Machine Learning-Based Prediction of Self-Healing Smart Concrete Properties

Abstract

Self-Healing Smart Concrete has arisen as a promising solution to mitigate the detrimental effects of cracks and deterioration in concrete structures, enhancing their durability and longevity. It is a type of concrete that consists of substances or microbes which have the ability to self-heal fractures that may form over time as a result of a variety of circumstances including stress, weathering, or structural damage. As the laboratory experiments can be costly and time-consuming for analyzing the characteristics of Smart Concrete, machine learning algorithms can help to develop better formulations for the Self-Healing concrete. In this study, the machine learning (ML) tools are compared based on number of parameters to help determine the most suitable tool for creating predictive models. A total of 14 parameters were selected for comparison and 3 ML algorithms were identified through a detailed Literature Review viz. Random Forests Regressor, Extra Trees Regressor and Elastic Net Regressor. The results showed that the Extra Trees Regressor performed better in predictions giving 97.63% accuracy and with Standard Deviation value of about 0.005314 followed by Random Forests and Elastic Net Regressor. Therefore, Extra Trees Regressor can be applied to develop predictive model for assessing the performance of self-healing smart concrete.