Abstract <p>Hydraulic motors are widely used in high-end equipment such as shield tunneling machines and dredgers due to their advantages of high torque density and low-speed stability. Accurate friction modeling is the foundation for understanding the strong nonlinear dynamic friction features of hydraulic motors and the key to control their stable operation. However, in low-speed and heavy-load conditions, the numerous friction pairs and complex interface lubrication behaviors in hydraulic motors cause significant challenges in establishing accurate friction torque models through physical models such as steady-state or dynamic friction models. For that, this study developed an innovative hybrid physics-data driven method to model the friction torque of hydraulic motors accurately, which is achieved by a modified LuGre physical model combined with a neural network data processing method named as modified LuGre guided neural network (MLuGre-GNN). This method could be divided into three parts: Firstly, a modified LuGre model suitable for the hydraulic motor is proposed as the physical model. Secondly, the steady-state part of the modified LuGre model is transmitted as prior physical features to guide data-driven modeling. Thirdly, a Micro-dynamic network and a Dynamic-friction network are constructed as the data-driven model for dynamic friction modeling. The above steps have achieved the integration of the physics-driven model and the data-driven model. Finally, the experiments are conducted to validate the effectiveness and accuracy of the hybrid physics-data driven modeling method. Compared to existing models, the hybrid modeling framework has significantly improved prediction accuracy, noise resistance, and robustness. The proposed method could offer a new way for accurate dynamic friction modeling of hydraulic motors.</p> Graphical abstract <p></p>