Performance prediction and analysis of engineered cementitious composites based on machine learning
Abstract
This study presents the implementation of machine learning (ML) techniques for mechanical properties prediction and analysis of polyethylene fiber-reinforced ECC (PE-ECC). A comprehensive database including different mechanical properties of PE-ECC was first constructed, with total 50 compressive strengths, 123 tensile strengths and 123 tensile strain capacities being assembled. Grey relational analysis was used to investigate the sensitivity of the critical parameters of PE-ECC’s mechanical properties. The evaluation results showed that the supplementary cementitious materials-to-binder ratio, water-to-binder ratio, sand-to-binder ratio, and fiber reinforcing index have significant effects on the mechanical properties of PE-ECC. Three representative ML techniques were utilized and demonstrated good predictive performance. A parametric study was further undertaken to quantify the effects of the selected parameters on the mechanical properties of PE-ECC based on the ML models. This study aims to help researchers and engineers estimate material properties of PE-ECC more effectively and provide supports for ECC design.