A digital twin–driven dynamic shelf life approach for mitigating food loss across fruit supply chains

Static shelf-life labels poorly reflect real cold-chain temperature dynamics and box-level heterogeneity, leading to unreliable quality control and avoidable food loss. This study develops a box-level digital twin–enabled dynamic shelf-life framework that integrates multi-source temperature sensing with deep learning to reconstruct in-box product temperatures and predict shelf-life evolution for strawberries, lychees, oranges, and apples. The digital twin-enabled model captures thermal inertia and spatial variability within logistics units and is evaluated under four cold-chain deployment scenarios. Compared with air-temperature-based modeling, the digital twin-enabled approach substantially improves temperature prediction accuracy (RMSE and MAE reduced by ∼60%), yielding more physically consistent shelf-life trajectories, especially during disturbance-prone stages (postharvest handling and loading/unloading) and under cold-chain disruption. Sensitivity analysis shows that uncertainty in baseline deterioration kinetics dominates shelf-life prediction error, while temperature sensitivity plays a secondary, amplifying role. Overall, digital twin-enabled dynamic shelf-life approach provides a data driven framework for temperature-informed shelf-life management, supporting targeted interventions to reduce food loss in fruit supply chains.