Physics-informed digital twin for sucrose loss prevention and transparent governance in regional sugar beet supply chains

Abstract

Objectives: This study aims to develop and validate an intelligent Digital Twin architecture that integrates Physics-Informed Neural Networks (PINN) for predictive sucrose degradation forecasting with fairness mechanisms for independent agricultural producers and transparent data governance in regional sugar beet supply chains. Prior work: Traditional logistics models treat agricultural cargo as static objects, ignoring the non-linear biochemical degradation of root crops during temporary field storage and the information asymmetry between producers and processing enterprises. While recent contributions in smart regional development have emphasised the importance of E-Governance and institutional transparency, existing quality forecasting and equity balancing models have evolved separately, resulting in fragmented decision-support systems. Approach: The proposed framework integrates three complementary modules: (i) a PINN embedding Arrhenius-based degradation kinetics for predictive sucrose loss estimation, trained on a historical industrial dataset and validated against the industrial baseline of constant daily losses. By resolving the quality observability gap through PINN-based hidden-state estimation and by reconciling delayed post-reception laboratory results with predictive estimates, the architecture ensures continuous adaptation without accumulating systemic drift error; (ii) a multi-period Growers' Equity balancing algorithm with historical compensation, designed to prevent systematic discrimination over a multi-year horizon; and (iii) an event-driven microservice architecture with an RFID-enabled transactional pipeline for end-to-end data transparency and macro-anomaly detection. Results: In a controlled proof-of-concept benchmark based on norm-derived historical degradation trajectories, the PINN reduces the Mean Absolute Error of sucrose concentration reconstruction by a factor of 12 compared to the simple averaging heuristic (0.031% vs. 0.383%). This indicates that explicit temperature-dependent kinetics substantially improves hidden-state reconstruction under sparse laboratory observations. Implications: By resolving the pre-reception quality observability gap through PINN-based hidden-state estimation and the post-reception analytical lag through retrospective reconciliation, the architecture ensures continuous adaptation without accumulating systemic drift error. For practitioners, this provides a transparent, empirically grounded decision-making environment; for researchers, it demonstrates that physics-informed machine learning can be successfully coupled with deterministic optimisation to enable proactive quality preservation and institutional fairness. Value: The scientific novelty lies in the integration of physics-informed machine learning with logistical optimisation and institutional fairness mechanisms, supported by empirical validation on an 11-year industrial dataset. By eliminating information asymmetry and creating a transparent cyber-physical space, the Digital Twin provides a robust foundation for E-Governance and sustainable development in regional agro-industrial clusters.