlog65.5674.8167.7870.0074.0769.2678.15
    hepatitis69.0077.6771.3377.5072.0070.1774.17
    horse-colic77.4278.1180.0080.1581.6281.3682.35
    ionosphere83.4885.4586.9085.1985.7784.0485.76
    kr-vs-kp95.1895.4996.6290.2996.8196.5997.03
    labor70.6761.8373.5068.3368.3372.3376.33
    mushroom99.8598.5699.8899.9099.9099.8699.83
    sick96.7494.6296.9894.4897.7797.7597.11
    sonar58.9355.5750.0758.2955.0059.1458.29
    spambase85.9288.8787.4484.2089.6888.9890.39
    tic-tac-toe77.1769.7474.5471.6374.6374.6378.36
    vote89.8995.3794.2190.5994.2193.9894.21
    climate91.4891.4891.4891.4891.4891.4891.48
    colic79.9781.0981.0982.4781.0982.4781.09
    monks90.7590.7393.5183.3593.5193.5193.28
    steel-plates-faults100.0089.64100.0092.01100.00100.00100.00
    平均值80.8781.4781.8981.2082.6482.2684.06
    下载: 导出CSV 
    | 显示表格
    表 4  工人质量0.6时的噪声比的威尔科克森测试结果
    Table 4  Noise ratio summary of Wilcoxon tests with pj = 0.6
    MVPLSTCCCAVNCCENCSTLNC
    MV
    PL
    STC
    CC
    AVNC
    CENC
    STLNC
    下载: 导出CSV 
    | 显示表格
    表 5  工人质量0.6时的模型质量的威尔科克森测试结果
    Table 5  Model quality summary of Wilcoxon tests with pj = 0.6
    MVPLSTCCCAVNCCENCTTLNC
    MV
    PL
    STC
    CC
    AVNC
    CENC
    STLNC
    下载: 导出CSV 
    | 显示表格

    表2 ~ 5给出了第1个系列实验的详细对比, 验证了STLNC算法在提高数据集标记质量和模型质量上的有效性. 具体结论如下:

    1) STLNC算法纠正后的数据集的平均噪声比为12.21%, 低于MV (27.45%)、PL (21.97%)、STC (19.77%)、CC (18.60%)、AVNC (13.69%)和CENC (15.08%).

    2) STLNC算法纠正后的数据集训练的目标模型的平均模型质量为84.06%, 高于MV (80.87%)、PL (81.47%)、STC (81.89%)、CC (81.20%)、AVNC (82.64%)和CENC (82.26%).

    3) 根据威尔科克森符号秩检验的结果, STLNC算法在噪声比和模型质量两个指标上都要显著优于对比的众包标记噪声纠正算法.

    在第2个系列实验中, 表6 ~ 9详细展示了在工人质量${p_j} \in \left[ {0.55,0.75} \right] $的情况下各个众包标记噪声纠正算法在纠正后数据集噪声比和模型质量上的实验结果, 以及威尔科克森符号秩检验中的算法差异显著性的对比结果. 根据表6表7中的实验结果, 在数据集的噪声比和模型质量两个度量指标上, STLNC算法的效果都是最好的. 同时, 根据表8表9中的威尔科克森符号秩检验的结果, STLNC算法在纠正后数据集的噪声比上显著优于对比算法MV、PL、STC、CC, 与AVNC、CENC算法性能相当. 而在模型质量上显著优于对比算法MV、PL、STC、CC、CENC, 与AVNC算法性能相当.

    表 6  工人质量[0.55, 0.75]时的噪声比对比结果 (%)
    Table 6  Noise ratio comparisons with pj ∈ [0.55, 0.75] (%)
    数据集MVPLSTCCCAVNCCENCSTLNC
    biodeg14.2221.1416.0213.8413.4613.0812.89
    breast-cancer16.4326.2220.9819.9323.4324.8324.48
    breast-w20.463.7210.014.154.154.433.72
    credit-a18.4120.5814.9313.6213.7713.0412.17
    credit-g17.7029.6022.7022.9021.6022.3024.60
    diabetes20.1822.6624.0922.2723.4422.6623.44
    heart-statlog16.3020.3715.1920.0016.6716.6718.52
    hepatitis12.2620.6514.1914.8416.7712.9012.26
    horse-colic17.6615.4913.8618.7514.6714.1315.22
    ionosphere17.3818.8013.689.6911.1110.8313.96
    kr-vs-kp17.4325.195.6011.551.311.882.44
    labor17.5429.8217.5412.2821.0521.0514.04
    mushroom18.074.944.841.670.100.110
    sick13.941.784.983.761.461.542.04
    sonar15.3837.5021.6325.9619.2322.6020.67
    spambase19.3237.5415.119.047.007.046.67
    tic-tac-toe20.6727.4519.3117.5415.7614.416.47
    vote22.076.9010.578.974.374.834.60
    climate22.968.5222.9610.748.528.528.52
    colic16.5819.5715.4917.9315.2214.4015.49
    monks17.1312.737.1823.382.784.862.78
    steel-plates-faults22.4634.837.3215.920.260.260.21
    平均值17.9320.2714.4614.4911.6411.6511.15
    下载: 导出CSV 
    | 显示表格
    表 7  工人质量[0.55, 0.75]时的模型质量对比结果 (%)
    Table 7  Model quality comparisons with pj ∈ [0.55, 0.75] (%)
    数据集MVPLSTCCCAVNCCENCSTLNC
    biodeg74.5881.8776.3679.9981.5980.2581.78
    breast-cancer69.4371.8169.5070.2771.2871.6471.64
    breast-w90.7694.5491.1094.3493.8592.4094.69
    credit-a82.1785.3684.7885.6585.6584.6484.93
    credit-g69.5069.8070.5069.6071.1069.0072.00
    diabetes71.6774.4471.1574.2173.1372.8774.56
    heart-statlog70.0078.8976.3075.9379.6378.5280.37
    hepatitis76.1779.1777.3380.5076.8377.8379.00
    horse-colic82.5080.3581.5782.6383.5183.0182.68
    ionosphere80.6283.4882.3388.0387.7386.9084.07
    kr-vs-kp97.9495.3497.6692.8698.2898.0098.06
    labor78.3368.1778.3364.3377.1777.1784.33
    mushroom99.9998.5299.9599.96100.00100.0099.95
    sick97.7296.9597.4895.4797.6497.8397.14
    sonar63.7968.9367.1470.3669.2969.5070.86
    spambase86.6588.8788.0784.8390.3788.9690.76
    tic-tac-toe77.8373.0077.5876.3077.8976.8580.08
    vote93.3593.7994.9892.6895.2495.0094.77
    climate91.4891.4891.4891.4891.4891.4891.48
    colic83.7477.9084.1979.1782.5382.6381.84
    monks98.3785.2198.6088.16100.00100.00100.00
    steel-plates-faults99.90100.00100.0099.69100.00100.00100.00
    平均值83.4883.5484.3883.4785.6585.2086.14
    下载: 导出CSV 
    | 显示表格
    表 8  工人质量[0.55, 0.75]时的噪声比的威尔科克森测试结果
    Table 8  Noise ratio summary of Wilcoxon tests with pj ∈ [0.55, 0.75]
    MVPLSTCCCAVNCCENCSTLNC
    MV
    PL
    STC
    CC
    AVNC
    CENC
    STLNC
    下载: 导出CSV 
    | 显示表格
    表 9  工人质量[0.55, 0.75]时的模型质量的威尔科克森测试结果
    Table 9  Model quality summary of Wilcoxon tests with pj ∈ [0.55, 0.75]
    MVPLSTCCCAVNCCENCSTLNC
    MV
    PL
    STC
    CC
    AVNC
    CENC
    STLNC
    下载: 导出CSV 
    | 显示表格

    根据上述两个系列的实验结果, 在不同的工人质量下的仿真数据集上都验证了本文提出的STLNC算法在提高数据集的标记质量和模型质量上的有效性, 并且STLNC算法在两个评估指标上都整体优于对比的5个众包标记噪声纠正算法.

    为了分析实例的多噪声标记集信息熵阈值$ T$的取值对STLNC算法的影响, 同时节省版面和避免重复, 本文随机选择了仿真数据集“ionosphere”在工人质量0.6的情况下进行实验, 同时使用了相同的过滤器CF保证每轮实验的干净集和噪声集相同. 最后评估了在不同$ T$值的情况下STLNC算法在纠正后数据集上噪声比的情况.

    在本文中设置$T $值的目的是协助实例纠正策略, 防止集成标记质量高的实例被实例纠正策略给错误纠正. 从图2可以看出, 随着$ T$值的变化, 噪声比呈现梯度上升, 从而验证了本文设置$T $值的想法, $T $值的取值范围为0.1 ~ 0.5, 为了精确筛选高标记质量实例, 本文设置$T$为较小值0.1.

    图 2  不同T值的STLNC在ionosphere数据集上的噪声比结果
    Fig. 2  Noise ratio of STLNC with different T values on ionosphere dataset

    为了进一步验证STLNC算法的有效性, 本文使用CEKA平台自带的真实众包数据集Leaves和文献[22]中使用的真实众包数据集LabelMe进行实验. 同时, 在真实众包数据集上的实验, 本文还针对STLNC算法过滤阶段过滤器的选取进行了讨论. 最后, 针对STLNC中构建空间结构关系阶段和实例选择与纠正阶段进行了消融实验.

    Leaves数据集包含384个带有63维特征的实例. 每个实例都是一张树叶图片, 由众包工人根据树叶的特征进行标注, 因此Leaves数据集中每个实例都带有多个含噪声的标记. 而LabelMe数据集则是包含1000个带有200维特征的实例. 每个实例都是一张拍摄的图像, 同样发布在AMT平台上, 由不同众包工人进行标注.

    本文从Leaves和LabelMe中提取了8个只包含二类的数据集, 这8个数据集分别记为Leaves1, Leaves2, Leaves3, Leaves4, LabelMe1, LabelMe2, LabelMe3和LabelMe4, 这些数据集的具体细节如表10所示. 例如: “LabelMe1”的目标是判断实例为高速公路还是街道的图片. 该数据集中包含199个实例, 其中89个为正例(高速公路), 110个为负例(街道). 为获得每个实例的多噪声标记集, 共有50个众包工人标注了395个标记. 需要注意的是, 在本文的实验中, 针对LabelMe数据集, 首先使用文献[23]中图像处理的方法将图像数据转换为文本词向量, 然后通过文本分类器对数据集进行训练和预测. 其中, 使用MNB (Multinomial naive Bayes)算法[24]作为PL、STC、CC、AVNC和CENC的基分类器, 而STLNC的基分类器则替换为MNB、CNB (Complement naive Bayes)[25]和OVA (One-versus-all-but-one)[25].

    表 10  8个真实众包数据的详细描述
    Table 10  Description of eight real-world crowdsourced datasets
    数据集分类任务#Instances#Positives#Negatives#Labelers#Labels
    Leaves1maple/alder1429646701093
    Leaves2maple/tilia1409644741044
    Leaves3alder/eucalyptus93464758407
    Leaves4alder/poplar89464354400
    LabelMe1highway/street1998911050395
    LabelMe2highway/forest2278913854476
    LabelMe3highway/opencountry2408915154375
    LabelMe4highway/insidecity2058911649339
    下载: 导出CSV 
    | 显示表格

    图3 ~ 6分别展示了各个算法在8个真实众包数据集上纠正后的噪声比和模型质量的详细对比结果. 从对比结果可以看出: 在Leaves数据集上, STLNC算法的效果整体优于MV、PL、STC、CC、AVNC以及CENC算法, 而在LabelMe数据集上, STLNC算法的优势更加显著. 因此, 通过上述实验的比较, STLNC同时提高了数据集的标记质量和模型质量, 验证了该算法在真实众包场景的有效性.

    图 3  Leaves数据集上的噪声比对比结果
    Fig. 3  Noise ratio comparisons on Leaves datasets
    图 4  Leaves数据集上的模型质量结果对比结果
    Fig. 4  Model quality comparisons on Leaves datasets
    图 5  LabelMe数据集上的噪声比对比结果
    Fig. 5  Noise ratio comparisons on LabelMe datasets
    图 6  LabelMe数据集上的模型质量对比结果
    Fig. 6  Model quality comparisons on LabelMe datasets

    为了分析过滤器的选取对STLNC算法的影响, 本文选取了传统机器学习领域经典的噪声过滤器CF、IPF (Iterative partitioning filtering)[26]和MVF (Majority vote filter)[27], 三个过滤器具体描述如下:

    1) CF: CF的思想是将整个数据集分为大小相等的$n $个子集, 对于其中的每一个子集, 都将剩余的$n-1 $个子集合并作为基分类器的训练集并进行训练, 然后对该子集中的实例进行预测, 若预测标记和集成标记不同, 则该实例为噪声, 并从整个数据集中去除. 上述步骤重复$n $次, 直到整个数据集的实例被预测. 在本实验中CF的$n $值设置为10, 基分类器使用MNB.

    2) IPF: IPF的思想将整个数据集分为大小相等的$n $个子集, 在每个子集上训练一个分类器来对数据集中每个实例进行预测, 再根据投票策略判断实例是否为噪声, 如果是则移出数据集. 执行上述的步骤直到筛选出的噪声实例的比例小于阈值时停止. 在本实验中IPF的$n $为5, 阈值为0.01, 投票策略为多数投票, 所使用的基分类器为MNB.

    3) MVF: MVF的思想是将整个数据集分为大小相等的$n $个子集, 对于其中的每一个子集, 都将剩余的$n-1 $个子集合并作为$ m$个基分类器的训练集并进行训练, 然后对该子集中的实例进行预测, 再根据多数投票的策略判断该实例是否为噪声. 上述步骤重复$n $次, 直到整个数据集的实例被预测. 在本实验中MVF的$n $值设置为10, $m $值设置为3, 所使用的基分类器为MNB、CNB和OVA.

    根据以上噪声过滤器设置, 本文随机选取了真实众包数据集LabelMe4进行了实验, 并评估了使用不同过滤器的STLNC在纠正后的数据集上的噪声比和训练模型质量.

    图7(a)图7(b)分别展示了使用各个过滤器的STLNC算法在LabelMe4上的噪声比和模型质量的详细对比结果. 从对比结果可以看出: 使用CF、IPF和MVF过滤器的STLNC在噪声比和模型质量上结果相似. 由此分析可得, STLNC在选用不同过滤器时表现稳定, 对算法性能影响不大.

    图 7  STLNC基于不同过滤器在LabelMe4数据集上的实验结果
    Fig. 7  Experimental results of STLNC with different filters on LabelMe4 dataset

    在第2节中, 本文介绍了在构建空间结构关系阶段和实例选择与纠正阶段的主要改进为: 1)使用众包数据信息对密度峰值聚类进行加权, 构建空间结构关系; 2)使用集成分类器, 同时设计实例选择和纠正策略. 为了详细说明STLNC算法的这两个核心阶段的重要性, 本文还在LabelMe4上进行了消融实验. 消融实验对比算法说明如下:

    1) STLNC-(1): 表示该算法与STLNC相比, 未使用加权的密度峰值聚类构建空间结构关系, 只使用原始的密度峰值聚类算法.

    2) STLNC-(2): 表示该算法与STLNC相比, 未使用集成分类器和实例选择与纠正策略. 只使用单个分类器MNB对噪声实例进行预测, 将预测后的标记就作为纠正的结果.

    3) STLNC-(1)(2): 表示该算法与STLNC相比, 未使用加权的密度峰值聚类构建空间结构关系以及未使用集成分类器和实例选择与纠正策略. 只使用原始的密度峰值聚类算法构建空间结构关系, 使用了单个分类器MNB对噪声实例进行预测, 将预测后的标记就作为纠正的结果.

    图8(a)图8(b)中可以看出, STLNC算法在纠正后数据集上的噪声比和训练的模型质量表现最好, STLNC-(1)和STLNC-(2)算法表现其次, STLNC-(1)(2)的算法效果最差. 从而可以得出结论: STLNC中两个核心阶段的改进均能提高STLNC算法的性能.

    图 8  STLNC在LabelMe4数据集上的消融实验结果
    Fig. 8  Results of STLNC ablation experiment on LabelMe4 dataset

    本文针对众包数据集经过标记集成后仍然存在标记噪声问题, 提出了一种基于自训练的众包标记噪声纠正算法STLNC. 主要创新包括: 1) 在构建空间结构关系阶段, 本文将实例的多噪声标记集转化为标记分布, 根据标记分布信息改进了密度峰值聚类算法, 从而精确地构建数据集中低密度实例指向高密度实例的空间结构关系; 2) 在实例选择与纠正阶段, 根据构建的空间结构关系, 设计了噪声实例选择策略, 同时引入集成学习的思想和标记置信度计算, 设计了新的实例纠正策略, 提高了纠正的准确性.

    根据在22个仿真的标准数据集以及8个真实的众包数据集上的实验结果, STLNC与PL、STC、CC、AVNC、CENC五种目前最先进的噪声纠正算法相比, 其性能在数据集噪声比和模型质量两个度量指标上更好, 从而验证了所提出的众包标记噪声纠正算法的有效性和优越性. 但是, 本文方法在算法的过滤阶段并未进行深入的研究, 如何获取更加精确的干净集和噪声集仍然是未来将要面临的难题之一. 因此, 下一步工作将围绕设计更好的过滤策略, 提高过滤效果展开.

  • 图  1  复材产品制造工艺的生产流程图

    Fig.  1  Production flow chart of the manufacturing process of composite products

    图  2  机械臂铺片的铺放形式

    Fig.  2  Laying forms of robot arm lay-up sheets

    图  3  拾取末端设计举例

    Fig.  3  Examples of pick-up end effector design

    图  4  铺放末端设计举例

    Fig.  4  Examples of lay-up end effector design

    图  5  单机械臂铺层研究案例

    Fig.  5  Study cases of single robot arm lay-up

    图  6  面向工业生产的多机械臂协同铺层研究案例

    Fig.  6  Multi-robot arms collaborative lay-up study cases for industrial production

    图  7  铺带 (丝) 头结构简图[16]

    Fig.  7  Simplified diagram of the structure of tape (fiber) lay-up head[16]

    图  8  红外热成像检测, 经许可转载自文献 [187], ©Elsevier, 2021

    Fig.  8  Infrared thermal imaging detection, reproduced with permission from reference [187], ©Elsevier, 2021

    表  1  机械臂在传统工业场景和复材产品制造场景应用特点对比

    Table  1  Comparison of the application characteristics of robot arm in traditional industrial scenario and composite products manufacturing scenario

    对比特点传统工业场景复材产品制造场景
    喷涂点焊搬运装配铺片铺带(丝)
    相同之处重复定位精度
    位置跟踪要求
    不同之处操作是否接触
    操作材料特性气体高温硬质硬质柔软粘性柔软粘性
    是否需要加热
    是否有接触力
    末端构造喷嘴焊钳夹持各类工具夹持悬垂拾取专有铺放头
    下载: 导出CSV

    表  2  不同拾取原理的优劣对比

    Table  2  Comparison of the advantages and disadvantages of different pick-up principles

    拾取原理对材料的损坏程度成本实现难度易操作性
    针刺
    低温
    真空吸取
    下载: 导出CSV

    表  3  单机械臂铺层研究案例对比

    Table  3  Comparison of single robot arm lay-up study cases

    研究机构研究重点路径规划运动规划工艺参数系统软件使用的机械臂相关文献
    德国宇航中心全过程自动化 基于视觉生成 系统生成 未知 独立开发 KUKA[4546, 7172]
    汉堡科技大学 工艺流程优化 未知 未知 未知 未知 ABB[7376]
    慕尼黑工业大学 全过程自动化 人类专家设计 控制器生成 人类专家设计 CFK-Tex.Office KUKA KR-500[3234, 77]
    布里斯托大学 铺放自动化 未知 未知 人类专家设计 未知 ABB[62]
    德国宇航中心 全过程自动化 系统生成 系统生成 未知 独立开发 KUKA[7879]
    南丹麦大学 铺放自动化 基于模拟方法 系统生成 未知 独立开发 KUKA KR-360[60, 8084]
    下载: 导出CSV

    表  4  多机械臂协同铺层研究案例对比

    Table  4  Comparison of multi-robot arms collaborative lay-up study cases

    研究机构机械臂数量研究内容路径规划运动规划系统软件使用的机械臂相关文献
    南卡罗莱纳大学3路径规划 运动规划算法生成控制器生成独立开发KUKA-iiwa[8587, 8991]
    斯图加特大学3系统搭建 路径规划人类专家设计系统生成独立开发ABB[64, 92]
    德国宇航中心2系统搭建 路径规划算法生成系统生成独立开发KUKA-KR270[93101]
    空客集团2系统搭建 末端开发人类专家设计系统生成独立开发KUKA[4142, 106]
    林雪平大学2技术验证 末端开发未知未知未知KUKA-KR10, ABB[107108]
    慕尼黑工业大学2系统搭建 路径规划算法生成系统生成独立开发Staubli, KUKA[24]
    思克莱德大学1技术验证人类专家设计系统生成独立开发KUKA-KR6[110]
    维也纳技术大学2技术验证人类专家设计系统生成未知自制[111112]
    下载: 导出CSV

    表  5  铺带(丝)头中采用的切割方式对比

    Table  5  Comparison of cutting methods used in tape (fiber) lay-up heads

    切割方式成本优点缺点
    机械道具切割结构简单, 切割效率高, 适用于多种复杂环境,
    维修更换比较方便
    难以控制切割深度且切口毛糙, 损伤预浸料,
    无法保证切口质量
    激光切割较高切割效率高, 非接触式切割, 产品边缘光滑平整,
    激光对位精准, 切割精度高
    温度较高, 使复合材料发生变质且
    切割深度不易控制
    水喷射切割设备结构简单, 操作容易, 工作机构具有喷头体积小、
    后坐力小、移动方便、生产效率高等特点
    给整个铺带环境带来大量污染液体,
    影响复合材料成型, 铺带工作不便
    超声波切割较高切割效率高, 切口平整; 合适的切割速度、
    切割深度满足不同工况下的切割
    易受负载、温度等因素影响, 引起谐振频率、
    等效阻抗等参数漂移变化
    下载: 导出CSV

    表  6  铺带(丝)头中采用的加热方式对比

    Table  6  Comparison of heating methods used in tape (fiber) lay-up heads

    加热方式成本优点缺点
    电阻丝加热加热均匀, 实现简单热损失大, 功率密度低, 使用寿命短
    激光加热激光加热效率高, 响应快温度难以控制, 容易产生局部过热
    热风加热温度场均匀, 调节范围广加热升温时间长, 热效率较低
    红外加热热效率高, 加热均匀, 响应速度快辐射面存在一定限制, 温度场不均匀
    下载: 导出CSV

    表  7  路径规划方法对比

    Table  7  Comparison of path planning methods

    分类方法优点缺点
    参考路径生成自然路径法可以避免纤维起皱, 轨迹可铺放性良好计算量大, 仅适用于低曲率表面
    定角度路径法原理及计算过程简单仅适用于整体曲率波动较小的曲面
    变角度路径法能够自适应芯模曲面不规则情况算法计算量大
    路径密化等距偏置算法算法简单, 能够覆盖整个芯模表面在复杂表面上可能存在间隙和重叠
    等角度算法算法实现简单, 适应各种复杂构件易存在间隙和重叠
    下载: 导出CSV

    表  8  轨迹规划及仿真软件主要功能[179]

    Table  8  Main functions of path planning and simulation software[179]

    关键技术研究内容研究目标
    轨迹规划根据构件3D表面设计相应路径规划算法, 自适应生成铺放轨迹满足构件结构的方向性、铺放顺序和铺叠层数要求
    铺放路径覆盖根据曲面上相邻路径的间距, 对铺丝路径的覆盖性进行检验与优化实现对模具的满覆盖、不重叠, 满足空隙容差
    边界处理根据构件的边界轮廓信息, 设计边界处理算法, 控制边缘和角部的铺放方式与形态 确保铺放边界质量和表面光洁度
    后置处理数控代码生成、代码优化与合成、加工仿真技术等机器人能够识别执行的指令
    下载: 导出CSV

    表  9  现有自动化缺陷检测技术优劣对比

    Table  9  Comparison of the advantages and disadvantages of existing automated defect detection technologies

    检测技术 使用设备 安装方式 优点 缺点 相关文献
    激光辅助检测 激光投影仪 固定支架安装 实时性好、精度高、分辨率高 投影仪与模具间的相对位置精度要求高,
    对效率提升不明显
    [189191]
    红外热成像检测 热成像仪 集成在铺放头 检测成本低 对环境温度要求严格, 精度难以保证 [192195]
    基于轮廓数据检测 激光轮廓仪 集成在铺放头或安装在
    机械臂末端
    检测结果准确, 不易受环境影响 计算量大, 需要极高性能计算平台,
    仅能检测外部缺陷
    [196200]
    机器视觉检测 工业相机 集成在铺放头 检测效果好, 自动化程度高 检测系统适应性不高 [201212]
    下载: 导出CSV
  • [1] 林刚, CINTE21. 构建“硬科技”优势——2021全球碳纤维复合材料市场报告. 纺织科学研究, 2022, 33(5): 46−66

    Lin Gang, CINTE21. Building a “hard technology” advantage-global carbon fibre composites market report 2021. Textile Science Research, 2022, 33(5): 46−66
    [2] 中国电子学会. 中国机器人产业发展报告(2022年), 世界机器人大会, 中国, 2022.

    Chinese Institute of Electronics. China Robotics Industry Development Report (2022), World Robot Conference, China, 2022.
    [3] 亿欧智库. 2022中国工业机器人市场研究报告. 机器人产业, 2022, (4): 83−95 doi: 10.3969/j.issn.2096-0182.2022.04.014

    EqualOcean. 2022 China industrial robotics market research report. Robot Industry, 2022, (4): 83−95 doi: 10.3969/j.issn.2096-0182.2022.04.014
    [4] Lei T, Rong Y M, Wang H, Huang Y, Li M. A review of vision-aided robotic welding. Computers in Industry, 2020, 123: Article No. 103326 doi: 10.1016/j.compind.2020.103326
    [5] Sun Y W, Jia J J, Xu J T, Chen M S, Niu J B. Path, feedrate and trajectory planning for free-form surface machining: A state-of-the-art review. Chinese Journal of Aeronautics, 2022, 35(8): 12−29 doi: 10.1016/j.cja.2021.06.011
    [6] Urhal P, Weightman A, Diver C, Bartolo P. Robot assisted additive manufacturing: A review. Robotics and Computer-Integrated Manufacturing, 2019, 59: 335−345 doi: 10.1016/j.rcim.2019.05.005
    [7] Cong Y, Chen R H, Ma B T, Liu H S, Hou D D, Yang C G. A comprehensive study of 3-D vision-based robot manipulation. IEEE Transactions on Cybernetics, 2023, 53(3): 1682−1698 doi: 10.1109/TCYB.2021.3108165
    [8] 秦方博, 徐德. 机器人操作技能模型综述. 自动化学报, 2019, 45(8): 1401−1418

    Qin Fang-Bo, Xu De. Review of robot manipulation skill models. Acta Automatica Sinica, 2019, 45(8): 1401−1418
    [9] 曾超, 杨辰光, 李强, 戴诗陆. 人−机器人技能传递研究进展. 自动化学报, 2019, 45(10): 1813−1828

    Zeng Chao, Yang Chen-Guang, Li Qiang, Dai Shi-Lu. Research progress on human-robot skill transfer. Acta Automatica Sinica, 2019, 45(10): 1813−1828
    [10] 齐志刚, 黄攀峰, 刘正雄, 韩冬. 空间冗余机械臂路径规划方法研究. 自动化学报, 2019, 45(6): 1103−1110

    Qi Zhi-Gang, Huang Pan-Feng, Liu Zheng-Xiong, Han Dong. Research on path planning method of spatial redundant manipulator. Acta Automatica Sinica, 2019, 45(6): 1103−1110
    [11] Chutima P. A comprehensive review of robotic assembly line balancing problem. Journal of Intelligent Manufacturing, 2022, 33(1): 1−34 doi: 10.1007/s10845-020-01641-7
    [12] Rajak D K, Pagar D D, Menezes P L, Linul E. Fiber-reinforced polymer composites: Manufacturing, properties, and applications. Polymers, 2019, 11(10): Article No. 1667 doi: 10.3390/polym11101667
    [13] Elkington M, Ward C, Sarkytbayev A. Automated composite draping: A review. In: Proceedings of the SAMPE Seattle 2017. Seattle, USA: SAMPE North America, 2017.
    [14] Björnsson A, Jonsson M, Johansen K. Automated material handling in composite manufacturing using pick-and-place systems——A review. Robotics and Computer-Integrated Manufacturing, 2018, 51: 222−229 doi: 10.1016/j.rcim.2017.12.003
    [15] Lukaszewicz D H J A, Ward C, Potter K D. The engineering aspects of automated prepreg layup: History, present and future. Composites Part B: Engineering, 2012, 43(3): 997−1009 doi: 10.1016/j.compositesb.2011.12.003
    [16] Brasington A, Sacco C, Halbritter J, Wehbe R, Harik R. Automated fiber placement: A review of history, current technologies, and future paths forward. Composites Part C: Open Access, 2021, 6: Article No. 100182 doi: 10.1016/j.jcomc.2021.100182
    [17] Zhang W X, Liu F, Jiang T, Yi M H, Chen W Q, Ding X L. Overview of current design and analysis of potential theories for automated fibre placement mechanisms. Chinese Journal of Aeronautics, 2022, 35(4): 1−13 doi: 10.1016/j.cja.2021.04.018
    [18] Chen J P, Fu K K, Li Y. Understanding processing parameter effects for carbon fibre reinforced thermoplastic composites manufactured by laser-assisted automated fibre placement (AFP). Composites Part A: Applied Science and Manufacturing, 2021, 140: Article No. 106160 doi: 10.1016/j.compositesa.2020.106160
    [19] de Campos A A, Henriques E, Magee C L. Technological improvement rates and recent innovation trajectories in automated advanced composites manufacturing technologies: A patent-based analysis. Composites Part B: Engineering, 2022, 238: Article No. 109888 doi: 10.1016/j.compositesb.2022.109888
    [20] Soares B A R, Henriques E, Ribeiro I, Freitas M. Cost analysis of alternative automated technologies for composite parts production. International Journal of Production Research, 2019, 57(6): 1797−1810 doi: 10.1080/00207543.2018.1508903
    [21] Jayasekara D, Lai N Y G, Wong K H, Pawar K, Zhu Y D. Level of automation (LOA) in aerospace composite manufacturing: Present status and future directions towards industry 4.0. Journal of Manufacturing Systems, 2022, 62: 44−61 doi: 10.1016/j.jmsy.2021.10.015
    [22] Potter K, Ward C. Draping processes for composites manufacture. Advances in Composites Manufacturing and Process Design. Amsterdam: Woodhead Publishing, 2015. 93−109
    [23] Cevotec. Composite tank reinforcements [Online], available: https://www.cevotec.com/industries-applications/fpp-composite-tanks/, July 29, 2023
    [24] Michl F, Coquel M. Fully-automated production of complex CFRP parts using fibre-patch-preforming technology. JEC Com posites Magazine, 2014, 87: 108−110
    [25] Malhan R K, Shembekar A V, Kabir A M, Bhatt P M, Shah B, Zanio S, et al. Automated planning for robotic layup of composite prepreg. Robotics and Computer-Integrated Manufacturing, 2021, 67: Article No. 102020 doi: 10.1016/j.rcim.2020.102020
    [26] Larsen L, Kim J. Path planning of cooperating industrial robots using evolutionary algorithms. Robotics and Computer-Integrated Manufacturing, 2021, 67: Article No. 102053 doi: 10.1016/j.rcim.2020.102053
    [27] Khodunov A A, Bogachev V V, Borodulin A S. Advances in tailored fiber placement technology. Journal of Physics: Conference Series, 2021, 1990: Article No. 012041
    [28] Lindback J E, Björnsson A, Johansen K. New automated composite manufacturing process: Is it possible to find a cost effective manufacturing method with the use of robotic equipment? In: Proceedings of the 5th International Swedish Production Symposium. Linkoping, Sweden: 2012. 523−531
    [29] Kordi M T, Husing M, Corves B. Development of a multifunctional robot end-effector system for automated manufacture of textile preforms. In: Proceedings of the IEEE/ASME International Conference on Advanced Intelligent Mechatronics. Zurich, Switzerland: IEEE, 2007. 1−6
    [30] Reinhart G, Straßer G. Flexible gripping technology for the automated handling of limp technical textiles in composites industry. Production Engineering, 2011, 5(3): 301−306 doi: 10.1007/s11740-011-0306-1
    [31] Reinhart G, Strassr G, Ehinger C. Highly flexible automated manufacturing of composite structures consisting of limp carbon fibre textiles. SAE International Journal of Aerospace, 2010, 2(1): 181−187
    [32] Angerer A, Ehinger C, Hoffmann A, Reif W, Reinhart G. Design of an automation system for preforming processes in aerospace industries. In: Proceedings of the IEEE International Conference on Automation Science and Engineering. Trieste, Italy: IEEE, 2011. 557−562
    [33] Ehinger C, Reinhart G. Robot-based automation system for the flexible preforming of single-layer cut-outs in composite industry. Production Engineering, 2014, 8(5): 559−565 doi: 10.1007/s11740-014-0546-y
    [34] Reinhart G, Ehinger C. Novel robot-based end-effector design for an automated preforming of limb carbon fiber textiles. In: Proceedings of the 1st Conference of the German Academic Society for Production Engineering (WGP). Berlin, Germany: Springer, 2013. 131−142
    [35] Löchte C, Kunz H, Schnurr R, Dietrich F, Raatz A, Dilger K, et al. Form-flexible handling technology for automated preforming. In: Proceedings of the 19th International Conference on Composite Materials. Montreal, Canada: Concordia University, 2013.
    [36] Löchte C, Kunz H, Schnurr R, Langhorst S, Dietrich F, Raatz A, et al. Form-flexible handling and joining technology (FormHand) for the forming and assembly of limp materials. Procedia CIRP, 2014, 23: 206−211 doi: 10.1016/j.procir.2014.10.086
    [37] Kunz H, Löchte C, Dietrich F, Raatz A, Fischer F, Dröder K, et al. Novel form-flexible handling and joining tool for automated preforming. Science and Engineering of Composite Materials, 2015, 22(2): 199−213 doi: 10.1515/secm-2013-0326
    [38] Dröder K, Dietrich F, Löchte C, Hesselbach J. Model based design of process-specific handling tools for workpieces with many variants in shape and material. CIRP Annals, 2016, 65(1): 53−56 doi: 10.1016/j.cirp.2016.04.109
    [39] Apmann H. Automatic Handling of CFRP-material for Frame and Stringer Production, SAE Technical Paper 2008-01-2289, SAE, USA, 2008.
    [40] Apmann H, Hemmen A, Herkt M. Automatic Handling of Carbon Fiber Preforms for CFRP Parts in Aerospace, SAE Technical Paper 2012-01-1864, SAE, USA, 2012.
    [41] Apmann H, Busse M, Du J Y, Köhnke P. Automated manufacture of fibre metal laminates to achieve high rate of production. Lightweight Design Worldwide, 2017, 10(4): 28−33
    [42] Ucan H, Apmann H, Graßl G, Krombholz C, Fortkamp K, Nieberl D, et al. Production technologies for lightweight structures made from fibre-metal laminates in aircraft fuselages. CEAS Aeronautical Journal, 2019, 10(2): 479−489 doi: 10.1007/s13272-018-0330-3
    [43] Ucan H, Scheller J, Nguyen C, Nieberl D, Beumler T, Haschenburger A, et al. Automated, quality assured and high volume oriented production of fiber metal laminates (FML) for the next generation of passenger aircraft fuselage shells. Science and Engineering of Composite Materials, 2019, 26(1): 502−508 doi: 10.1515/secm-2019-0031
    [44] Braun G, Buchheim A, Fischer F, Gerngross T. Handgeführter Endeffektor Für Die Automatisierte Handhabung von Textilen Zuschnitten, DLR-IB 435-2013/88, Deutsches Zentrum für Luft- und Raumfahrt e.V. (DLR), Germany, 2013.
    [45] Kuehnel M, Schuster A, Buchheim A, Gergross T, Kupke M. Automated near-net-shape preforming of carbon fiber reinforced thermoplastics (CFRTP). In: Proceedings of the Paper for the I.C.S of the JEC Europe 2014. Paris, France: 2014.
    [46] Kuhnel M, Schuster A, Rähtz C, Kupke M. Near net shape thermoplastic preforming with continuously automated cutting and robotic pick and place processes. In: Proceedings of the International Conference and Exhibition on Thermoplastic Composites. Bremen, Germany: 2016.
    [47] Vistein M, Faber J, Schmidt-Eisenlohr C, Reiter D. Automated handling of auxiliary materials using a multi-kinematic gripping system. Procedia Manufacturing, 2019, 38: 1276−1283 doi: 10.1016/j.promfg.2020.01.220
    [48] Gunnarsson G G, Nielsen O W, Schlette C, Petersen H G. Fast and simple interacting models of drape tool and ply material for handling free hanging, pre-impregnated carbon fibre material. In: Proceedings of the 15th Informatics in Control, Automation and Robotics. Porto, Portugal: Springer, 2018. 1−25
    [49] Fleischer J, Förster F, Crispieri N V. Intelligent gripper technology for the handling of carbon fiber material. Production Engineering, 2014, 8(6): 691−700 doi: 10.1007/s11740-014-0549-8
    [50] Förster F, Ballier F, Coutandin S, Defranceski A, Fleischer J. Manufacturing of textile preforms with an intelligent draping and gripping system. Procedia CIRP, 2017, 66: 39−44 doi: 10.1016/j.procir.2017.03.370
    [51] Wirth B, Coutandin S, Fleischer J. Disturbance force estimation for a low pressure suction gripper based on differential pressure analysis. In: Proceedings of the Annals of Scientific Society for Assembly, Handling and Industrial Robotics. Berlin, Germany: Springer, 2020. 263−273
    [52] Wirth B, Schwind T, Friedmann M, Fleischer J. Automated stack singulation for technical textiles using sensor supervised low pressure suction grippers. In: Proceedings of the Annals of Scientific Society for Assembly, Handling and Industrial Robotics 2022. Cham, Germany: Springer, 2022. 329−340
    [53] Brink M, Ohlendorf J H, Thoben K D. Development of a handling system with integrated sensors for textile preforms using additive manufacturing. Procedia Manufacturing, 2018, 24: 114−119 doi: 10.1016/j.promfg.2018.06.016
    [54] Björnsson A, Jonsson M, Eklund D, Lindbäck J E, Björkman M. Getting to grips with automated prepreg handling. Production Engineering, 2017, 11: 445−453 doi: 10.1007/s11740-017-0763-2
    [55] Brinker J, Prause I, Kosse P, Früh H C, Printz S, Henke C, et al. Automated handling and draping of reinforcing textiles-challenges and developments. New Advances in Mechanisms, Mechanical Transmissions and Robotics. Cham, Germany: Springer, 2017. 485−493
    [56] Brinker J, Paris J, Müller M, Hüsing M, Corves B. Mechanism type synthesis approach for automated handling and multiaxial draping of reinforcing textiles. New Trends in Mechanism and Machine Science: Theory and Industrial Applications. Cham, Germany: Springer, 2017. 523−532
    [57] Brinker J, Müller M, Paris J, Husing M, Corves B. Mechanism design for automated handling and multiaxial draping of reinforcing textiles. In: Proceedings of the ASME International Design Engineering Technical Conferences and Computers and Information in Engineering Conference. Charlotte, USA: ASME, 2016. Article No. V05BT07A040
    [58] Corves B, Brinker J, Prause I, Hüsing M, Abbas B, Krieger H, et al. AutoHD——Automated handling and draping of reinforcing textiles. In: Proceedings of the Conference on Mechanisms, Transmissions and Applications. Cham, Germany: Springer, 2015. 301−309
    [59] Wang B W. Design and Development of a Soft Robotic Gripper for Fabric Material Handling [Master thesis], University of Windsor, Canada, 2020.
    [60] Krogh C, Jakobsen J, Sherwood J A. Development of a computationally efficient fabric model for optimization of gripper trajectories in automated composite draping. arXiv preprint arXiv: 1810.07619, 2018.
    [61] Elkington M, Bloom D, Ward C, Chatzimichali A, Potter K. Hand layup: Understanding the manual process. Advanced Manufacturing: Polymer & Composites Science, 2015, 1(3): 138−151
    [62] Elkington M, Ward C, Potter K D. Automated layup of sheet prepregs on complex moulds. Journal of Advanced Materials, 2016, 3 : 70−84
    [63] Elkington M, Almas E, Ward-Cherrier B, Pestell N, Lloyd J, Ward C, et al. Real time defect detection during composite layup via tactile shape sensing. Science and Engineering of Composite Materials, 2021, 28(1): 1−10 doi: 10.1515/secm-2021-0001
    [64] Szcesny M, Heieck F, Carosella S, Middendorf P, Sehrschön H, Schneiderbauer M. The advanced ply placement process——An innovative direct 3D placement technology for plies and tapes. Advanced Manufacturing: Polymer & Composites Science, 2017, 3(1): 2−9
    [65] Richrath M, Franke J, Ohlendorf J H, Thoben K D. Effector for automated direct textile placement in rotor blade production. Lightweight Design Worldwide, 2017, 10(5): 42−47 doi: 10.1007/s41777-017-0039-8
    [66] Schouterden G, Cramer J, Demeester E, Kellens K. Development of a membrane-shaped MR-based composite draping tool. Procedia CIRP, 2019, 86: 167−172 doi: 10.1016/j.procir.2020.01.048
    [67] Denkena B, Schmidt C, Werner S, Schwittay D. Development of a shape replicating draping unit for continuous layup of unidirectional non-crimp fabrics on complex surface geometries. Journal of Composites Science, 2021, 5(4): Article No. 93 doi: 10.3390/jcs5040093
    [68] Helber F, Amann A, Carosella S, Middendorf P. Intrinsic fibre heating: A novel approach for automated dry fibre placement. IOP Conference Series: Materials Science and Engineering, 2018, 460: Article No. 012064
    [69] Björnsson A, Lindback J E, Johansen K. Automated removal of prepreg backing paper——A sticky problem. In: Proceedings of the SAE 2013 AeroTech Congress & Exhibition. New York, USA: SAE International, 2013.
    [70] Bruns C, Micke-Camuz M, Bohne F, Raatz A. Process design and modelling methods for automated handling and draping strategies for composite components. CIRP Annals, 2018, 67(1): 1−4 doi: 10.1016/j.cirp.2018.04.014
    [71] Schuster A, Larsen L, Fischer F, Glück R, Schneyer S, Kühnel M, et al. Smart manufacturing of thermoplastic CFRP skins. Procedia Manufacturing, 2018, 17: 935−943 doi: 10.1016/j.promfg.2018.10.147
    [72] Beyrle M, Endraß M, Kühnel M, Schuster A, Stefani T, Glück R, et al. Automated production and joining of high performance structures out of carbon fiber reinforced thermoplastics. In: Proceedings of the Composites and Advanced Materials Expo. Orlando, USA: 2017.
    [73] Roth F, Eschen H, Schüppstuhl T. The loop gripper: A soft gripper for honeycomb materials. Procedia Manufacturing, 2021, 55: 160−167 doi: 10.1016/j.promfg.2021.10.023
    [74] Eschen H, Harnisch M, Schuppstühl T. Flexible and automated production of sandwich panels for aircraft interior. Procedia Manufacturing, 2018, 18: 35−42 doi: 10.1016/j.promfg.2018.11.005
    [75] Eschen H, Kalscheuer F, Schüppstuhl T. Optimized process chain for flexible and automated aircraft interior production. Procedia Manufacturing, 2020, 51: 535−542 doi: 10.1016/j.promfg.2020.10.075
    [76] Kalscheuer F, Eschen H, Schüppstuhl T. Towards semi automated pre-assembly for aircraft interior production. In: Proceedings of the Annals of Scientific Society for Assembly, Handling and Industrial Robotics 2021. Cham, Germany: Springer, 2022. 203−213
    [77] Angerer A, Ehinger C, Hoffmann A, Reif W, Reinhart G, Strasser G. Automated cutting and handling of carbon fiber fabrics in aerospace industries. In: Proceedings of the IEEE International Conference on Automation Science and Engineering. Toronto, Canada: IEEE, 2010. 861−866
    [78] Gerngross T, Nieberl D. Automated manufacturing of large, three-dimensional CFRP parts from dry textiles. CEAS Aeronautical Journal, 2016, 7(2): 241−257 doi: 10.1007/s13272-016-0184-5
    [79] Nägele L, Macho M, Angerer A, Hoffmann A, Vistein M, Schönheits M, et al. A backward-oriented approach for offline programming of complex manufacturing tasks. In: Proceedings of the 6th International Conference on Automation, Robotics and Applications (ICARA). Queenstown, New Zealand: IEEE, 2015. 124−130
    [80] Ellekilde L P, Wilm J, Nielsen O W, Krogh C, Kristiansen E, Gunnarsson G G, et al. Design of automated robotic system for draping prepreg composite fabrics. Robotica, 2021, 39(1): 72−87 doi: 10.1017/S0263574720000193
    [81] Serpina G G G, Petersen H G. Mathematical modeling of a highly underactuated tool for draping fiber plies on double curved molds. In: Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). Xi'an, China: IEEE, 2021. 1666−1672
    [82] Krogh C, Glud J A, Jakobsen J. Modeling of prepregs during automated draping sequences. AIP Conference Proceedings, 2017, 1896: Article No. 030036
    [83] Krogh C, Glud J A, Jakobsen J. Modeling the robotic manipulation of woven carbon fiber prepreg plies onto double curved molds: A path-dependent problem. Journal of Composite Materials, 2019, 53(15): 2149−2164 doi: 10.1177/0021998318822722
    [84] Krogh C, Sherwood J A, Jakobsen J. Generation of feasible gripper trajectories in automated composite draping by means of optimization. Advanced Manufacturing: Polymer & Composites Science, 2019, 5(4): 234−249
    [85] Malhan R K, Kabir A M, Shembekar A V, Shah B, Gupta S K, Centea T. Hybrid cells for multi-layer prepreg composite sheet layup. In: Proceedings of the IEEE 14th International Conference on Automation Science and Engineering (CASE). Munich, Germany: IEEE, 2018. 1466−1472
    [86] Malhan R K, Kabir A M, Shah B, Centea T, Gupta S K. Automated prepreg sheet placement using collaborative robotics. In: Proceedings of the North America Society for the Advancement of Material and Process Engineering (SAMPE) Long Beach Conference. Long Beach, USA: SAMPE, 2018.
    [87] Malhan R K, Kabir A M, Shah B, Gupta S K. Identifying feasible workpiece placement with respect to redundant manipulator for complex manufacturing tasks. In: Proceedings of the International Conference on Robotics and Automation (ICRA). Montreal, Canada: IEEE, 2019. 5585−5591
    [88] Malhan R K, Kabir A M, Shah B, Centea T, Gupta S K. Determining feasible robot placements in robotic cells for composite prepreg sheet layup. In: Proceedings of the ASME 14th International Manufacturing Science and Engineering Conference. Erie, USA: ASME, 2019. Article No. V001T02A025
    [89] Manyar O M, Desai J, Deogaonkar N, Joesph R J, Malhan R, McNulty Z, et al. A simulation-based grasp planner for enabling robotic grasping during composite sheet layup. In: Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). Xi'an, China: IEEE, 2021. 930−937
    [90] Malhan R K, Joseph R J, Shembekar A V, Kabir A M, Bhatt P M, Gupta S K. Online grasp plan refinement for reducing defects during robotic layup of composite prepreg sheets. In: Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). Paris, France: IEEE, 2020. 11500−11507
    [91] Malhan R K, Thakar S, Kabir A M, Rajendran P, Bhatt P M, Gupta S K. Generation of configuration space trajectories over semi-constrained cartesian paths for robotic manipulators. IEEE Transactions on Automation Science and Engineering, 2023, 20(1): 193−205 doi: 10.1109/TASE.2022.3144673
    [92] Helber F, Carosella S, Middendorf P. Multi-robotic composite production of complex and large-scaled components for the automotive industry. In: Proceedings of the Advances in Automotive Production Technology——Theory and Application. Berlin, Germany: Springer, 2021. 369−376
    [93] Krebs F, Larsen L, Braun G, Dudenhausen W. Design of a multifunctional cell for aerospace CFRP production. The International Journal of Advanced Manufacturing Technology, 2016, 85: 17−24 doi: 10.1007/s00170-014-6022-1
    [94] Tekles N, Reiner M, Krebs F. Model-based elastic deformation compensation for a multi-robot work cell. In: Proceedings of the IEEE 15th International Conference on Control and Automation (ICCA). Edinburgh, UK: IEEE, 2019. 530−536
    [95] Tekles N, Krebs F, Reiner M. Inverse model command shaper for a flexible gantry robot. In: Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). Vancouver, Canada: IEEE, 2017. 1636−1642
    [96] Eckardt M, Buchheim A, Gerngross T. Investigation of an automated dry fiber preforming process for an aircraft fuselage demonstrator using collaborating robots. CEAS Aeronautical Journal, 2016, 7(3): 429−440 doi: 10.1007/s13272-016-0199-y
    [97] Schuster A, Kupke M, Larsen L. Autonomous manufacturing of composite parts by a multi-robot system. Procedia Manufacturing, 2017, 11: 249−255 doi: 10.1016/j.promfg.2017.07.238
    [98] Brandt L, Eckardt M. Automated handling and positioning of large dry carbon fibre cut-pieces with cooperating robots in rear pressure bulkhead production. In: Proceedings of the CEAS Conference. Bucharest, Romania: CEAS, 2017.
    [99] Schuster A, Frommel C, Deden D, Brandt L, Eckardt M, Glück R, et al. Simulation based draping of dry carbon fibre textiles with cooperating robots. Procedia Manufacturing, 2019, 38: 505−512 doi: 10.1016/j.promfg.2020.01.064
    [100] Deden D, Frommel C, Glück R, Larsen L, Malecha M, Schuster A. Towards a fully automated process chain for the lay-up of large carbon dry-fibre cut pieces using cooperating robots. In: Proceedings of the SAMPE Europe Conference. Nantes, France: SAMPE, 2019.
    [101] Frommel C, Krebs F, Haase T, Vistein M, Schuster A, Larsen L, et al. Automated manufacturing of large composites utilizing a process orchestration system. Procedia Manufacturing, 2020, 51: 470−477 doi: 10.1016/j.promfg.2020.10.066
    [102] Larsen L, Kim J, Kupke M. Intelligent path panning towards collision-free cooperating industrial robots. In: Proceedings of the 11th International Conference on Informatics in Control. Vienna, Austria: SciTePress, 2014.
    [103] Angerer A, Hoffmann A, Larsen L, Vistein M, Kim J, Kupke M, et al. Planning and execution of collision-free multi-robot trajectories in industrial applications. In: Proceedings of the 47th International Symposium on Robotics. Munich, Germany: VDE, 2016. 1−7
    [104] Larsen L, Pham V L, Kim J, Kupke M. Collision-free path planning of industrial cooperating robots for aircraft fuselage production. In: Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). Seattle, USA: IEEE, 2015. 2042−2047
    [105] Larsen L, Kaspar M, Schuster A, Vistein M, Kim J, Kupke M. Full automatic path planning of cooperating robots in industrial applications. In: Proceedings of the 13th IEEE Conference on Automation Science and Engineering (CASE). Xi'an, China: IEEE, 2017. 523−530
    [106] Vistein M, Deden D, Glück R, Schneyer S. Automated production of large fibre metal laminate aircraft structure parts. Procedia Manufacturing, 2019, 38: 1300−1307 doi: 10.1016/j.promfg.2020.01.160
    [107] Björnsson A, Jonsson M, Johansen K. Automation of composite manufacturing using off-the-shelf solutions, three cases from the aerospace industry. In: Proceedings of the 20th International Conference on Composite Materials. Copenhagen, Denmark: Aalborg University, 2015.
    [108] Björnsson A, Lindback J E, Eklund D, Jonsson M. Low-cost automation for prepreg handling-two cases from the aerospace industry. SAE International Journal of Materials and Manufacturing, 2016, 9(1): 68−74
    [109] Zhang X W, Chi X F, Ji C C. Discrete path planning of carbon fiber patch placement with complex surface. Textile Research Journal, 2023, 93(17−18): 4010−4022 doi: 10.1177/00405175231169045
    [110] Yang M M, Yu L J, Wong C B, Mineo C, Yang E F, Bomphray I, et al. A cooperative mobile robot and manipulator system (Co-MRMS) for transport and lay-up of fibre plies in modern composite material manufacture. The International Journal of Advanced Manufacturing Technology, 2022, 119: 1249−1265 doi: 10.1007/s00170-021-08342-2
    [111] Flixeder S, Glück T, Kugi A. Force-based cooperative handling and lay-up of deformable materials: Mechatronic design, modeling, and control of a demonstrator. Mechatronics, 2017, 47: 246−261 doi: 10.1016/j.mechatronics.2016.10.003
    [112] Hartl-Nesic C, Glück T, Kugi A. Surface-based path following control: Application of curved tapes on 3-D objects. IEEE Transactions on Robotics, 2021, 37(2): 615−626 doi: 10.1109/TRO.2020.3033721
    [113] Belhaj M, Dodangeh A, Hojjati M. Experimental investigation of prepreg tackiness in automated fiber placement. Composite Structures, 2021, 262: Article No. 113602 doi: 10.1016/j.compstruct.2021.113602
    [114] Zhao F H, Liu Z Q, Chen R F, Hao Y, Ma Z H. The effect of temperature field on the characteristics of carbon fiber reinforced thermoplastic composites in the laying and shaping process. The International Journal of Advanced Manufacturing Technology, 2022, 121: 7569−7589 doi: 10.1007/s00170-022-09795-9
    [115] de Sá Rodrigues J, Gonçalves P T, Pina L, Gomes de Almeida F. Modelling the heating process in the transient and steady state of an in situ tape-laying machine head. Journal of Manufacturing and Materials Processing, 2022, 6(1): Article No. 8 doi: 10.3390/jmmp6010008
    [116] Zhang P, Sun R L, Zhao X Y, Hu L J. Placement suitability criteria of composite tape for mould surface in automated tape placement. Chinese Journal of Aeronautics, 2015, 28(5): 1574−1581 doi: 10.1016/j.cja.2015.06.002
    [117] 赵尧旭. 热塑性复合材料机器人铺放设备及工艺研究 [硕士学位论文], 哈尔滨工业大学, 中国, 2019.

    Zhao Yao-Xu. Research on Robotic Placement Machine and Placement Process of Thermoplastic Composite [Master thesis], Harbin Institute of Technology, China, 2019.
    [118] 解五一. 复合材料自动铺带机器人设计及铺带过程控制 [硕士学位论文], 武汉大学, 中国, 2019.

    Xie Wu-Yi. Design of Automatic Composite Tape Laying Robot and Process Control of Taping [Master thesis], Wuhan University, China, 2019.
    [119] Han Z Y, Hu F C, Lu H, Fu H Y. Design of ultrasonic cutting bench for automatic tape laying machine. Applied Mechanics and Materials, 2014, 552: 80−85 doi: 10.4028/www.scientific.net/AMM.552.80
    [120] Grimshaw M N, Grant C M C G, Diaz C M J M L. Advanced technology tape laying for affordable manufacturing of large composite structures. In: Proceedings of the 46th International SAMPE Symposium. Long Beach, USA: SAMPE, 2001. 2484−2494
    [121] Budelmann D, Detampel H, Schmidt C, Meiners D. Interaction of process parameters and material properties with regard to prepreg tack in automated lay-up and draping processes. Composites Part A: Applied Science and Manufacturing, 2019, 117: 308−316 doi: 10.1016/j.compositesa.2018.12.001
    [122] Ren S L, Lu H, Wang Y Z, Fu H Y. Development of PLC-based tension control system. Chinese Journal of Aeronautics, 2007, 20(3): 266−271 doi: 10.1016/S1000-9361(07)60043-0
    [123] Izco L, Isturiz J, Motilva M. High Speed Tow Placement System for Complex Surfaces with Cut/Clamp/ & Restart Capabilities at 85 m/min (3350 IPM), SAE Technical Paper 2006-01-3138, Aerospace Manufacturing and Automated Fastening Conference and Exhibition, USA, 2006.
    [124] Denkena B, Schmidt C, Weber P. Automated fiber placement head for manufacturing of innovative aerospace stiffening structures. Procedia Manufacturing, 2016, 6: 96−104 doi: 10.1016/j.promfg.2016.11.013
    [125] Singh Y, Singh J, Sharma S, Sharma A, Singh Chohan J. Process parameter optimization in laser cutting of coir fiber reinforced epoxy composite——A review. Materials Today: Proceedings, 2022, 48: 1021−1027 doi: 10.1016/j.matpr.2021.06.344
    [126] Liu X L, Han Z Y, Zhao Z Y, Sun S Z. Thermal analysis of cylindrical molds using thermoplastic composite during robotic fiber placement. Journal of Physics: Conference Series, 2021, 1765: Article No. 012023
    [127] 蒋威, 周悦, 杨飞, 黄志高, 陈诚, 周华民. 红外辅助自动纤维铺放工艺对连续玻璃纤维增强聚丙烯复合材料结构与性能影响. 复合材料学报, 2023, 40(4): 2015−2025

    Jiang Wei, Zhou Yue, Yang Fei, Huang Zhi-Gao, Chen Cheng, Zhou Hua-Min. Infrared-assisted automated fiber placement process on the structure and properties of continuous glass fiber reinforced polypropylene composites. Acta Materiae Compositae Sinica, 2023, 40(4): 2015−2025
    [128] 杨宇. 门式六通道纤维铺放系统的研究与装备实现 [硕士学位论文], 武汉理工大学, 中国, 2019.

    Yang Yu. Research and Equipment Realization of Gantry Type Six-channel Fiber Laying System [Master thesis], Wuhan University of Technology, China, 2019.
    [129] Sabido A M. Development of an Automated Fiber Placement Process for the Fabrication of Thermoplastic Composite Laminates [Master thesis], University of South Carolina, USA, 2019.
    [130] Zhang W X, Liu F, Lv Y X, Ding X L. Modelling and layout design for an automated fibre placement mechanism. Mechanism and Machine Theory, 2020, 144: Article No. 103651 doi: 10.1016/j.mechmachtheory.2019.103651
    [131] Hauber D E, Langone R J, Martin J P, Miller S F, Pasanen M J. Composite Tape Laying Apparatus and Method, U.S. Patent 7063118, June 2006
    [132] Mischler P L, Tingley M C, Hoffmann K. Compaction Roller for A Fiber Placement Machine, U.S. Patent 7810539, October 2010
    [133] 王磊. 纱架与铺丝头一体化纤维铺放系统研究 [硕士学位论文], 哈尔滨工业大学, 中国, 2015.

    Wang Lei. Research on Integration Technique of Creels and Fiber Placment Head for Automated Fiber Placement Machine [Master thesis], Harbin Institute of Technology, China, 2015.
    [134] Liu F, Zhang W X, Shang J F, Yi M H, Wang S R, Ding X L. A planar underactuated compaction mechanism with self-adaptability for automated fiber placement heads. Aerospace, 2022, 9(10): Article No. 586 doi: 10.3390/aerospace9100586
    [135] Saboukhi A. Designing and Implementing a Small-size Automated Fiber Placement (AFP) Head Capable of Depositing Thermoset Layers on V-shape Structures [Master thesis], Concordia University, Canada, 2023.
    [136] Li L N, Xu D, Wang X G, Tan M. A survey on path planning algorithms in robotic fibre placement. In: Proceedings of the 27th Chinese Control and Decision Conference (CCDC). Qingdao, China: IEEE, 2015. 4704−4709
    [137] Rousseau G, Wehbe R, Halbritter J, Harik R. Automated fiber placement path planning: A state-of-the-art review. Computer-Aided Design & Applications, 2018, 16(2): 172−203
    [138] Shinno N, Shigematsu T. Method for Controlling Tape Affixing Direction of Automatic Tape Affixing Apparatus, U.S. Patent 5041179, August 1991
    [139] 尹书云. 自由型面自动铺丝线型规划约束研究 [硕士学位论文], 武汉理工大学, 中国, 2013.

    Yin Shu-Yun. Research on Constraint for Fibers Arrangement Pattern Planning in Auto Fiber Placement on Free-surface [Master thesis], Wuhan University of Technology, China, 2013.
    [140] Zhang J F, Xu D L, Wang Z H. Modeling approach for fiber placement routes on complex surface. Applied Mechanics and Materials, 2014, 686: 560−566 doi: 10.4028/www.scientific.net/AMM.686.560
    [141] Xiao H, Han W, Tang W B, Duan Y G. An efficient and adaptable path planning algorithm for automated fiber placement based on meshing and multi guidelines. Materials, 2020, 13(18): Article No. 4209 doi: 10.3390/ma13184209
    [142] Hély C, Birglen L, Xie W F. Feasibility study of robotic fibre placement on intersecting multi-axial revolution surfaces. Robotics and Computer-Integrated Manufacturing, 2017, 48: 73−79 doi: 10.1016/j.rcim.2017.02.005
    [143] Qu W W, Gao J X, Yang D, He R M, Yang Q, Cheng L, et al. Automated fiber placement path generation method based on prospective analysis of path performance under multiple constraints. Composite Structures, 2021, 255: Article No. 112940 doi: 10.1016/j.compstruct.2020.112940
    [144] 张鹏, 尹来容, 周振华, 黄龙. 基于近似测地线的分层次自动铺带轨迹规划方法. 机械工程学报, 2020, 56(23): 226−238 doi: 10.3901/JME.2020.23.226

    Zhang Peng, Yin Lai-Rong, Zhou Zhen-Hua, Huang Long. A multi-level trajectory planning method based on quasi-geodesic for automated tape placement. Journal of Mechanical Engineering, 2020, 56(23): 226−238 doi: 10.3901/JME.2020.23.226
    [145] Punera D, Mukherjee P. Recent developments in manufacturing, mechanics, and design optimization of variable stiffness composites. Journal of Reinforced Plastics and Composites, 2022, 41(23−24): 917−945 doi: 10.1177/07316844221082999
    [146] Parnas L, Oral S, Ceyhan Ü. Optimum design of composite structures with curved fiber courses. Composites Science and Technology, 2003, 63(7): 1071−1082 doi: 10.1016/S0266-3538(02)00312-3
    [147] IJsselmuiden S T, Abdalla M M, Gurdal Z. Optimization of variable-stiffness panels for maximum buckling load using lamination parameters. AIAA Journal, 2010, 48(1): 134−143 doi: 10.2514/1.42490
    [148] Blom A W, Tatting B F, Hol J M A M, Gürdal Z. Fiber path definitions for elastically tailored conical shells. Composites Part B: Engineering, 2009, 40(1): 77−84 doi: 10.1016/j.compositesb.2008.03.011
    [149] 李玥华, 富宏亚, 韩振宇, 韩德东. 两类非可展曲面零件自动纤维铺放变角度轨迹规划算法. 计算机辅助设计与图形学学报, 2013, 25(10): 1523−1529

    Li Yue-Hua, Fu Hong-Ya, Han Zhen-Yu, Han De-Dong. Variable-angle trajectory planning algorithm for automated fiber placement of two non-developable surfaces. Journal of Computer-Aided Design & Computer Graphics, 2013, 25(10): 1523−1529
    [150] 李玥华. 热塑性预浸丝变角度铺放及其轨迹规划的研究 [博士学位论文], 哈尔滨工业大学, 中国, 2013.

    Li Yue-Hua. Research on Thermoplastic Towpreg Variable Angle Placement and Trajectory Planning [Ph.D. dissertation], Harbin Institute of Technology, China, 2013.
    [151] Shirinzadeh B, Cassidy G, Oetomo D, Alici G, Ang Jr M H. Trajectory generation for open-contoured structures in robotic fibre placement. Robotics and Computer-Integrated Manufacturing, 2007, 23(4): 380−394 doi: 10.1016/j.rcim.2006.04.006
    [152] Schueler K, Miller J, Hale R. Approximate geometric methods in application to the modeling of fiber placed composite structures. Journal of Computing and Information Science in Engineering, 2004, 4(3): 251−256 doi: 10.1115/1.1736685
    [153] Zhu Y J, Yao K Z. Optimization path planning algorithm based on STL file reconstruction for automated fiber placement. In: Proceedings of the Chinese Intelligent Systems Conference. Singapore: Springer, 2019. 379−387
    [154] Li L, Wang X G, Xu D, Tan M. A placement path planning algorithm based on meshed triangles for carbon fiber reinforce composite component with revolved shape. International Journal on Control Systems and Applications, 2014, 1(1): 23−32
    [155] 王小平, 周宇, 刘付国. 三角网格面自动铺丝定角度路径规划算法. 南京航空航天大学学报, 2020, 52(3): 378−387

    Wang Xiao-Ping, Zhou Yu, Liu Fu-Guo. Fixed-angle method for automatic fiber placement on triangular mesh surface. Journal of Nanjing University of Aeronautics & Astronautics, 2020, 52(3): 378−387
    [156] Bruyneel M, Zein S. A modified fast marching method for defining fiber placement trajectories over meshes. Computers & Structures, 2013, 125: 45−52
    [157] 刘志强, 顾献安, 郭昊, 奚浩, 王明强, 李军利. 碳纤维螺旋桨自动铺放成形轨迹规划方法. 中国机械工程, 2020, 31(17): 2079−2084 doi: 10.3969/j.issn.1004-132X.2020.17.010

    Liu Zhi-Qiang, Gu Xian-An, Guo Hao, Xi Hao, Wang Ming-Qiang, Li Jun-Li. A trajectory planning method for automatic placement of carbon n fiber screw propellers. China Mechanical Engineering, 2020, 31(17): 2079−2084 doi: 10.3969/j.issn.1004-132X.2020.17.010
    [158] 赵安安, 何大亮, 王晗, 郭俊刚, 柯映林. 复杂曲面上的自动铺放路径规划方法. 北京航空航天大学学报, 2022, 48(4): 595−601

    Zhao An-An, He Da-Liang, Wang Han, Guo Jun-Gang, Ke Ying-Lin. Automatic paving path planning method on complex surfaces. Journal of Beijing University of Aeronautics and Astronautics, 2022, 48(4): 595−601
    [159] Yan L, Chen Z C, Shi Y Y, Mo R. An accurate approach to roller path generation for robotic fibre placement of free-form surface composites. Robotics and Computer-Integrated Manufacturing, 2014, 30(3): 277−286 doi: 10.1016/j.rcim.2013.10.007
    [160] Wang P Y, Li Y, Wang X F, Xiao J. Research on fiber placement trajectory design algorithm for the free-form surface with given ply orientation information. Polymers and Polymer Composites, 2011, 19(2−3): 203−208 doi: 10.1177/0967391111019002-323
    [161] Qu W W, He R M, Wang Q, Cheng L, Yang D, Gao J X, et al. Algorithms for constructing initial and offset path of automated fiber placement for complex double-curved surfaces. Applied Composite Materials, 2021, 28(3): 855−875 doi: 10.1007/s10443-021-09901-2
    [162] Lu J R, Xu H J, Jiang Z H, Li K M, Hu J. Design of a composite truncated elliptical rotary shell based on variable-angle trajectories. Composite Structures, 2022, 294: Article No. 115772 doi: 10.1016/j.compstruct.2022.115772
    [163] Scheirer N, Holland S D, Krishnamurthy A. Fiber layup generation on curved composite structures. Computer-Aided Design, 2021, 136: Article No. 103031 doi: 10.1016/j.cad.2021.103031
    [164] Xu K, Hao X Z, Lin J M. Automated fibre placement path generation for complex surfaces via digital image deconvolution algorithm. Composites Part A: Applied Science and Manufacturing, 2022, 163: Article No. 107246 doi: 10.1016/j.compositesa.2022.107246
    [165] Wang K, Wang X P, Gan J Q, Jiang S K. A general method of trajectory generation based on point-cloud structures in automatic fibre placement. Composite Structures, 2023, 314: Article No. 116976 doi: 10.1016/j.compstruct.2023.116976
    [166] 张洋, 钟翔屿, 包建文. 先进树脂基复合材料自动丝束铺放技术研究现状及发展方向. 航空制造技术, 2013, 56(23): 131−136 doi: 10.3969/j.issn.1671-833X.2013.23.026

    Zhang Yang, Zhong Xiang-Yu, Bao Jian-Wen. Research status and future trend of automated fiber placement technology for advanced polymer matrix composites. Aeronautical Manufacturing Technology, 2013, 56(23): 131−136 doi: 10.3969/j.issn.1671-833X.2013.23.026
    [167] Oromiehie E, Prusty B G, Compston P, Rajan G. The influence of consolidation force on the performance of AFP manufactured laminates. In: Proceedings of the 21st International Conference on Composite Materials. Xi'an, China: Chinese Society for Composite Materials, 2017. 1−11
    [168] 徐志明. 复合材料自动铺放技术的研究进展及其工艺特点分析. 电气自动化, 2018, 40(1): 88−91

    Xu Zhi-Ming. Research progress of automatic placement of composite materials and analysis of its technological characters. Electrical Automation, 2018, 40(1): 88−91
    [169] Khan M A, Mitschang P, Schledjewski R. Parametric study on processing parameters and resulting part quality through thermoplastic tape placement process. Journal of Composite Materials, 2013, 47(4): 485−499 doi: 10.1177/0021998312441810
    [170] 刘林, 文立伟, 李勇, 肖军. 基于PMAC随动控制模式下自动铺带切割的研究. 宇航材料工艺, 2007, 37(5): 46−49 doi: 10.3969/j.issn.1007-2330.2007.05.012

    Liu Lin, Wen Li-Wei, Li Yong, Xiao Jun. Research on tape-cutting for automated tape laying based on tracking control mode of PMAC. Aerospace Materials & Technology, 2007, 37(5): 46−49 doi: 10.3969/j.issn.1007-2330.2007.05.012
    [171] Gao J C, Pashkevich A, Caro S. Optimization of the robot and positioner motion in a redundant fiber placement workcell. Mechanism and Machine Theory, 2017, 114: 170−189 doi: 10.1016/j.mechmachtheory.2017.04.009
    [172] FarzanehKaloorazi M, Bonev I A, Birglen L. Simultaneous path placement and trajectory planning optimization for a redundant coordinated robotic workcell. Mechanism and Machine Theory, 2018, 130: 346−362 doi: 10.1016/j.mechmachtheory.2018.08.022
    [173] Zhang X M, Xie W F, Hoa S V. Semi-offline trajectory synchronized algorithm of the cooperative automated fiber placement system. Robotics and Computer-Integrated Manufacturing, 2018, 51: 53−62 doi: 10.1016/j.rcim.2017.11.015
    [174] Hassan M, Liu D K, Xu D L. A two-stage approach to collaborative fiber placement through coordination of multiple autonomous industrial robots. Journal of Intelligent & Robotic Systems, 2019, 95: 915−933
    [175] He K, Nie H P, Yan C. The intelligent composite panels manufacturing technology based on tape-laying automatic system. Procedia CIRP, 2016, 56: 610−613 doi: 10.1016/j.procir.2016.10.120
    [176] Yao Y X. Adaptive Position/Force Control and Calibration of Robotic Manipulators as Applied to Automated Composite Tape-laying [Ph.D. dissertation], The University of Wisconsin-Madison, USA, 1988.
    [177] Jiang J X, He Y X, Wang H, Ke Y L. Modeling and experimental validation of compaction pressure distribution for automated fiber placement. Composite Structures, 2021, 256: Article No. 113101 doi: 10.1016/j.compstruct.2020.113101
    [178] Shirinzadeh B, Hui Tan B, Tronche D. Planning and simulation for robotic fibre placement. In: Proceedings of the 30th International Symposium on Robotics. Tokyo, Japan: International Federation of Robotics, 1999. 161−168
    [179] 朱珮旗, 樊红日, 钱波. 复合材料自动铺丝软件技术研究与应用综述. 软件工程与应用, 2022, 11(6): 1521−1533 doi: 10.12677/SEA.2022.116157

    Zhu Pei-Qi, Fan Hong-Ri, Qian Bo. Review of research and application about composite automated fibre placement software technology. Software Engineering and Applications, 2022, 11(6): 1521−1533 doi: 10.12677/SEA.2022.116157
    [180] Shirinzadeh B, Alici G, Foong C W, Cassidy G. Fabrication process of open surfaces by robotic fibre placement. Robotics and Computer-Integrated Manufacturing, 2004, 20(1): 17−28 doi: 10.1016/S0736-5845(03)00050-4
    [181] Shirinzadeh B, Wei Foong C, Hui Tan B. Robotic fibre placement process planning and control. Assembly Automation, 2000, 20(4): 313−320 doi: 10.1108/01445150010353242
    [182] Druiff P P, Ma K, Visrolia A, Arruda M, Palardy-Sim M, Bolduc S, et al. A smart interface for machine learning based data-driven automated fibre placement. In: Proceedings of the Composites and Advanced Materials Expo. Dallas, USA: 2021.
    [183] Wanigasekara C, Oromiehie E, Swain A, Prusty B G, Nguang S K. Machine learning-based inverse predictive model for AFP based thermoplastic composites. Journal of Industrial Information Integration, 2021, 22: Article No. 100197 doi: 10.1016/j.jii.2020.100197
    [184] Zimmerling C, Poppe C, Stein O, Kärger L. Optimisation of manufacturing process parameters for variable component geometries using reinforcement learning. Materials & Design, 2022, 214: Article No. 110423
    [185] Croft K, Lessard L, Pasini D, Hojjati M, Chen J H, Yousefpour A. Experimental study of the effect of automated fiber placement induced defects on performance of composite laminates. Composites Part A: Applied Science and Manufacturing, 2011, 42(5): 484−491 doi: 10.1016/j.compositesa.2011.01.007
    [186] Halbritter A, Harper R. Big parts demand big changes to the fiber placement status quo. In: Proceedings of the SME Composites Manufacturing. Mesa, USA: 2012.
    [187] Juarez P D, Gregory E D. In situ thermal inspection of automated fiber placement for manufacturing induced defects. Composites Part B: Engineering, 2021, 220 : Article No. 109002
    [188] Harik R, Saidy C, Williams S J, Gurdal Z, Grimsley B. Automated fiber placement defect identity cards: Cause, anticipation, existence, significance, and progression. In: Proceedings of the SAMPE 2018 Technical Conference and Exhibition. Long Beach, USA: SAMPE, 2018.
    [189] Shadmehri F, Ioachim O, Pahud O, Brunel J E, Landry A, Hoa S V, et al. Laser-vision inspection system for automated fiber placement (AFP) process. In: Proceedings of the 20th International Conference on Composite Materials. Copenhagen, Denmark: Aalborg University, 2015.
    [190] Rudberg T, Cemenska J. Incorporation of laser projectors in machine cell controller reduces ply boundary inspection time, on-part course identification and part probing. SAE International Journal of Aerospace, 2012, 5(1): 74−78 doi: 10.4271/2012-01-1886
    [191] Rudberg T, Nielson J, Henscheid M, Cemenska J. Improving AFP cell performance. SAE International Journal of Aerospace, 2014, 7(2): 317−321 doi: 10.4271/2014-01-2272
    [192] Juarez P D, Cramer K E, Seebo J P. Advances in in situ inspection of automated fiber placement systems. In: Proceedings of the SPIE 9861, Thermosense: Thermal Infrared Applications XXXVIII. Baltimore, USA: SPIE, 2016. Article No. 986109
    [193] Denkena B, Schmidt C, Völtzer K, Hocke T. Thermographic online monitoring system for automated fiber placement processes. Composites Part B: Engineering, 2016, 97: 239−243 doi: 10.1016/j.compositesb.2016.04.076
    [194] Yadav N, Oswald-Tranta B, Gürocak M, Galic A, Adam R, Schledjewski R. In-line and off-line NDT defect monitoring for thermoplastic automated tape layup. NDT & E International, 2023, 137: Article No. 102839
    [195] Chen H Z, Zhang Z J, Yin W L, Wang Q, Li Y F, Zhao C Y. Surface defect characterization and depth identification of CFRP material by laser line scanning. NDT & E International, 2022, 130: Article No. 102657
    [196] Cemenska J, Rudberg T, Henscheid M. Automated in-process inspection system for AFP machines. SAE International Journal of Aerospace, 2015, 8(2): 303−309 doi: 10.4271/2015-01-2608
    [197] 马少博. 复合材料自动铺放过程表层缺陷检测与识别方法研究 [硕士学位论文], 南京航空航天大学, 中国, 2020.

    Ma Shao-Bo. Surface Defect Inspection Technology of Automated Fiber Placement Manufacturing Process [Master thesis], Nanjing University of Aeronautics and Astronautics, China, 2020.
    [198] Tang Y P, Wang Q, Wang H, Li J X, Ke Y L. A novel 3D laser scanning defect detection and measurement approach for automated fibre placement. Measurement Science and Technology, 2021, 32(7): Article No. 075201
    [199] Tang Y P, Wang Q, Cheng L, Li J X, Ke Y L. An in-process inspection method integrating deep learning and classical algorithm for automated fiber placement. Composite Structures, 2022, 300: Article No. 116051 doi: 10.1016/j.compstruct.2022.116051
    [200] Nguyen D H, Sun X C, Tretiak I, Valverde M A, Kratz J. Automatic process control of an automated fibre placement machine. Composites Part A: Applied Science and Manufacturing, 2023, 168: Article No. 107465 doi: 10.1016/j.compositesa.2023.107465
    [201] Tao Y C, Jia S H, Duan Y G, Zhang X H. An online detection method for composite fibre tow placement accuracy. Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture, 2016, 230(9): 1614−1621 doi: 10.1177/0954405416640189
    [202] Chen M J, Jiang M, Liu X L, Wu B L. Intelligent inspection system based on infrared vision for automated fiber placement. In: Proceedings of the IEEE International Conference on Mechatronics and Automation (ICMA). Changchun, China: IEEE, 2018. 918−923
    [203] 蔡志强. 基于图像处理的纤维铺放缺陷检测研究 [硕士学位论文], 南京航空航天大学, 中国, 2017.

    Cai Zhi-Qiang. Research on Defect Detection Based on Image Processing for AFP [Master thesis], Nanjing University of Aeronautics and Astronautics, China, 2017.
    [204] Sacco C, Radwan A B, Harik R, Van Tooren M. Automated fiber placement defects: Automated inspection and characterization. In: Proceedings of the SAMPE 2018 Technical Conference and Exhibition. Long Beach, USA: SAMPE, 2018.
    [205] Sacco C. Machine Learning Methods for Rapid Inspection of Automated Fiber Placement Manufactured Composite Structures [Master thesis], University of South Carolina, USA, 2019.
    [206] Zambal S, Heindl C, Eitzinger C, Scharinger J. End-to-end defect detection in automated fiber placement based on Artificially generated data. In: Proceedings of the SPIE 11172, Fourteenth International Conference on Quality Control by Artificial Vision. Mulhouse, France: SPIE, 2019. 371−378
    [207] Meister S, Möller N, Stüve J, Groves R M. Synthetic image data augmentation for fibre layup inspection processes: Techniques to enhance the data set. Journal of Intelligent Manufacturing, 2021, 32(6): 1767−1789 doi: 10.1007/s10845-021-01738-7
    [208] Manyar O M, Cheng J Y, Levine R, Krishnan V, Barbič J, Gupta S K. Physics informed synthetic image generation for deep learning-based detection of wrinkles and folds. Journal of Computing and Information Science in Engineering, 2023, 23(3): Article No. 030903
    [209] Szarski M, Chauhan S. An unsupervised defect detection model for a dry carbon fiber textile. Journal of Intelligent Manufacturing, 2022, 33(7): 2075−2092 doi: 10.1007/s10845-022-01964-7
    [210] Schmidt C, Hocke T, Denkena B. Artificial intelligence for non-destructive testing of CFRP prepreg materials. Production Engineering, 2019, 13(5): 617−626 doi: 10.1007/s11740-019-00913-3
    [211] Zhang Y D, Wang W, Liu Q, Guo Z H, Ji Y C. Research on defect detection in automated fiber placement processes based on a multi-scale detector. Electronics, 2022, 11(22): Article No. 3757 doi: 10.3390/electronics11223757
    [212] 柯岩, 傅云, 周玮珠, 朱伟东. 基于Transformer的复合材料多源图像实例分割网络. 红外与激光工程, 2023, 52(2): Article No. 20220338

    Ke Yan, Fu Yun, Zhou Wei-Zhu, Zhu Wei-Dong. Transformer-based multi-source images instance segmentation network for composite materials. Infrared and Laser Engineering, 2023, 52(2): Article No. 20220338
    [213] Rath J E, Graupner R, Schüppstuhl T. Processing strategies for dieless forming of fiber-reinforced plastic composites. Machines, 2023, 11(3): Article No. 365 doi: 10.3390/machines11030365
    [214] Priyadharshini M, Balaji D, Bhuvaneswari V, Rajeshkumar L, Sanjay M R, Siengchin S. Fiber reinforced composite manufacturing with the aid of artificial intelligence——A state-of-the-art review. Archives of Computational Methods in Engineering, 2022, 29(7): 5511−5524 doi: 10.1007/s11831-022-09775-y
    [215] Yadav N, Schledjewski R. Review of in-process defect monitoring for automated tape laying. Composites Part A: Applied Science and Manufacturing, 2023, 173: Article No. 107654 doi: 10.1016/j.compositesa.2023.107654
    [216] Cong Y, Tian D Y, Feng Y, Fan B J, Yu H B. Speedup 3-D texture-less object recognition against self-occlusion for intelligent manufacturing. IEEE Transactions on Cybernetics, 2019, 49(11): 3887−3897 doi: 10.1109/TCYB.2018.2851666
    [217] Manyar O M, Kanyuck A, Deshkulkarni B, Gupta S K. Visual servo based trajectory planning for fast and accurate sheet pick and place operations. In: Proceedings of the ASME 17th International Manufacturing Science and Engineering Conference. West Lafayette, USA: ASME, 2022. Article No. V001T04A019
    [218] Döbrich O, Brauner C. Machine vision system for digital twin modeling of composite structures. Frontiers in Materials, 2023, 10: Article No. 1154655 doi: 10.3389/fmats.2023.1154655
    [219] Glück R, Korber M. Automated control and simulation of dynamic robot teams in the domain of CFK production. arXiv: 2210.11213, 2022.
    [220] Manyar O M, McNulty Z, Nikolaidis S, Gupta S K. Inverse reinforcement learning framework for transferring task sequencing policies from humans to robots in manufacturing applications. In: Proceedings of the IEEE International Conference on Robotics and Automation (ICRA). London, UK: IEEE, 2023. 849−856
    [221] Si W Y, Wang N, Li Q C, Yang C G. A framework for composite layup skill learning and generalizing through teleoperation. Frontiers in Neurorobotics, 2022, 16: Article No. 840240 doi: 10.3389/fnbot.2022.840240
  • 期刊类型引用(0)

    其他类型引用(5)

  • 加载中
图(8) / 表(9)
计量
  • 文章访问数:  802
  • HTML全文浏览量:  704
  • PDF下载量:  380
  • 被引次数: 5
出版历程
  • 收稿日期:  2023-03-20
  • 录用日期:  2023-08-31
  • 网络出版日期:  2024-04-19
  • 刊出日期:  2024-05-29

目录

/

返回文章
返回