Prediction of crack repair percentage in self-healing concrete using machine learning.

Concrete is the most widely used construction material, yet it is highly susceptible to micro-crack formation, which can critically reduce its durability and structural performance. This challenge has drawn significant research interest toward the self-healing capability of concrete as a sustainable solution. However, the effectiveness of self-healing in concrete is still mainly evaluated through time-consuming and costly experimental procedures, with no universally accepted standard methods, limiting its practical implementation. To address this problem and harness the potential of artificial intelligence for modeling complex nonlinear behaviors, this study develops three hybrid predictive models: artificial neural network (ANN) optimized with genetic algorithm (GA), particle swarm optimization (PSO), and the Levenberg-Marquardt (LM) algorithm. Model performance was assessed using multiple statistical indices, and the results demonstrated that all three hybrid models outperformed the reference study by Zhuang et al., with the ANN-LM model yielding the highest prediction accuracy. The findings highlight that integrating optimization algorithms with ANN provides a robust and reliable framework for predicting the crack healing percentage in self-healing concrete, offering valuable insights for the design and improvement of durable and sustainable cement-based materials.