<p>Digital twin models are of great significance to the performance optimization, predictive maintenance, and operation efficiency improvement of robot systems. However, robot systems integrate multiple disciplines such as mechanics, electronics, and control, and this complexity poses great challenges to the construction of high-precision digital twin models. Digital twin models of robots constructed by traditional modeling methods generally suffer from limited accuracy, and the models are isolated from each other, resulting in low reusability and poor scalability. These limitations create significant difficulties in the process of maintenance, adjustment and upgrading. To address these problems, this paper proposes a comprehensive framework for high-fidelity digital twin modeling. By leveraging the Modelica language, the framework realizes unified modeling of multi-domain physical systems. It supports efficient model reuse and expansion through modular design with standardized interfaces, encapsulated components, and hierarchical assembly strategies. Crucially, the framework integrates data-driven models with mechanistic models, thereby substantially enhancing the overall model accuracy. Experimental results on an HSRJR605 6-DOF industrial robot show that the proposed approach achieves high fidelity, with the average Root Mean Squared Error (RMSE) reduced by 87.3% and the average Mean Absolute Error (MAE) reduced by 89.2% compared with a pure mechanism model. The framework proposed in this paper can not only accurately map the physical behavior of the robot system but also improve the maintainability and scalability of the model through modular design, providing methodological support with both theoretical completeness and practical applicability for the engineering implementation of robot digital twin systems.</p>