Digital twins for food processing and shelf-life: linking mechanistic models, sensors, and AI to predict quality, safety, and waste through near-real-time decision support

Abstract

Hybrid digital twins (DTs) are emerging as decision-support systems for perishable food supply chains because they can transform shipment-specific exposure histories into individualized forecasts of quality evolution, safety risk, and remaining shelf-life (RSL). This paper presents a focused hybrid DT framework for modified-atmosphere-packaged (MAP) fresh-cut produce under refrigerated cold-chain conditions. The framework is centered on three core elements: (i) temperature-linked state estimation, (ii) mechanistic modeling of MAP headspace dynamics together with microbial and quality kinetics, and (iii) near-real-time prediction of remaining shelf-life for inventory prioritization. A staged pilot-to-implementation protocol is described, including instrumented data capture, sensor calibration, hybrid model commissioning under operational implementation-faithful validation, defined update intervals, synchronization of model states with incoming observations, and near-real-time decision-support outputs for FEFO reprioritization, rerouting, setpoint adjustment, and diversion/hold decisions. Collectively, the framework clarifies how improved state estimation tightens RSL uncertainty, enabling earlier and more targeted interventions that can reduce spoilage and improve service performance. The paper concludes with adoption requirements, including governance, calibration maintenance, update-cycle design, and decision integration, as well as research priorities such as benchmarking, domain-shift robustness, uncertainty-calibrated decision rules, and MAP observability design.