Exploring the use of artificial intelligence (AI) for corrosion-related tasks: Using large language models (LLMs) to analyse organic coatings images

Large language models (LLMs) have increasingly been applied as a novel and accessible alternative to traditional machine learning (ML). Their main advantage lies in the fact that they do not require model training or data pre-processing; instead, they are guided through natural language instructions. In this study, we evaluate the suitability of LLMs for corrosion-related tasks and introduce an approach that combines LLMs with hard-coded logic to enhance reliability and performance. The methodology was tested on three tasks: classifying images of organic coatings in ‘blister’/‘no blister’ classes, ranking specimens based on corrosion severity, and classifying images into severity classes. The results demonstrate performance comparable to that of trained ML models, with ‘blister’/‘no blister’ classification achieving accuracies up to 0.93. These findings highlight the potential of LLMs, when integrated with expert-driven strategies, as practical and scalable tools for corrosion assessment.