Prediction of coating degradation based on “Environmental Factors–Physical Property–Corrosion Failure” two-stage machine learning

Abstract The corrosion failure prediction of coating materials in diverse environments is of great significance for service performance evaluation. This work proposes a two-stage machine learning method that makes use of various data, including environmental factors, physical properties, and coating...

Full description

Saved in:
Bibliographic Details
Main Authors: Weiting Chen, Lingwei Ma, Yiran Li, Dequan Wu, Kun Zhou, Jinke Wang, Zhibin Chen, Xin Guo, Zongbao Li, Thee Chowwanonthapunya, Xiaogang Li, Dawei Zhang
Format: Article
Language:English
Published: Nature Portfolio 2025-06-01
Series:npj Materials Degradation
Online Access:https://doi.org/10.1038/s41529-025-00614-6
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract The corrosion failure prediction of coating materials in diverse environments is of great significance for service performance evaluation. This work proposes a two-stage machine learning method that makes use of various data, including environmental factors, physical properties, and coating barrier performance, to accurately evaluate the corrosion degradation of coatings. In the first stage, a 1-year outdoor exposure experiment of polyurethane coatings was conducted in nine representative climatic environments. A semi-supervised collaborative training regression model is established between key environmental data and physical properties of coatings (i.e., glossiness, adhesion, water contact angle, and yellowness). In the second stage, using the predicted physical property data as inputs, a machine learning model is constructed that links physical properties to the barrier performance of coatings and develops binary classification models that can distinguish between intact and damaged coatings. This two-stage modeling strategy provides enhanced prediction accuracy and scientific interpretability by incorporating intermediate physical property parameters.
ISSN:2397-2106