Industrial robots often experience significant inaccuracies in their movements due to joint friction, which is challenging to model accurately as it depends on various factors such as load, temperature, and time. Within this paper, we investigate the research question how analytical (white-box) and data-driven (black-box) friction models can be effectively combined with live data during the operation of an industrial robot. Existing research falls short in providing methods for continuously improving both white- and black-box friction models during operation. In this paper, we apply and refine a method to construct digital shadows that integrate structural and behavioral models, with operational data, and various live measurements to enhance our understanding of joint friction and ultimately improve robot accuracy. Our methodology involves modeling the industrial robot’s mechanical structure, incorporating friction-related behavioral, implementing Asset Administration Shells (AAS), and live-data linking and visualization in a dashboard. The resulting digital shadows enable the continuous, multi-physics-aware identification of joint friction, using both white-box and black-box models to enhance the robot’s position accuracy.