Scientists at the Max Planck Institute for Iron Research have developed a new machine learning model for corrosion-resistant alloys and are publishing their results in the journal Science Advances.
2.5 billion US dollars annually – that is the economic damage caused by corrosion worldwide. Science and industry have always been looking for new alloys that are corrosion resistant and for coatings that protect alloys from corrosion. In the search, artificial intelligence (AI) is increasingly being used to predict the corrosion behavior of materials and thus find optimal alloy compositions. However, the predictive power of previous AI models is limited because not all relevant data can be taken into account. Scientists at the Düsseldorf Max Planck Institute for Iron Research (MPIE) have developed a new machine learning model that can predict corrosive failure 15% more accurately than previous models and suggests new resistant alloys.
Originally developed for the critical area of pitting corrosion in high-strength alloys, the model can be expanded to cover all alloy properties. The researchers published their findings in the journal Science Advances.
Texts and numbers in an AI model
“The corrosion resistance of each alloy depends on its composition and how it is manufactured and processed. However, previous AI models could only process composition based on numerical data. However, since the production and processing of the alloy is documented textually, this data was not included in AI models. That’s why the informative value of previous AI models was limited,” explains Dr. Kasturi Narasimha Sasidhar, first author of the publication and former postdoctoral researcher at the Max Planck Institute for Iron Research.
The research team uses language processing methods similar to ChatGPT and combines them with machine learning (ML). The MPIE scientists were able to develop a machine learning model that processes numerical data and natural language fully automatically and can now better predict how alloys behave in the event of corrosion or which alloys are corrosion-resistant.
“Initially, we trained the learning model with data about corrosion properties and alloy composition. “Now the model is able to independently recognize corrosion-resistant alloys, even if the individual elements were not originally entered into the model,” says Dr. Michael Rohwerder, co-author of the publication and head of the “Corrosion” group at the Max Planck Institute for Iron Research.
Outlook: Automated data mining and image processing
So far, the AI model has been based on data collected manually by scientists. Your goal now is to automate the process of data mining and seamlessly integrate it into your model. In addition, the model should also be expanded to include microscopy images so that all relevant information sources, text, numbers and images, are incorporated into the AI model and thus further increase its informative value.
You can access the original publication here.