Digital twin integration for dynamic quality loss control in fruit supply chains

Abstract

Effective cold chain management is imperative for minimizing food loss and maintaining quality in perishable logistics. This study integrates digital twin (DT) and artificial intelligence (AI) technologies to establish a “five-dimensional model” for cold supply chains, featuring a two-step approach that improve temperature prediction accuracy for shelf-life estimation. In the first step, a long short-term memory (LSTM) based model—trained solely on experimentally verified temperature data—accurately forecasts in-box conditions. Subsequently, a literature-based kinetic model applies well-established parameters to estimate remaining shelf life. By placing a single sensor at the pallet level and applying our box-level digital twin model, we achieved a temperature prediction error below ±0.3 °C (2σ), which translated into a shelf-life estimation error of under ±1.2 days for highly perishable fruits such as strawberries and lychees. Simulations also reveal the integrated DT–AI system reduces food loss by 8.6 %, 12.1 %, 13.6 %, and 15.5 % for strawberries, lychees, oranges, and apples, respectively, surpassing simpler ambient-based methods in both accuracy and food safety—particularly for highly perishable produce. Although hierarchical scaling of DTs (box, pallet, container) indicates increasing deviations at larger units, this trade-off between model precision and resource efficiency renders the solution practical across diverse cold-supply scenarios. Future work may incorporate end-point quality assessments and advanced management modules to further enhance reliability, reduce waste, and foster sustainability in global food logistics.