Researchers from the Indian Institute of Science (IISC) and Qatar Science and Technology Research Centre (QSRTC) have presented innovative and fully automated approaches to detect and quantify corrosion in industrial equipment, leveraging advanced machine learning and image analysis.
Those methods detailed in the degradation of NPJ materials promise great advances in monitoring and mitigating corrosion, a costly and dangerous issue for industries ranging from power generation to oil and gas.
“Corrosion quietly undermines the integrity of these systems, puts life and livelihoods at risk, and burdens society with escalating maintenance costs,” said Professor Phaneendra K. Yalavarthy, Department of Computational and Data Science, IISC, senior author of the study.
The new AI-powered algorithm analyzes high-resolution microscope images of metal surfaces to assess two important indicators of corrosion: the thickness of the corrosive deposit and the porosity within it (small holes).
By quantifying these characteristics, the algorithm makes changes in the chemical environment, such as increased chloride content and acidity, to exacerbate corrosion.
“We've identified certain pH levels that indicate corrosion is getting worse. For example, if the pH is below 2.8-3, it means that the corrosion has reached a very serious stage,” explained Yalavarthy.
Co-author Ashwin Rajkumar said, “It was impressive to see such a strong correlation between local pH and the threshold that marks escalation to the corrosion stages, particularly the most dangerous stages. This gives a predictive marker of severity.”
The researcher's method leaves traditionally supervised machine learning using unsupervised approaches, particularly k-means clustering. This efficiently segments microscopic images into areas of sediment and pores without relying on pre-labeled data.
This flexibility is essential as the structure and appearance of corrosive products can vary widely. In initial tests involving low deposit corrosion of steam generator tubes, the model correctly assessed the severity of corrosion for approximately 73% of the time, surpassing manual analysis of speed and consistency.
Going forward, the team aims to verify methods on broader, more complex datasets to ensure robustness in the entire industrial environment.
The technology is promising for real-time, data-driven corrosion assessments, allowing for safer and more efficient operation of critical infrastructure. Because it is integrated with digital monitoring, such advances could reconstruct the maintenance and safety of industries around the world.
– end
