Hybrid AI Models for Sustainable and Efficient Nonwoven and Meltblown Production

Combining the Strengths of AI and Simulation

In the »HyPNos« project, we are developing hybrid process models for meltblown production that combine numerical simulations, Artificial Intelligence (AI), and expert knowledge. This approach addresses the limitations of existing methods: while purely AI-based models require large amounts of high-quality production data, purely numerical simulations are computationally expensive and demand extensive domain expertise.

By combining the strengths of both approaches, our hybrid models provide the foundation for the systematic optimization of meltblown processes. Our goal is to reduce energy and material consumption, thereby improving both the environmental and economic efficiency of production. At the same time, we aim to maintain or further enhance product quality, for example with respect to basis weight, tensile strength, and filtration performance.

Hybrid Modeling and Digital Process Support

Our work follows an interdisciplinary approach that closely integrates simulation, measurement technology, data analytics, and Artificial Intelligence:

Systematic Process Analysis
We begin by identifying the key process parameters and their interrelationships within the meltblown process using structured methods such as cause-and-effect diagrams. Based on this analysis, we define the requirements for data acquisition, process models, and the Machine Learning (ML) workflow.

Numerical Simulation of Subprocesses
For key process steps – including polymer spinning, fiber entanglement in turbulent airflows, and web formation – we combine existing simulation tools into an integrated simulation chain. This provides both high-quality training data and physically meaningful intermediate variables that are difficult or impossible to measure experimentally.

AI Models Based on Real Production Data
At pilot-scale and industrial production facilities, we collect, structure, and analyze process and quality data. Building on these datasets, we evaluate and apply various machine learning methods, including neural networks and Gaussian process regression, to model both individual subprocesses and the overall meltblown process.

Hybrid AI Models
In the next step, simulation data, experimental measurements, and expert knowledge are combined in hybrid models. We also employ advanced approaches such as Physics-Informed Machine Learning (PIML), which directly incorporates physical laws into the learning process. The resulting models are robust, partially physics-based, and simultaneously data-driven.

Digital Decision Support for Process Optimization (Interactive Dashboard)
Finally, the developed models are integrated into a digital decision-support tool. Through an interactive dashboard, users can adjust process parameters, target values, and operational constraints while immediately visualizing the resulting effects on product quality, energy consumption, and material usage.

Benefits for Industry: Greater Efficiency, Sustainability, and Speed in Nonwoven Production

The methods and tools developed in the project support companies in the nonwoven industry across several application areas:

Optimizing Production Processes
Our approach enables more effective process parameter settings and, consequently, more efficient plant operation. This helps reduce specific energy consumption, minimize material losses, and decrease production waste.

Accelerating Product Development
By combining virtual testing with data-driven predictions, we significantly reduce the need for experimental trials. New material formulations and process windows can be evaluated more quickly and optimized in a targeted manner.

Enabling the Use of Sustainable Raw Materials
Our hybrid models facilitate the integration of recycled and bio-based polymers by enabling their processing behavior to be systematically analyzed and optimized.

Improving Utilization of Existing Production Capacity
We help companies identify new application areas for meltblown nonwovens more rapidly and assess their technical feasibility and economic viability.

Transferability of the Methods
The hybrid AI modeling concepts and system architectures developed within the project are not limited to meltblown processes. They can also be transferred to other nonwoven manufacturing technologies and a wide range of continuous production processes.

Project Partner, Duration, and Funding

The »HyPNos« project is carried out in collaboration with the Institute for Textile Technology (ITA) at RWTH Aachen University. In close cooperation with an industrial advisory board, we ensure that the methods and results developed throughout the project are aligned with the practical needs of the nonwoven industry and are specifically prepared for transfer into industrial applications.

»HyPNos« runs from September 1, 2024, to February 28, 2027, and is funded by the Industrial Collective Research (IGF – Industriellen Gemeinschaftsforschung) program. The project is aimed primarily at Small and Medium-sized Enterprises (SMEs) in the nonwoven industry and their supply chain. The IGF program is funded by the German Federal Ministry for Economic Affairs and Energy (BMWE – Bundesministerium für Wirtschaft und Energie) and supports SMEs in Germany through pre-competitive collaborative research projects.