Physics-informed digital twins for sustainability: a data-efficient framework for physically consistent state tracking

Purpose This study presents a data-efficient and physically constrained digital twin framework for monitoring degradation and performance loss in engineered systems, with emphasis on sustainability-oriented applications where sensing availability, historical data, and computational resources may be limited. The objective is to establish a structurally interpretable and physically consistent approach to long-term degradation state tracking under sparse data conditions. Design/methodology/approach The proposed digital twin represents system sustainability using a compact internal degradation state governed by irreversible kinetic principles. Physical constraints, including monotonicity, boundedness, and asymptotic saturation, are embedded directly into the learning formulation to ensure physically admissible evolution. A lightweight neural network surrogate approximates the state trajectory while remaining constrained by the governing physical law. The framework is evaluated using five-fold cross-validation and validated against publicly available lithium-ion battery degradation datasets to assess robustness, reproducibility, and experimental consistency. Findings The results demonstrate stable and physically admissible state evolution across all validation folds, with degradation trajectories remaining monotonic, bounded, and insensitive to data partitioning. The physics-informed formulation achieves predictive accuracy comparable to purely data-driven approaches while preserving interpretability and structural consistency. Experimental validation confirms the ability of the framework to capture realistic degradation behaviour under sparse sampling without producing non-physical artifacts. Originality/value The framework embeds irreversible physical structure directly into a compact and identifiable degradation-state representation, prioritizing physical consistency, interpretability, and data efficiency. This provides a transparent and extensible foundation for sustainability-oriented digital twin implementation in intelligent engineering systems.