Fraunhofer Institute for Industrial Mathematics ITWM
Fraunhofer Institute for Industrial Mathematics ITWM
Deep Learning oder Machine Learning in der Textilindustrie
© iStockphoto / freepik
Deep Learning oder Machine Learning in der Textilindustrie

© iStockphoto / freepik

Machine Learning in Textile Industry

Intelligent machine systems and machine learning (ML) are dramatically changing the way we work. The basic concepts have been known for quite some time. As early as 1959, Arthur Samuel introduced the term machine learning as a »field of study that gives computers the ability to learn without being explicitly programmed«.

 

Data-Based Machine Learning

In practice, machine learning means that statistical methods are used to develop algorithms that learn to recognize patterns and structures in given data. This can be used, for example, to predict the product quality from the process parameters using machine learning methods. The quality of the ML algorithms and thus of the prognosis depends decisively on the quality and quantity of the available data sets. 

CFD-Simulation einer virtuellen Garnspule im Färbebad
© Fraunhofer ITWM
CFD-Simulation einer virtuellen Garnspule im Färbebad

Hybrid Simulation-Based Machine Learning

In recent years, researchers have focussed to be able to process ever larger amounts of data in real time. Especially for applications in the textile industry, however, the problem is often not the efficient processing of large amounts of data, but that there is not enough or not enough usable data available. For the design of processes and products in the textile industry, extensive experience knowledge is available, but usually not documented as symbolic knowledge and therefore not accessible for data-driven optimization with ML methods.

Therefore, we develop and use a hybrid approach to optimize production processes in the textile industry with ML methods. If purely data-driven machine learning methods cannot be used due to too little data or the lack of formalization of existing experience knowledge, we supplement these with simulations. We describe the processes with the help of physical models and implement them numerically. The simulations then provide the missing data to develop suitable ML algorithms and to dovetail them with existing measurement data. In this concept ML closes the gap between physically based simulation of the production processes and the quality measure of the end products, which in many cases is not accessible to a physical model.

The optimization of winding machines with regard to a better dyeing of the wound bobbins illustrates this innovative hybrid ML process in AiF’s DensiSpul project.

Related Links: