Explainable machine learning quantification and optimization of self-healing bacterial concrete for crack reduction in RC structures

Concrete is the key material for modern civilization; despite widespread criticism, how to better sustain its longevity is a huge sustainability issue. Cracking is widespread in the concrete structure and has significantly deteriorated over time, leading to repetitive, expensive repairs. Although many artificial intelligence (AI) techniques have been used to anticipate a variety of concrete qualities, the use of AI to predict the self-healing capacity of engineering cementitious composites is uncommon. To this end, four machine learning (ML) models were developed to predict the ability of calcium lactate to self-heal. To predict crack width after healing (CWAH), 344 data points were taken from the literature, with seven contributing factors: amount of cement, fine aggregate, coarse aggregate, water content, calcium lactate, Bacterial Spores Content, and CWAH. The data points were separated into two sets: training (80%) and testing (20%). To measure the efficacy of the developed ML models, several performance indicators were used, including R 2 , MSE, RMSE, MAE, and MedAE. During the testing phase, the best-optimized ML model (Hist Gradient Boosting) was identified with R 2 = 0.967, MSE = 0.002, RMSE = 0.0471, MAE = 0.0307, and MedAE = 0.0154. In contrast, the typical linear regression model fails to meet this condition, producing numerous negative predicted values. The sensitivity analysis results show that calcium lactate has the most significant impact on fracture width following self-healing. HGBR exhibits the highest predictive accuracy, with Calcium Lactate as the most influential factor in crack healing. These proposed ML models could be employed as a cutting-edge alternative for predicting the final CWAH, allowing engineers to analyze crack-reduction potential.