Multiscale modelling and data‑Driven optimisation of sustainable bioepoxy nanocomposites reinforced with cellulose nanocrystals and nanobiochar

This thesis investigates sustainable bioepoxy nanocomposites reinforced with cellulose nanocrystals (CNC) and novel wheat straw-derived nanobiochar (NBC) through an integrated approach combining multiscale modelling, experimental characterisation, and interpretable machine learning. Analytical micromechanics and finite element analysis (FEA) were used to determine the interphase properties and elastic moduli of untreated and esterified CNC/epoxy systems. Representative volume elements generated in Digimat-FE and analysed in ANSYS showed that explicitly modelling an interphase region produced modulus predictions consistent with experimental observations. Increasing the CNC content from 2.5 wt% to 5 wt% increased composite stiffness, and esterification further enhanced both the modulus and interphase thickness. Response surface optimisation identified an optimal configuration with a 6 nm interphase, an aspect ratio of 10, a CNC modulus of 1000 Pa, and an interphase modulus of 0.65 GPa to predict the reduction in modulus resulting from excessive esterification treatment.

Complementary experiments examined the reinforcement efficiency of CNC and NBC in bioepoxy. CNC increased the Young’s modulus by 13.6% at 5 wt% to 3.5 ± 0.2 GPa and improved mode I fracture toughness by 37.2% at 3 wt% to 110 ± 11 J m⁻², while dispersion challenges and agglomeration reduced performance at higher loadings. NBC outperformed CNC, achieving a 24.2% modulus increase at 10 wt% and a 72.7% fracture toughness improvement at 7 wt%. Scanning electron microscopy confirmed crack pinning and deflection as the governing toughening mechanisms, and dynamic mechanical analysis showed glass transition temperature increases up to 69.1 ± 0.1 °C with NBC.

An ordinary state-based peridynamic model was developed to simulate the transverse fracture mechanics of NBC-coated flax fibre composites. The spray coating was computationally resolved as a two-ring functionally graded interphase with varying radial stiffness and fracture toughness. The peridynamic simulations successfully captured progressive damage evolution, transverse crack kinking, and interfacial debonding, demonstrating that the NBC coating acts as a sacrificial, damagetolerant layer. This non-local framework accurately replicated the 15% increase in peak transverse strength observed experimentally in the 5 wt% NBC-treated composites compared to the untreated baseline.

An interpretable machine learning framework using six tree-based models was developed to predict and optimise CNC/epoxy mechanical properties. Gradient Boosting achieved the highest accuracy, with SHAP analysis identifying matrix modulus, CNC content, and aspect ratio as the dominant predictors. Multi-objective optimisation indicated an optimal CNC loading between 3.4 and 4.5 wt%.