Prediction of self-healing ability of recurring cracks in Engineered Cementitious Composites with a machine learning based computational approach

Engineered Cementitious Composites (ECCs) with inherent self-healing mechanisms based on ongoing hydration have gained significant attention; however, the occurrence of repeated self-healing in ECC remains inadequately explored, posing challenges in real-world engineering applications, but laboratory experiments to investigate recurring crack self-healing in ECCs are time-consuming and costly. In this study, in order to address these issues, we present a novel hybrid model that combines an Evolutionary Algorithm (EA) with a Back-propagation Neural Network (BPNN) to optimize the initial weights and biases of the network. The model is trained, validated, and tested using detailed laboratory experimental data, which are comprehensively described in this paper. The computational results demonstrate the effectiveness of the proposed hybrid model in accurately predicting ECCs’ recurring crack self-healing ability, providing an efficient tool for its prognosis. Additionally, a Shapley Additive Explanations (SHAP) analysis is conducted to investigate the impacts of previous crack widths and mix components on recurring crack self-healing ability. The findings herein advance our understanding of the influencing factors associated with recurring cracks and have the potential to reduce the time and cost associated with experimental programs, while facilitating the selection of optimal mixture components to enhance the design of durable ECC structures. • EA-optimized BPNN model accurately predicts ECC’s crack self-healing ability. • SHAP analysis reveals impact of inputs on recurring crack self-healing. • Relative importance of factors shifts across recurring self-healing stages. • Computational approach reduces time and cost of ECC self-healing research.