Researchers from the Indian Institute of Science (IISc) and Qatar Science and Technology Research Center (QSRTC) have unveiled an innovative, fully automated approach for detecting and quantifying corrosion in industrial equipment, harnessing advanced machine learning and image analysis.
Their method, detailed in npj Materials Degradation, promises significant progress in monitoring and mitigating corrosion, a costly and dangerous problem for industries from power generation to oil and gas.
“Corrosion silently compromises the integrity of these systems, putting lives and livelihoods at risk and burdening society with escalating maintenance costs,” noted Professor Phaneendra K. Yalavarthy from IISc’s Department of Computational and Data Sciences, a senior author of the study.
The new AI-powered algorithm analyses high-resolution microscope images of metal surfaces to assess two crucial indicators of corrosion: the thickness of corrosive deposits and the porosity (tiny holes) within them.
By quantifying these features, the algorithm infers changes in the chemical environment—such as increasing chloride content and acidity—that signal worsening corrosion.
“We identified specific pH levels that indicate when corrosion is worsening. For example, when the pH drops below 2.8-3, it means corrosion has reached a very severe stage,” explained Yalavarthy.
Co-author Ashwin RajKumar added, “It was striking to see such a strong correlation between local pH and corrosion stage, especially the threshold that marks escalation to the most dangerous phase. This gives us a predictive marker for severity.”
The researchers’ method departs from conventional supervised machine learning by using an unsupervised approach—specifically, k-means clustering—which efficiently segments microscopy images into regions of deposits and pores without relying on pre-labeled data.
This flexibility is vital, as corrosion products can vary widely in structure and appearance. Initial testing, including under-deposit corrosion of steam generator tubes, saw the model correctly assess corrosion severity about 73% of the time, outperforming manual analysis in speed and consistency.
Looking ahead, the team aims to validate their technique on broader and more complex datasets to ensure robustness across industrial settings.
The technology holds promise for real-time, data-driven corrosion assessment, enabling safer and more efficient operation of critical infrastructure. Integrated with digital monitoring, such advances could reshape industrial maintenance and safety worldwide.
– Ends