Machine learning helps develop protective coatings for steel materials

Steel materials are exposed to extreme temperatures around 600°C in some renewable energy technologies, such as heat transfer fluids in concentrated solar power, steam turbines, and hydrogen combustion. A promising solution for their protection lies in cost-effective and highly efficient coatings applicable as paint and known as Slurry Coatings.

"In order to accelerate the development and facilitate the lifetime prediction of these coatings, we are exploring the potential of machine learning methodologies. As part of this research, we used Symbolic Regression to model the effect of the thickness of the suspension (paint) applied to the steel surface. This approach provides a useful tool for industrial applications and is proving to be effective for complex systems that depend on several parameters," explained Vladislav Kolarik from Fraunhofer ICT, who believes that data-driven modelling using symbolic regression has proven to be an effective tool compared to physical models when the system depends on several parameters whose intensity of influence is unknown.

Concurrently, a Machine Learning algorithm underwent training to automatically assess the relevant parameters characterising the coating from the electron microscopy images. A methodology was developed to facilitate training when dealing with a limited number of available micrographs. The obtained parameters, in turn, serve as input parameters for Symbolic Regression model calculations. The algorithm is currently being tested on a set of micrographs of a selected coating system. Initial findings have been presented at scientific conferences, and publications are in progress, targeting both the computer science and materials science communities.

Text: Martina Šaradínová, PR specialist for R&D

Photo: VK archive