Resource T-EXDIZ - Data-supported efficiency increase in the development of textile products through experimentable digital twins using the example of tufting

The aim of the completed research project “Ressource T-EXDIZ” was to develop an experimentable digital material twin that, in cooperation with the experimentable digital process twin and the use of AI in product development, enables a significant increase in resource efficiency.

One of the biggest challenges in such a digitalization project is the lack of end-to-end digitalization of the companies. The basis for the use of AI / ML tools is a uniform, automatically analyzable database. In many cases, this is not available across the board. As a result, AI tools can only be used in exceptional situations.

With the Experimentable Digital Material Twin (EDMT), expert knowledge on the development of tufting products was transferred into digital form, enabling an a priori assessment to improve product development and a more efficient use of physical resources. To develop the EDMT, a material data room was set up in which the EDMT was mapped. This consisted of the process-relevant material and process parameters. The combination of the real system with the developed Experimentable Digital Process Twin (EDPT) (T-EXDIZ IGF21166) and the EDMT developed here resulted in an important building block for a cyber-physical system (CPS) for estimating material and process parameters. For a CPS, many potentially relevant material, process and product parameters had to be integrated, which led to a steadily increasing complexity of the overall system. To master this complexity, new correlations and dependencies between material, process and product parameters were determined with the help of correlation and cluster analyses as well as machine learning methods. The resulting AI-supported data-driven information and knowledge acquisition enabled the prediction of suitable setting parameters for new products and thus efficient product development and low-loss use of physical resources.

During the project, a list of parameters was developed that lists the influencing factors that the project team and the Project Advisory Committee believe have the greatest impact on the product properties. In addition to the influences, the measurability, units and measurement methods were also recorded here. This list was repeatedly amended and supplemented over the course of the project. Building on this basis, the DMT was developed in the form of an ontology and an ML model was trained on development and production data from industry and research. At the end of the project, the resulting EDMT was able to predict a selected set of parameters from material, product and process parameters for new products and reveal dependencies that were not represented in the training data, despite a still manageable database. This allows the conclusion to be drawn that the methods used with a larger database can include more parameters and deliver even more reliable results and can therefore be a valuable tool in product development. The tufting process selected as an example offered great potential for the development of new applications in the field of technical textiles.

Both the tufting industry and its suppliers as well as the machine industry and AI companies were represented in the project support committee. As part of the project, this committee contributed knowledge in workshops and gave advice on the development of the EDMT.

The final report on the research project is available for download in the TFI publication series 2024 / 127.