Modelling of drying kinetics of groundnuts in a convective cabinet tray dryer using physics-informed neural networks

Abstract

The drying kinetics of groundnut samples were experimentally investigated using a convective cabinet tray dryer under controlled operating conditions. Moisture ratio data obtained from experiments were used to develop drying kinetic models using physics-informed neural networks (PINNs), namely Standard PINN, Hybrid PINN, and Adaptive PINN architectures. In addition, six widely used thin-layer drying models were implemented for comparative evaluation against the PINN-based predictions. The drying rate was obtained directly from the predicted moisture ratio through automatic differentiation, thereby eliminating the need for separate experimental determination of rate of drying data. Among all evaluated approaches, the Adaptive PINN demonstrated superior predictive performance, achieving a coefficient of determination (R²) of 0.991, with a root mean square error (RMSE) of 0.028 and a mean absolute error (MAE) of 0.022. Furthermore, the effective moisture diffusivity increased with drying temperature, ranging from 2.14 × 10⁻¹⁰ to 3.64 × 10⁻¹⁰ m² s⁻¹ across the investigated air velocities. The activation energy for moisture diffusion ranged from 11.01 to 11.73 kJ mol⁻¹, indicating a moderate energy requirement for the drying process. The results demonstrate that physics-informed neural network (PINN) modeling, particularly the Adaptive PINN framework, provides an accurate and physically consistent approach for representing groundnut drying kinetics. The proposed methodology enables reliable prediction of drying behavior and offers a useful framework for drying process analysis and optimization.