TY - JOUR A1 - Kohler, Markus A1 - Mitsios, Dionysios A1 - Endisch, Christian T1 - Reconstruction-based visual anomaly detection in wound rotor synchronous machine production using convolutional autoencoders and structural similarity JF - Journal of Manufacturing Systems N2 - Manufacturing wound rotor synchronous machines (WRSMs) for electric vehicle traction systems necessitates rigorous quality inspection to ensure optimal product performance and efficiency. This paper presents a novel visual anomaly detection method for monitoring the needle winding process of WRSMs, utilizing unsupervised learning with convolutional autoencoders (CAEs) and the structural similarity index measure (SSIM). The method identifies deviations from the desired orthocyclic winding pattern during each stage of the winding process, enabling early detection of winding errors and preventing resource wastage and potential damage to the product or winding machinery. Trajectory-synchronized frame extraction aligns the visual inspection system with the winding trajectory, ensuring precise monitoring traceable to a specific point in the winding process. We present the comprehensive Winding Anomaly Dataset (WAD), which comprises images of WRSM rotor prototypes with and without winding faults recorded in different lighting conditions. The proposed reconstruction-based anomaly detection technique is trained on fault-free data only and utilizes the introduced masked mean structural dissimilarity index measure (MMSDIM) to focus on the relevant sections of the winding during inference. Comprehensive comparative analysis reveals that the CAE with unregularized latent space and the maximum mean discrepancy Wasserstein autoencoder (MMD-WAE) outperform the beta variational autoencoder (beta-VAE) in terms of anomaly detection performance, with the CAE and WAE delivering comparable results. Extensive testing confirms the approach’s effectiveness, achieving 95.6 % recall at 100 % precision, an AUROC of 99.9 %, and an average precision of 99.1 % on the challenging WAD, considerably outperforming state-of-the-art visual anomaly detection models. This work thus offers a robust solution for WRSM production quality monitoring and promotes the incorporation of visual inspection in electric drive manufacturing systems. UR - https://doi.org/10.1016/j.jmsy.2024.12.005 Y1 - 2024 UR - https://doi.org/10.1016/j.jmsy.2024.12.005 UR - http://nbn-resolving.de/urn/resolver.pl?urn:nbn:de:bvb:573-55371 SN - 1878-6642 VL - 2025 IS - 78 SP - 410 EP - 432 PB - Elsevier CY - Amsterdam ER - TY - CHAP A1 - Kohler, Markus A1 - Herrera, Christian A1 - Gerngroß, Martin A1 - Kennel, Ralph A1 - Endisch, Christian T1 - Empirical Investigation and Feed-Forward Control of Wire Tension in Needle Winding Processes T2 - 2024 IEEE 33rd International Symposium on Industrial Electronics (ISIE), Proceedings UR - https://doi.org/10.1109/ISIE54533.2024.10595747 Y1 - 2024 UR - https://doi.org/10.1109/ISIE54533.2024.10595747 SN - 979-8-3503-9408-5 PB - IEEE CY - Piscataway ER - TY - CHAP A1 - Kohler, Markus A1 - Fendt, David A1 - Endisch, Christian T1 - Modeling Geometric Wire Bending Behavior in Needle Winding Processes Using Circular Arcs with Tangential Linear Functions T2 - ITEC+2022: 2022 IEEE/AIAA Transportation Electrification Conference and Electric Aircraft Technologies Symposium (ITEC+EATS) UR - https://doi.org/10.1109/ITEC53557.2022.9814041 KW - wire forming KW - manufacturing KW - production systems KW - digital twin KW - machine learning Y1 - 2022 UR - https://doi.org/10.1109/ITEC53557.2022.9814041 SN - 978-1-6654-0560-7 SP - 894 EP - 901 PB - IEEE CY - Piscataway (NJ) ER - TY - CHAP A1 - Kohler, Markus A1 - Gerngroß, Martin A1 - Endisch, Christian T1 - A Test Bench Concept and Method for Image-Based Modeling of Geometric Wire Bending Behavior in Needle Winding Processes T2 - 2022 IEEE 31st International Symposium on Industrial Electronics (ISIE) UR - https://doi.org/10.1109/ISIE51582.2022.9831494 KW - wire forming KW - manufacturing KW - computer vision KW - machine vision KW - image processing KW - curve fitting KW - splines Y1 - 2022 UR - https://doi.org/10.1109/ISIE51582.2022.9831494 SN - 978-1-6654-8240-0 SN - 2163-5145 SP - 1113 EP - 1120 PB - IEEE CY - Piscataway ER - TY - CHAP A1 - Gerngroß, Martin A1 - Kohler, Markus A1 - Endisch, Christian A1 - Kennel, Ralph T1 - Model-based control of nonlinear wire tension in dynamic needle winding processes T2 - Proceedings, 2020 IEEE International Conference on Industrial Technology UR - https://doi.org/10.1109/ICIT45562.2020.9067168 Y1 - 2020 UR - https://doi.org/10.1109/ICIT45562.2020.9067168 SN - 978-1-7281-5754-2 SP - 381 EP - 388 PB - IEEE CY - Piscataway ER -