Physics-Informed Neural Network for prediction of post-harvest firmness of avocados

This study employs physics-informed neural networks (PINNs) to model avocado firmness during ripening, based on a physiological framework where firmness loss is governed by enzymatic cell wall degradation, with kinetics modulated by the metabolic rate. PINNs are trained on literature datasets from diverse cultivars and countries under constant storage conditions, revealing comparable climacteric ripening kinetics. The approach is extended to dynamic storage scenarios using a time/temperature/gas-conditions superposition principle. Variations due to orchard origin, climate, and harvest time are captured through differences in the initial pre-climacteric state, which cannot be described by a single state variable. • Physics-Informed-Neural-Networks applied to modeling Avocado firmness. • PhINNs can estimate common model parameters from multiple datasets. • Firmness shows universal behavior if rescaled to CO2 production. • Model needs extension for difference between CA and regular air storage.