文献类型：期刊文献

英文题名：A denoising approach based on CEEMDAN-nlDAE for 3D reconstruction of on-machine measurement data in real shop-floor environments

作者：Li, Jin[1];Jiang, Zaixiang[1];Zhou, Weiwei[2];Guo, Yongning[2]

机构：[1]East China Univ Sci & Technol, Sch Mech & Power Engn, Shanghai 200237, Peoples R China;[2]CNNC SUFA Technol Ind Co Ltd, Dept Mech Engn, Suzhou 215011, Peoples R China

年份：2026

卷号：270

外文期刊名：MEASUREMENT

收录：;EI(收录号：20260920153480);WOS:【SCI-EXPANDED(收录号:WOS:001702347400003)】；

基金：This work was supported by the National Natural Science Foundation of China (Grant Nos. 52575060 and 12327807) .

语种：英文

外文关键词：On-machine measurement; 3D reconstruction; Real shop-floor environment; CEEMDAN-nlDAE; Denoising

摘要：On-machine measurement (OMM) eliminates errors caused by workpiece re-clamping and has thus emerged as a crucial technology for adaptive machining. Compared to coordinate measuring machine (CMM) operated in noise-free environments, the accuracy of OMM is typically compromised under harsh working conditions. Especially OMM systems equipped with precision sensors such as chromatic confocal sensors are highly sensitive to environmental disturbances. This study focuses on OMM of inner surfaces in narrow spaces and develops a four-degree-of-freedom (4-DOF) scanning system equipped with a chromatic confocal sensor. To address the challenges of complex noise during data acquisition in real shop-floor environments, we propose a denoising approach that combines complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) and a noise-learning-based denoising autoencoder (nlDAE). The model was trained using standard gauge block measurements and its experimental validation was conducted on the precision sealing surfaces of a gate valve body on a turning lathe. Experimental results show that the proposed approach reduces the flatness measurement error from 47.7 mu m to 13.9 mu m with a repeatability of about 1.5 mu m, and the angle measurement error from 1.258 degrees to 0.017 degrees with a repeatability of about 0.002 degrees. This approach demonstrates significant potential for precision OMM in harsh shop-floor environments and is particularly suitable for sensing systems where supervised learning is limited by the lack of reliable noise-free data, such as non-contact optical, spaceborne, and underwater measurement systems.