New opportunity: Materials genome strategy for engineered cementitious composites (ECC) design

Abstract

Engineered cementitious composites (ECC) is considered as one of the most promising cement-based materials in construction industry. However, classical micromechanics-based design theory of ECC qualitatively indicates whether a mix meets the pseudo strain-hardening condition, and cannot provide detailed ECC mix compositions targeting specific mechanical performance requirements. This study presents a Materials Genome Initiative (MGI)-oriented strategy to design ECC through properties prediction and optimization design based on machine learning (ML). The genomic characteristics of ECC materials were summarized and analyzed, demonstrating that the mechanical properties of ECC could be closely related with its raw material attributes and mixture proportions. A comprehensive data-driven ML framework for the design of ECC was proposed, which integrates database construction with data treatment, feature selection, interpretable ML model prediction and inverse optimization design. A preliminary study towards designing ECC mixture proportions that meet specific performance requirements was further conducted. Three ML models were developed to predict the tensile properties of ECC based on the constructed database. The weight ratios of the binder, the sand-to-binder ratio, the water-to-binder ratio and the volumetric ratio, diameter, and length of fibers were employed as the input features for the ML modelling. With the best prediction models, two design scenarios with four objectives including tensile strength, tensile ductility, carbon footprint, and material cost of ECC were optimized using non-dominated sorting genetic algorithm II (NSGA-II). As a result, the proposed ML framework could achieve reliable properties prediction and rapid, quantitative, and intelligent design for ECC, which were validated by the experiments. This work introduces the concept of MGI and lays the groundwork of an AI-guided performance-based design of ECC, while further research remains.