Non-destructive laser-based quantification of corrosion rates in reinforced concrete using ensemble machine learning

Corrosion in reinforced concrete materials is a critical factor affecting the durability of structures and often leads to building collapse. In this research, we employed non-linear machine learning methods, including kernel-based support vector regression (SVR) and decision tree regression (DTR), to predict the corrosion rate in concrete based on the intensity of elemental constituents obtained via laser-induced breakdown spectroscopy (LIBS). To enhance model performance, adaptive boosting mechanisms were utilised to build ensemble models, effectively predicting corrosion rates. Model performance was assessed using standard metric performance indicators: mean absolute error (MAE), correlation coefficient (CC), root mean square error (RMSE), and Nash Sutcliffe efficiency (NSE). The standalone SVR model demonstrated consistent performance across varying input feature combinations during the testing phase, with NSE = 0.9611 and CC = 0.9813. Overall, the ensemble models outperformed the standalone models, with SVR-ADB-Combo-3 achieving NSE = 0.9618 and CC = 0.9821, and DTR-ADB-Combo-2 achieving NSE = 0.9839 and CC = 0.9951, exhibiting the best prediction performance. The findings of this study will serve as a valuable guide for researchers and policymakers in designing construction materials and setting standards, contributing to improved durability and safety in reinforced concrete structures.