Sim2Know: new paradigm of digital twins to design and inform human-centric knowledge system

Abstract

The novel framework, Sim2Know, tackles two major challenges in adaptively designing and informing a human-centric knowledge system: the lack of labeled real-world training data and the difficulty of capturing implicit knowledge. First, a digital twin demonstrator is developed to generate high-quality synthetic training data. Next, we propose a hybrid training approach that combines transfer learning from pre-trained self-supervised models with synthetic data augmentation, achieving a precision rate of 90.31 % in identifying 11 essential human action patterns in metal additive manufacturing. Finally, the human-centric knowledge system is designed to capture implicit knowledge through contextualizing human machine interaction beyond explicit domain knowledge.