Unsupervised machine learning for automated corrosion staging using optical microscopy images
Abstract Corrosion poses a substantial economic burden, and machine learning is increasingly being explored for its potential in staging, predictive maintenance, and data-driven decision making. This study presents an unsupervised automated corrosion staging method based on image processing and mach...
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Nature Portfolio
2025-07-01
|
| Series: | npj Materials Degradation |
| Online Access: | https://doi.org/10.1038/s41529-025-00635-1 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Corrosion poses a substantial economic burden, and machine learning is increasingly being explored for its potential in staging, predictive maintenance, and data-driven decision making. This study presents an unsupervised automated corrosion staging method based on image processing and machine learning using optical microscopy (OM) images. It detects and computes (i) the local porosity in a neighborhood of 5 μm Ã- 5 μm at pore locations, and (ii) the deposit thickness in (μm). The local porosity and deposit thickness were used to estimate the chloride concentration factor, associated pH, and the corrosion stage. The approach was tested on 48 ex-service OM images of under-deposit corrosion (UDC). A thickness-based approach yielded an accuracy of ~73% in classifying the corrosion stage in UDC compared to previous time-consuming approaches. This is a significant step in automating the evaluation of the corrosion stage, enabling scalable data-driven corrosion assessment across critical infrastructures. |
|---|---|
| ISSN: | 2397-2106 |