Ensemble machine learning model for corrosion initiation time estimation of embedded steel reinforced self-compacting concrete

Corrosion initiation time of embedded steel is an important service life parameter, which depends on concrete material make-up, exposure environment, and duration of exposure. Early and accurate determination of corrosion initiation time will aid in designing durable reinforced concrete, saves cost and time. This study leveraged on the power of ensemble machine learning by combining the performances of different models in estimating the corrosion initiation time of steel embedded in self compacted concrete using corrosion potential measurement. The concrete specimens were prepared with limestone powder as supplementary addition to Portland cement and was exposed to 5% sodium chloride in accordance with the requirements of ASTM C876 – 15 for 8 months. During the exposure, corrosion potential of the embedded steel was measured, and the recorded datasets were used in training five different machine learning models. With cement, limestone powder, coarse aggregate, fine aggregate, water and exposure period.as input variables, five different models were developed to estimate the corrosion initiation time (determined from the corrosion potential measurements) of the embedded steel. With respect to model predictive performance, the acquired results demonstrated that the random forest (RF) ensemble model amongst other trained models performed best with 85/15 dataset percentage split for the training and testing. RF ensemble performed best with CC and RMSE of 99.01% and 18.2747 mV for training, and 98.67% and 25.0298 mV for testing respectively. Hence, due to its superior and robust performance, this study proposes RF ensemble model in the estimation of corrosion initiation time of embedded steel in reinforced limestone-cement blend concrete.