The production of building materials has been particularly energy-intensive up to now. The use of artificial intelligence could help in the development of alternative and climate-friendly materials. A team from the Bundesanstalt für Materialforschung und -prüfung (BAM) has developed an app that provides intuitive access. It could help to save CO2 in other areas as well.
The production of building materials contributes significantly to CO2 emissions around the globe. The production of cement alone is responsible for around eight per cent of global emissions of this greenhouse gas.
The energy-intensive manufacturing processes could be replaced by self-running chemical reactions: Alternative and more environmentally friendly cements do not have to be sintered in a blast furnace. The raw materials and chemicals react without heat input to form materials that have almost identical properties to cement. And they produce almost no CO2 emissions.
For this, however, it is essential to know exactly the properties of the alternative cements and the boundary conditions of their reaction. And this is where the difficulties begin: Cement is not only the most widely used material of mankind, but also a very complex building material: In particular, the raw materials of the CO2-friendly cements can differ greatly from one another, depending on their geological origin, and react differently with one another accordingly.
In purely mathematical terms, this quickly results in billions and more possible combinations. Researchers usually improve formulations on the basis of empirical observations in the laboratory. Traditional material science reaches its limits here simply because of the large number of combinations.
Artificial intelligence (AI)-based prediction of material properties can help here. Empirically observed properties of material samples or simulations are learned by an AI model in order to predict new and potentially better end products.
However, even the AI models still require large amounts of empirical information. Especially with cement, this "hunger for data" is a problem, because the reactions take place very slowly. Whether the desired result has been achieved can often only be assessed after several weeks of laboratory work. The development of an alternative cement would thus still take many years.
The use of sequential learning (SL) promises a breakthrough here: it has the potential to revolutionise materials research. The decisive difference to previous AI: SL also searches for new end products beyond the already empirically known materials and gets by with significantly less data overall. Paths that do not lead to the goal are discarded at an early stage and promising experiments are identified more quickly.
So far, SL has been successfully used, for example, in the development of pharmaceuticals or metallic glasses - in other words, for products whose synthesis takes place quickly or which can be easily captured in simulations.
A team from the Bundesanstalt für Materialforschung und -prüfung (BAM) led by Prof. Dr. Sabine Kruschwitz, in collaboration with Prof. Dietmar Stephan from the Technical University of Berlin, has now been able to show that the use of SL is also promising for cement research, although reactions there take place much more slowly. "We were able to prove that it is possible to find reliable and climate-neutral materials in less than eight months. Normally, the development cycle would have taken several years," explains Dr Christoph Völker, materials scientist at BAM, who developed the app. "The method could also be transferred to CO2-intensive areas such as steel, aluminium or asphalt production."
To make their approach generally usable, the researchers have now programmed an app to make it easier for the materials community to explore SL methods. The Sequential Learning App for Materials Discovery (SLAMD) offers low-threshold access to SL.
Similar to the many combinations of starting materials in the laboratory, there are almost infinite configuration possibilities for AI. With SLAMD, scientists can work on SL methods much faster than before on the development of climate-friendly materials via intuitive and interactive user interfaces.