Study Tests LLMs for Corrosion Assessment

Researchers Vincenzo Bongiorno, Michele Curioni, Niek Hijnen and colleagues evaluated large language models (LLMs) for corrosion-related image analysis in a study published in Corrosion Engineering, Science and Technology in 2026. They tested an approach combining LLMs with hard-coded logic on three tasks: classifying organic coating images as “blister” or “no blister,” ranking specimens by corrosion severity, and assigning images to severity classes. The “blister”/“no blister” classification reached accuracy of up to 0.93.

The researchers say LLMs can be guided by natural-language instructions without model training or data pre-processing, offering an accessible alternative to traditional machine learning. Across the corrosion tasks, their approach performed comparably to trained machine-learning models. The study describes LLMs combined with expert-driven strategies as practical and scalable tools for corrosion assessment.

Researchers Vincenzo Bongiorno, Michele Curioni, Niek Hijnen and colleagues evaluated large language models (LLMs) for corrosion-related image analysis in a study published in Corrosion Engineering, Science and Technology in 2026. They tested an approach combining LLMs with hard-coded logic on three tasks: classifying organic coating images as “blister” or “no blister,” ranking specimens by corrosion severity, and assigning images to severity classes. The “blister”/“no blister” classification reached accuracy of up to 0.93.

The researchers say LLMs can be guided by natural-language instructions without model training or data pre-processing, offering an accessible alternative to traditional machine learning. Across the corrosion tasks, their approach performed comparably to trained machine-learning models. The study describes LLMs combined with expert-driven strategies as practical and scalable tools for corrosion assessment.