A new methodology for the segmentation and evaluation of diffusion coatings for steel surface treatment has been developed by scientists from the IT4Innovations National Supercomputing Center at VSB-TUO and the Fraunhofer Institute for Chemical Technology ICT in a paper published in the journal High Temperature Corrosion of Materials. The proposed method can facilitate the preparation of coatings for various specific environments.
“The study focuses on the design of an automatic evaluation of the parameters of aluminide diffusion coatings applied to steel for surface protection. The aim is to prevent the adverse environmental effects on the material by appropriate choice of coating layer. The parameters of the coating layers are evaluated from electron microscope image data. The developed approach uses machine learning and combines real and synthetically generated data for accurate determination of material parameters,” said Petr Strakoš, one of the authors of the paper.
Aluminide diffusion coatings and their importance
Aluminide diffusion coatings are an effective and affordable method of protecting steel from corrosion at high temperatures in harsh environments. These coatings are applied as an aluminium slurry, which is applied by spraying or painting. Aluminide diffusion coatings are widely used, for example, in the new generation of steam turbines, where they allow operation at temperatures above 650 °C, and in the protection of steels in power plants using concentrated solar power. The coating not only protects components such as tubing or tanks from corrosion, but also prevents the formation of toxic hexavalent chromium compounds, that occurs when the chromium from the steel reacts with the molten salt.
Machine learning in coatings evaluation
Analysis of scanning electron microscope (SEM) images is crucial for evaluating the structure of coatings, but traditional techniques face challenges such as imaging artifacts, image noise, or and overlaps in grayscale values among different physical features such as resins, cracks, and pores. Researchers from IT4Innovations present a new methodology that uses deep learning and U-Net architecture to segment SEM images.
To train the model, the researchers created "ground truth" data using ImageJ software and supplemented it with synthetic data generated in Blender 3D. The combination of real and synthetic data allowed the model to achieve exceptional accuracy: 98.7% for the Fe2Al5 layer, 82.6% for the pores and 81.48% for the precipitates.
Results: effect of suspension composition
The model was used to evaluate coatings formed by suspensions of three different compositions. The evaluation revealed that using a slurry without a rheology modifier may lead to a thicker partial Fe2Al5 layer that is formed by inward diffusion. The total thickness of the coating did not affect the ratio of external to internal diffusion. The thinner diffusion coating contained fewer pores and chromium precipitates regardless of the suspension composition.
The future of research
The newly developed segmentation and analysis method using machine learning (deep learning with U-Net architecture) opens up new possibilities for the optimisation of steel surface treatments. The method allows the influence of suspension composition on the structure of coatings to be efficiently investigated and thus optimise their protective properties. A significant advantage of the method is the use of synthetic data for model training, which allows to overcome the limited availability of real labeled data and to improve the generalisation of the model (improve its ability to operate on unknown data of the same type).
The proposed method will enable efficient optimisation of coatings for high temperature and corrosion resistance, contributing to the development of specific materials for different challenging environments. It will find particular application in the energy sector, for example in the design of a new generation of steam turbines or in systems using concentrated solar power.
Scientific paper
Segmentation and Metallographic Evaluation of Aluminium Slurry Coatings Using Machine Learning Techniques
https://doi.org/10.1007/s11085-024-10321-3
This research was supported by the European Union under the REFRESH project (CZ.10.03.01/00/22_003/0000048) and the Ministry of Education, Youth and Sports of the Czech Republic through e-INFRA CZ (ID:90254).
