Digital twin of an industrial wire bonder: Synergy between physics-based models and neural networks

To realise the potential of digital twins, the model that constitutes a digital twin
must accurately predict the (dynamical) behaviour of the physical system. Traditional (nonlinear) dynamical models that are derived from first principles, however, often miss relevant dynamics of the physical system. Therefore, this article introduces the Extension and Augmentation-based (EA) model-updating method, which synergises physics-based models, (closed-loop) measurement data, and AI techniques to create accurate (grey-box) EA models (i.e., digital twins). Applied to an industrial wire bonder, the EA model predicts dynamical (settling) behaviour with high accuracy, enabling improved positioning accuracy and throughput through model-based control design.