Fourier-encoded neural surrogates for modelling multi-physics problems in composites processing

Composite manufacturing processes are governed by transient, nonlinear and coupled partial differential equations, making repeated high-fidelity simulations computationally expensive during process optimisation. This work develops a Fourier-encoded, physics-augmented neural-network framework for modelling multi-physics problems in composites, with a focus on frontal polymerisation. The approach combines physics-informed loss functions, model-specific Fourier feature encoding and a neural-network architecture to predict coupled temperature, degree of cure and mechanical deformation. The framework is applied to frontal polymerisation, a rapid out-of-autoclave curing process governed by coupled heat transfer, cure kinetics and mechanical response. A thermo-chemical neural network predicts the evolution of temperature and degree of cure, while a second Fourier-encoded mechanical network uses these predicted fields to infer displacement. It is observed that Fourier encoding significantly improves the representation of transient behaviour compared with a conventional feed-forward neural network. The results show close agreement with direct numerical reference solutions and demonstrate accurate prediction of propagating cure fronts, front velocity and thermo-chemo-mechanical deformation across material and process parameters. The proposed Fourier-encoded surrogate provides an efficient route for optimising the energy-efficient manufacturing process.