AI-TRANSPWOOD: AI methods and modelling accelerate the development of wood materials

Artificial intelligence and modelling are effective tools for computational research on new materials. The AI-TRANSPWOOD EU project focuses on transparent wood-based materials and their modelling. The project aims to effectively integrate advanced AI-based computational models with SSbD (Safe and Sustainable by Design) principles for the safe and sustainable design of wood-based composites.

The goal in developing new transparent wood composites is to integrate or even replace plastic and glass in the construction, automotive, electronics, and furniture industries. To support this, the project is developing user-centric tools, including alternative modelling methods and targeted LCA tools to measure product life cycles and environmental impacts, for consortium members and external industrial partners. From Finland, VTT, which is coordinating the project, and Aalto University are participating in this 13-partner project.

– The role of VTT and Aalto in the project focuses on the development of AI-based surrogate models, i.e., lighter AI models that replace the original model, as well as the broad development of various machine learning methods, says Professor Simo Särkkä.

Accelerated development of AI models with the LUMI supercomputer

The AI-TRANSPWOOD project utilises the GPU units on the LUMI supercomputer to develop surrogate and physics-driven AI models. They are trained using neural network models and PyTorch software to recognize various wood properties and screen them for the most promising candidates for new materials.

– Lightweight AI-based surrogate models are used for optimization instead of slow and heavy physics-based models, for example. This significantly speeds up model development, explains Aalto researcher Dr. Marcin Minkowski.

– The original physics-based models are used to generate data that is then used to train surrogate models. The original model can be revisited to check the results and validate the performance of faster models. The most promising surrogate models can be selected for development as AI models, and lighter models can be scaled up massively, for example, in a supercomputer computing environment. This is currently an established method for optimising and discovering new materials. If the original computational model is unsuitable for a high-performance computing environment, a surrogate model can also be used in this case, explains Joonas Linnosmaa, Senior AI Researcher at VTT.

CSC training and expert support familiar

VTT researchers are generally aware of and trained in the use of CSC resources. This enables the smooth and versatile use of computing systems in various projects. CSC’s computing services documentation is always available to help when needed. The informal weekly coffee meeting for researchers is a convenient way to network and, if necessary, ask for support for your own use. The multinational LUMI user support, LUST, which also assisted Marcin Minkowski, provides support for the international EuroHPC LUMI supercomputer.

– It is always extremely interesting and useful for CSC to be involved in industrial and academic RDI collaboration projects as an enabler of computational simulations and the use of new technologies. We want to provide researchers with the best possible tools. Smooth cooperation between companies, universities, research institutions and CSC is a particular focus of our development efforts, because it is essential for Finland that the world-class computing resources acquired to support domestic and European research projects are used efficiently in research and business applications, says Dan Still, Development Manager at CSC.

More information about the project: https://www.ai-transpwood-project.eu/partners

Image: Adobe Stock