Machine learning based multiscale calibration of mesoscopic constitutive models for composite materials: application to brain white matter

Abstract A modular pipeline for improving the constitutive modelling of composite materials is proposed.The method leveraged here development subject-specific spatially-varying brain white matter mechanical properties. For this application, microstructural information extracted from diffusion magnetic resonance imaging (dMRI) scans, and used to generate hundreds representative volume elements (RVEs) with randomly distributed fibre By automatically running finite element analyses on these RVEs, stress-strain curves corresponding multiple RVE-specific loading cases are produced. mesoscopic model homogenising RVEs’ behaviour then calibrated each RVE, producing a library parameters against set RVE characteristics. Finally, machine learning layer implemented predict directly any new microstructure. The results show that methodology can material properties high accuracy. More generally, overall framework allows efficient simulation when experimentally measured location-specific geometrical characteristics provided.