Corrosion is a major challenge impacting industries such as construction, petrochemicals, and metallurgy. This study presents an innovative approach to predicting corrosion rates in reinforced concrete by employing laser-induced breakdown spectroscopy (LIBS) and machine learning. Corroded samples were analyzed using LIBS to identify elemental compositions and their intensities, focusing on Si, Al, Fe, Ti, Na, Ca, and Cl as input features for the models. Three different machine learning models (Support Vector Regression (SVR), Gaussian Process Regression (GPR), and Decision Tree Regression (DTR)) were trained to predict corrosion rates. The input features were tested in three combinations: Combo-1 (Si, Al, Cl), Combo-2 (Si, Al, Fe, Ca, Cl), and Combo-3 (Si, Al, Fe, Ti, Na, Ca, Cl). The models' performance was evaluated using metrics such as Mean Absolute Error (MAE), Nash Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), and Correlation Coefficient (CC). The results showed that Combo-3 consistently achieved the highest prediction efficiency for SVR (NSE = 0.9734, CC = 0.9874) and GPR (NSE = 0.9785, CC = 0.9900) during testing, highlighting the importance of comprehensive input descriptors. However, for the DTR model, Combo-1 (NSE = 0.9997, CC = 0.9970) exhibited the highest prediction efficiency, demonstrating its capability to accurately predict with fewer descriptors. This research offers valuable insights for industries and policymakers aiming to mitigate the adverse effects of corrosion, emphasizing the potential of combining LIBS with machine learning for precise corrosion rate prediction.
Read full abstract