Bio-inspired discontinuous composite materials with a machine learning optimized architecture

Bio-inspired hierarchical discontinuous fibrous composite materials are investigated with the aim of achieving enhanced pseudo-ductility and elevated toughness. A novel methodology is proposed to search quickly and efficiently through the vast design space of the geometrical parameters of the discontinuities, combining advanced numerical simulations of the material's mechanical behavior with state-of-the-art Machine Learning approaches, such as Active Learning. A continuum mesoscale-based numerical model is developed to simulate the mechanical behavior of discontinuous composites under three-point bending loading and is utilized in a sequential Bayesian optimization scheme that iteratively searches for the material architecture that maximizes toughness. Five independent geometrical variables related to the size and exact topology of the discontinuities form a vast five-dimensional design space of more than 2.6 million possible combinations. In this space, the proposed methodology efficiently identifies, after 100 iterations, a remarkable optimal configuration that increases the material's toughness by more than 100%, with a knock-down effect on the ultimate bending strength of only 10%.