Implementation of hybrid neuro-fuzzy and self-turning predictive model for the prediction of concrete carbonation depth: A soft computing technique

• Main Authors: Salim Idris Malami, Faiz Habib Anwar, Suleiman Abdulrahman, S.I. Haruna, Shaban Ismael Albrka Ali, S.I. Abba
• Format: Article
• Language: English
• Published: Elsevier 2021-06-01
• Series: Results in Engineering
• Subjects: Carbonation; Artificial intelligence; Extreme learning machine; Adaptive neuro-fuzzy inference system; Support vector machine
• Online Access: http://www.sciencedirect.com/science/article/pii/S2590123021000293

Carbonation is one of the critical problems that affects the durability of reinforced concrete; it is a reaction between CO2 gas and Ca (OH)2 when H2O is available, which forms powdery CaCO3 that alters the microstructure of the concrete by reducing its pH level and initiating corrosion that reduces the structure's service life. This study provides experimental information on the carbonation depths of samples from 10 separate existing reinforced concrete structures, where five are located in the inland area (Nicosia), while the other five are in the coastal area (Kyrenia) of the Turkish Republic of North Cyprus. The study found that the inland buildings have a higher depth of carbonation compared to the coastal buildings. The building structures in North Cyprus exhibit a higher rate of carbonation than the expected threshold within their life span. Constant values of B were yielded, which is useful in predicting carbonation depth. Using AI, the potential Hybrid Neuro-fuzzy model, which is comprised of an Adaptive Neuro-fuzzy Inference System (ANFIS), Extreme Learning Machine (ELM), Support Vector Machine (SVM) and a Conventional Multilinear Regression (MLR) model, were employed for the estimation of carbonation depth using experimental data, including age, compressive strength, current density, and carbonation constant. Four different performance indexes were used to verify the modelling accuracy, namely Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Nash- Coefficient (NSE), and Correlation Coefficient (CC). The results indicated that the AI models (ANFIS, ELM, SVM) performed better than the linear model (MLR) with NSE-values higher than 0.97 in both the testing and training stages. The results also indicated that the prediction skills of ANFIS-M2 increased the performance accuracy of ELM-M2, SVM-M2, and MLR-M2, and the ANFIS-M1 model performed better than ELM-1, SVM-1 and MLR-1 models in terms of prediction accuracy. The final outcomes indicated the capability of the non-linear models (ANFIS, ELM, and SVM) in the prediction of Cd.