Digital twins in chemical engineering: An integrated framework for identification, implementation, online learning, and uncertainty assessment

A digital representation of a cyber–physical system (CPS) offers significant opportunities, containing dynamic process tracking in real-time, inferring parameters, and enabling constant learning to refine insights and predictive capabilities. This concept, referred to as a digital twin (DT), is increasingly applied in multiple sectors of industries but faces distinct issues depending on the application. In chemical engineering, for example, discrepancies between physical processes and models often arise due to unobservable state changes, while the complexity of these systems demands simplifications that introduce uncertainties. In addition, resource constraints in industrial environments further complicate implementation. This study presents a comprehensive framework for developing and implementing DTs in process units, with applications demonstrated in a gas-lift system and an electrical submersible pump (ESP). The framework incorporates techniques such as online learning, transfer learning, Bayesian inference, and Monte Carlo simulations, along with innovative methods such as dimensionality reduction and enhanced cognitive strategies to ensure adaptive capabilities. By addressing gaps in uncertainty management and integrating advanced learning methodologies, this framework supports the creation of reliable and flexible DTs, capable of operating efficiently in dynamic and complex environments. • A digital twin framework integrates AI for real-time process monitoring and prediction. • Uncertainty management enhances digital twin reliability in dynamic environments. • Gas lift and ESP case studies validate the framework’s adaptive learning approach. • Online learning and transfer learning improve model accuracy over time. • Statistical model reduction optimizes computational efficiency for real-time use.