The AutoNoM project, funded by the German Federal Ministry of Education and Research under grant 01IS19015A, ran from 1 September 2019 to 31 December 2022 and was carried out by a consortium of five partners. The consortium leader, atlan‑tec Systems GmbH, developed the data‑curation framework in line with the VDE/VDI 3714 guideline, while proCtec GmbH supplied process‑control interfaces and data acquisition hardware. Axon Machine Vision GmbH & Co. KG provided the optical measurement system for areal‑weight determination, and BNP Brinkmann GmbH & Co. KG served as the industrial application partner, offering production data and operational expertise. The research arm was the Institute for Textile Technology (ITA) at RWTH Aachen University, which performed the machine‑learning modelling and validation. Together, the partners defined the data requirements, built a single‑source‑of‑truth data lake, and implemented a dashboard for real‑time monitoring and historical analysis.
Technically, AutoNoM achieved the first global automation of production‑and‑quality data integration for nonwoven fabrics. The project introduced an automated data‑curation pipeline that removes duplicate records while accounting for measurement uncertainty, eliminates temporal dependencies through de‑biasing, and applies plausibility and redundancy filters. This ensures that only causally consistent, high‑quality data feed the machine‑learning models, thereby markedly improving robustness against over‑fitting and poor data quality. The curated data were used to train a soft‑sensor predictor that estimates the uniformity of areal weight in real time. The resulting set‑point system, demonstrated as a proof‑of‑concept, can be evolved into a fully autonomous real‑time optimisation (RTO) platform once additional safety measures per VDE/VDI 3714 are incorporated.
An economic‑technical optimisation model was also developed, linking production‑cost data with quality outcomes to identify the minimal‑cost operating envelope that satisfies the required areal‑weight tolerance. The model was iteratively refined in collaboration with BNP Brinkmann’s technologists and machine operators, ensuring that the optimisation aligns with practical constraints. The optical measurement system, integrated into the production line, provides high‑resolution areal‑weight data that feed both the soft‑sensor and the optimisation model, closing the loop between measurement, prediction, and control.
The AutoNoM demonstrator, built on the curated data and the soft‑sensor, achieved a significant reduction in areal‑weight variability, which in the industry typically fluctuates by 5–10 %. While the report does not specify the exact percentage reduction, the system’s ability to maintain consistent set‑points under varying raw‑material and environmental conditions demonstrates its practical value. The methodology is generic and can be transferred to other fibre‑based products such as woven fabrics, knitted textiles, paper, glass, and films, with initial discussions already underway with potential adopters in those sectors.
In summary, AutoNoM delivered a comprehensive, end‑to‑end solution that combines automated data curation, real‑time soft‑sensor prediction, and economic optimisation for the nonwoven fabric industry. The collaboration between industry partners, a machine‑vision specialist, a process‑control provider, and an academic research institute enabled the integration of cutting‑edge machine‑learning techniques into a production environment, paving the way for broader adoption of data‑driven manufacturing across the textile and related sectors.