Modeling of Temperature-dependent Joint Friction in Industrial Robots Using Neural Networks

The application of industrial robots (IR) in machining offers many potential advantages such as flexibility and a large workspace. However, IR show weaknesses in absolute and path accuracy compared to machine tools due to their functional structure. Model-based compensation techniques can be a possible solution, which requires precise modeling of the robot dynamics and its influences. Friction causes a large portion of the total torque, particularly at low speeds showing highly nonlinear behavior. Furthermore, friction is caused by many influencing variables that are not considered in simple analytical models, such as temperature or load. The aim of this work is to propose analytical and data-driven friction models that adapt to the change in joint temperatures due to prolonged robot movements. As observed with experiments, the effect of temperature rise is significant on friction when the robot is operated over a long duration. Robot trajectories, with varying velocities and temperature rise rates inside the joints, are generated and implemented on the first axis of the MABI MAX 100 robot for training of the proposed friction models. While trajectory execution, additional sensors on the robot are used to record the robot housing temperature and the environment temperature in the robot cell. As a result, the temperature-dependent as well as data-driven models outperformed currently used models.