|
Title |
Development of a Machine Learning-Based Model for the Prediction of Chloride Diffusion Coefficient Using Concrete Bridge Data Exposed to Marine Environments
|
|
Authors |
남우석(Woo-Suk Nam) ; 임홍재(Hong-Jae Yim) |
|
DOI |
https://doi.org/10.11112/jksmi.2024.28.5.20 |
|
Keywords |
기계학습; 염화물 확산계수; 해양 노출 환경; 콘크리트 교량; 정밀안전진단 데이터 Machine learning; Chloride diffusion coefficient; Marine exposure environment; Concrete bridge; Precision safety diagnosis data |
|
Abstract |
The chloride diffusion coefficient is a critical indicator for assessing the durability of concrete marine substructures. This study develops a prediction model for the chloride diffusion coefficient using data from concrete bridges located in marine exposure zones (atmospheric, splash, tidal), an aspect that has not been considered in previous studies. Chloride profile data obtained from these bridge substructures were utilized. After data preprocessing, machine learning models, including Random Forest (RF), Gradient Boosting Machine (GBM), and K-Nearest Neighbors (KNN), were optimized through hyperparameter tuning. The performance of these models was developed and compared under three different variable sets. The first model uses six variables: water-to-binder (W/B) ratio, cement type, coarse aggregate volume ratio, service life, strength, and exposure environment.
The second model excludes the exposure environment, using only the remaining five variables. The third model relies on just three variables: service life, strength, and exposure environment factors that can be obtained from precision safety diagnostics. The results indicate that including the exposure environment significantly enhances model performance for predicting the chloride diffusion coefficient in concrete bridges in marine environments.
Additionally, the three variable model demonstrates that effective predictions can be made using only data from precision safety diagnostics.
|