首页|基于融合特征t-SNE降维的控制图质量异常模式识别

基于融合特征t-SNE降维的控制图质量异常模式识别

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
为解决控制图质量异常模式识别中,实时质量数据呈现出高维非线性等复杂特征导致模型过拟合以及失真等问题,提出一种基于融合特征t分布随机近邻嵌入(t-distributed stochastic neighbor embedding,t-SNE)降维的控制图质量异常模式识别方法.首先,从生产过程动态数据流中提取其统计特征、形状特征并与原始特征进行融合,形成动态数据流的高维融合特征;然后利用t-SNE算法对融合特征进行降维,t-SNE算法能够有效地处理线性和非线性数据,并产生更有意义的聚类;进而利用一维卷积神经网络(one-dimensional convolutional neural networks,1DCNN)作为分类器实现复杂产品制造过程的质量异常模式识别;最后,通过仿真实验将本文所提方法与单一类型特征方法、融合特征方法以及融合特征主成分分析法(principal component analysis,PCA)、核主成分分析(kernel PCA,KPCA)和局部线性嵌入算法(locally linear embedding,LLE)降维方法的识别模型进行比较,并利用锂离子电池极片制造过程为例进一步说明本文模型的有效性与实用性.仿真与实例结果表明,本文所提算法具有更高的识别效率和精度,特别适用于处理在复杂产品制造过程背景下的高维非线性数据.
Control chart quality abnormal pattern recognition based on hybrid feature t-SNE fusion
To address the issues of overfitting and distortion caused by the complex charac-teristics of high-dimensional nonlinear real-time quality data in control chart quality abnormal pattern recognition,a control chart quality abnormal pattern recognition method based on fused feature t-distributed stochastic neighbor embedding(t-SNE)dimensionality reduction is pro-posed.Firstly,statistical features and shape features are extracted from the dynamic data stream of the production process,and fused with the original features to form high-dimensional fused features of the dynamic data stream.Then,the fused features are reduced using the t-SNE algorithm,which effectively handles both linear and nonlinear data and produces more meaning-ful clustering.Furthermore,a one-dimensional convolutional neural networks(1DCNN)is used as a classifier to achieve quality abnormal pattern recognition in complex product manufactur-ing processes.Finally,through simulation experiments,the proposed method is compared with single-type feature methods,fused feature methods,and fused feature principal component anal-ysis(PCA),kernel PCA(KPCA)and locally linear embedding(LLE)dimensionality reduction methods for the recognition models.The effectiveness and practicality of the proposed model are further demonstrated using the example of the lithium-ion battery electrode manufacturing process.Simulation and case study results show that the proposed algorithm has higher recogni-tion efficiency and accuracy,particularly suitable for handling high-dimensional nonlinear data in the context of complex product manufacturing processes.

control chartpattern recognitioncomplex product manufacturing processt-distributed stochastic neighbor embedding(t-SNE)convolutional neural networks(CNN)

王宁、郭梓昱、田淑珂、李可雨阳

展开 >

郑州大学商学院,郑州 450001

郑州大学国际质量发展研究院,郑州 450001

控制图 模式识别 复杂产品制造过程 t分布随机近邻嵌入(t-SNE) 卷积神经网络(CNN)

国家自然科学基金重点资助项目河南省高等学校青年骨干教师培养项目河南省高校哲学社会科学创新人才支持计划郑州大学人文社会科学优秀青年科研团队

U19042112021GGJS0062023-CXRC-192023-QNTD-01

2024

10.12011/SETP2023-1577

系统工程理论与实践
中国系统工程学会

系统工程理论与实践

CSTPCDCSSCI北大核心
影响因子：1.575
ISSN：1000-6788
年,卷(期)：2024.44(7)