Automated design of architectured polymer-concrete composites with high specific flexural strength and toughness using sequential learning

Architectured polymer-concrete composite (APCC) is a promising structural material with high mechanical performance while optimizing the design of APCC for a high flexural strength , high toughness, and light weight remains a challenge. This paper presents a machine learning-based approach to design APCC with high specific flexural strength and toughness. The proposed approach integrates sequential surrogate modelling , Latin hypercube sampling , and Lion Pride Optimization to predict and optimize the flexural properties of APCC. The proposed approach was implemented into designing APCC beams, which were fabricated via 3D printing and tested under flexural loads. Results show that the APCC beams achieved high flexural strength, high toughness, and light weight, simultaneously. The devised architecture of APCC arrested crack propagation and promoted energy dissipation . Parametric studies were performed to evaluate the effect of key design variables of APCC on flexural properties. This research advances the basic knowledge and capabilities of AI-assisted design of APCC. • A machine learning-based approach is developed to design and optimize APCC. • Sequential surrogate models and Latin hypercube sampling are used to predict APCC properties. • APCC beams are fabricated through 3D printing and tested under flexural loads. • APCC achieves high flexural strength, high toughness, and low density. • The effects of important design parameters and the underlying mechanisms are revealed.