CC BY-NC-ND 4.0Lück, MatthiasMatthiasLückLinde, Hendrik vonHendrik vonLinde2025-09-222025-09-222025https://publica.fraunhofer.de/handle/publica/496117https://doi.org/10.24406/publica-550710.1016/j.procir.2025.08.16610.24406/publica-55072-s2.0-105015302080In an increasingly competitive industrial landscape, optimizing production processes and ensuring quality assurance are paramount. This paper presents a robust method for intelligent quality assurance of process curves, leveraging unsupervised machine learning techniques to enhance the detection of process defects in manufacturing. Traditional monitoring approaches, often based on predefined rules, are limited in their ability to accurately classify defective products. The proposed methodology introduces a comprehensive data pipeline that includes resampling of process curves, application of Fourier transforms, and clustering techniques to effectively analyze large, unannotated datasets. Through this approach, we demonstrate that conventional evaluation methods can misclassify defective parts, highlighting the need for advanced data-driven solutions. Our findings indicate that integrating machine learning into quality assurance processes significantly improves the accuracy of defect detection, contributing to the goal of zero-defect production. This research not only offers a pragmatic solution for industrial quality assurance but also lays the groundwork for future advancements in the field.entrueFeature ExtractionFeature SelectionIndustrial ManufacturingProfile MonitoringQuality Controljournal article