Development of a data-driven framework for monitoring corrosion under droplets

Understanding localized corrosion under atmospheric droplets is critical, yet previous studies have mostly focused on single-droplet systems or general trends, leaving the role of individual droplets within multi-droplet environments yet to be explored. Here, we present a fully automated, image-based, data-driven framework for analyzing corrosion progression under thousands of droplets simultaneously. Using time-resolved optical imaging and pre-trained large vision models for droplet segmentation, we construct per-droplet color features and propose a probability-based representation of corrosion product formation in inner and outer regions of interest. This approach overcomes the limitations of binary classification by capturing the continuous and spatially heterogeneous nature of corrosion product formation. Applied to carbon steel exposed to over 1500 pre-sprayed 1 M NaCl droplets of various sizes, the method reveals that the probability of corrosion product presence strongly depends on droplet size, with larger droplets more likely to exhibit products both under and around the droplet footprint. Moreover, corrosion products in the outer region can appear independently of under-droplet corrosion, suggesting a role for inter-droplet interactions. By transforming raw imaging data into physically meaningful per-droplet metrics, this work offers a scalable platform for investigating localized corrosion kinetics and morphology in complex, real-world droplet populations, opening new opportunities for connecting droplet formation and population behavior to local and overall atmospheric corrosion rates.