In a world where annual economic losses from corrosion exceed $2.5 trillion, the search for corrosion-resistant alloys and protective coatings continues. Artificial intelligence (AI) is playing an increasingly important role in the development of new alloys. However, the expected ability of artificial intelligence models to predict corrosion behavior and suggest optimum alloy formulas is not achieved.
Scientists at the Max Planck Research Institute (MPIE) have developed a machine learning model that improves prediction accuracy by 15% compared to existing frameworks. This model reveals new yet realistic combinations of corrosion resistant alloys. Its special strength comes from the combination of numerical and textual data. Originally developed for the critical area of pitting resistance in high strength alloys, the versatility of this model can be extended to all alloy properties. The researchers have published their latest results in the journalism. Science Advances.
Combine texts and numbers
“Each alloy has unique properties in terms of corrosion resistance. These properties depend not only on the alloy’s own composition, but also on the alloy manufacturing process. Current machine learning models can only use numerical data. However, machining methodologies and experimental test protocols, often documented with textual descriptors, very important to explain corrosion,” explains Kasturi Narasimha Sasidhar, lead author of the paper and former MPIE postdoctoral fellow.
The research team used language processing techniques similar to ChatGPT along with machine learning (ML) techniques on numerical data and developed a fully automated natural language processing framework. In addition, the inclusion of textual data in the ML framework allows for the identification of alloy compositions with enhanced pitting corrosion resistance.
“We trained a deep learning model using internal data that includes information about corrosion properties and composition. Michael Roverder, co-author of the paper and leader of the Corrosion Group at MPIE, said, “The model now includes alloys critical for corrosion resistance, even if individual elements were not originally entered in the model. can determine their composition.”
Pushing the limits: automated data analysis and image processing
In the newly developed framework, Sasidhar and his team used manually collected data as text descriptors. Currently, their goal is to automate the intelligent data analysis process and seamlessly integrate it into the existing structure.
The inclusion of microscopic images marks another milestone by envisioning next-generation AI frameworks integrating text, numerical and image-based data. Source
Source: Port Altele
