Machine Learning in the textile industry
Not just databased machine learning
In databased machine learning, statistical learning algorithms were developed that recognize patterns and structures in given data. The quality of the ML algorithms depends decisively on the quality and quantity of the available data. In the textile industry, enough measurements are usually collected for quality control. However, in the rarest of cases there is sufficient usable data available which links the process parameters with the product quality. This means that we cannot use purely data-driven machine learning - especially in plant and process optimization for customer-specific production processes.
Hybrid simulation-based machine learning In order to design and optimize production processes in the textile industry using ML, a hybrid approach were developed and used. For the design of processes and products, the textile industry has extensive experience. This expertise was formalized by describing the processes using physical models and then implementing them numerically. Simulations then provide the missing data to develop suitable ML algorithms and to interlock them with existing measurements. In this concept ML closes the gap between physically based simulation of production processes and the - in many cases not accessible to a physical model - quality measure of the end products.
The new hybrid ML process will be demonstrated at the Techtextil using the optimization of cross-winding machines as an example for a better dyeing of the wound yarn bobbins as part of the AiF DensiSpul project.