Probabilistic Assessment of Cooling Towers Under Carbonation-Induced Corrosion Using a Categorical Boosting Machine Learning Model

Probabilistic Assessment of Cooling Towers Under Carbonation-Induced Corrosion Using a Categorical Boosting Machine Learning Model

Stať ve sborníku v databázi WoS či Scopus

Accurate prediction of carbonation depth in reinforced concrete remains a challenge due to material variability, environmental exposure, and nonlinear diffusion dynamics. This study presents a machine learning approach using a Categorical Boosting (CatBoost) model trained on 830 accelerated carbonation test samples to predict annual carbonation rates with high fidelity (R²â€¯= 0.985, MAE = ±0.98 mm). Feature attribution analysis using SHapley Additive exPlanations (SHAP) identifies CO₂ concentration, exposure duration, SiO₂ content, and water-to-cement ratio as the most influential factors. The model is applied to a 42-year-old cooling tower in the Czech Republic to assess corrosion initiation risk. Ac celerated test predictions are converted to natural conditions using a stochastic transformation factor (mean = 11.1, standard deviation = 4.7), and time-dependent reliability is evaluated via Monte Carlo simulation. Results show that the reliability index (β) drops below the target serviceability threshold (β = 1.5) at year 50, indicating increasing corrosion risk. The proposed framework integrates data-driven prediction with probabilistic assessment and demonstrates improved interpretability and predictive ac curacy over traditional deterministic models.

Accurate prediction of carbonation depth in reinforced concrete remains a challenge due to material variability, environmental exposure, and nonlinear diffusion dynamics. This study presents a machine learning approach using a Categorical Boosting (CatBoost) model trained on 830 accelerated carbonation test samples to predict annual carbonation rates with high fidelity (R²â€¯= 0.985, MAE = ±0.98 mm). Feature attribution analysis using SHapley Additive exPlanations (SHAP) identifies CO₂ concentration, exposure duration, SiO₂ content, and water-to-cement ratio as the most influential factors. The model is applied to a 42-year-old cooling tower in the Czech Republic to assess corrosion initiation risk. Ac celerated test predictions are converted to natural conditions using a stochastic transformation factor (mean = 11.1, standard deviation = 4.7), and time-dependent reliability is evaluated via Monte Carlo simulation. Results show that the reliability index (β) drops below the target serviceability threshold (β = 1.5) at year 50, indicating increasing corrosion risk. The proposed framework integrates data-driven prediction with probabilistic assessment and demonstrates improved interpretability and predictive ac curacy over traditional deterministic models.

CatBoost; Concrete carbonation; Cooling tower; Probabilistic assessment; Reliability index; SHAP values

CatBoost; Concrete carbonation; Cooling tower; Probabilistic assessment; Reliability index; SHAP values

SIMWANDA, L.; LEHKÝ, D.; ŠOMODÍKOVÁ, M.; SYKORA, M.

16.06.2025

The International Federation for Structural Concrete (fib)

9782940643295

Concrete structures: extend lifetime, limit impacts - 20th fib Symposium Proceedings - 16-18 June 2025 - Antibes, France

1958

1969

12

https://shop.fib-international.org/publications/fib-proceedings/1046-21th-fib-Symposium-Proceedings-in-Antibes-2025-France

Proceedings_SimwandaEtAl_ProbabilisticAssessmentOfCoolingTowers