Online Continual Learning Based Real-Time Dynamic Digital Twin Model in Industrial Systems

Digital twin (DT), as a new digital technology, has been widely used to improve the ability of industrial applications such as monitoring and fault diagnosis, by generating high-fidelity models of physical objects. However, existing DT models of industrial equipment cannot dynamically update with changing environments, particularly when it comes to utilizing real-time data streams. In this paper, we propose an edge-based DT framework in industrial systems. For dynamically updating the DT model of the equipment with real-time data streams, we design an online continual learning (OCL) based on DT update method. Meanwhile, an algorithm based on online experience replay (OER) is proposed to address the catastrophic forgetting problem of the proposed DT update method. Numerical results show the effectiveness of the proposed methods.