Digital Twin Technology for Intelligent System Simulation and Optimization

Abstract

Digital twin technology — the creation of dynamic, continuously synchronised virtual replicas of physical systems that enable simulation, monitoring,
and optimisation without interrupting physical operations — has emerged as a foundational paradigm for intelligent system management across
manufacturing, infrastructure, healthcare, and urban domains. Despite substantial progress in digital twin platforms, the systematic integration of
real-time machine learning-driven state estimation, physics-informed simulation, and reinforcement learning-based optimisation within a unified
intelligent digital twin framework remains architecturally incomplete. This study presents TwinAI, a unified intelligent digital twin framework integrating
a Kalman-neural hybrid state estimator for real-time physical system tracking, a physics-informed neural network surrogate model for high-fidelity
simulation, and a multi-objective reinforcement learning optimiser for digital twin-in-the-loop control optimisation. TwinAI was deployed and validated
across three intelligent system applications: a wind turbine farm digital twin (Vattenfall Horns Rev 3, 49 turbines, offshore Denmark), a precision
manufacturing cell digital twin (Aalborg University CNC lab, 6-axis robot, 12 sensors), and a smart building energy management digital twin
(University of Genoa main campus, 47 buildings, 2,847 IoT nodes). TwinAI achieved state estimation RMSE improvements of 34.7% over Kalman
filter baselines, simulation fidelity R2 of 0.987, and optimisation objective improvements of 19.4% energy efficiency, 12.8% manufacturing yield, and
23.1% building energy reduction over rule-based baselines. Digital twin synchronisation latency averaged 87 ms. These results establish TwinAI as a
high-fidelity, responsive, and optimisation-capable intelligent digital twin framework across diverse physical system domains.