A Robotic Milling Posture Optimization Method Based on Physics-Informed Dynamic Stiffness Model and Graph Optimization Algorithm

In robotic milling, the robot’s positional error arises from insufficient stiffness or flexural rigidity. These errors can be mitigated by optimizing the robot’s posture to enhance stiffness and reduce the influence of flexing. Existing posture optimization methods are based on static stiffness models and fail to account for dynamic deflection induced by time-varying milling forces. To address this limitation, a physics-informed prediction model is proposed to predict modal properties of a robotic milling system. The tool-tip is modeled as a three-degree-of-freedom vibration system, which can be used for informing a posture optimization method to minimize the dynamic deflection. Specifically, the proposed prediction model captures the relationship between modal properties and robot configuration similarity by a customized kernel with a multioutput framework, while introducing physical constraints to tighten error bounds and enhance prediction accuracy. Additionally, a directed graph of robot configurations is constructed to support the posture optimization algorithm. This algorithm employs a novel cost function alongside a local optimization mechanism with uncertainty to optimize the milling posture, ensuring enhanced milling accuracy. Experimental results validate the effectiveness of the proposed method, demonstrating improvements in contour accuracy by 32.7% and 35.8%, respectively, for two different skin parts compared with baseline methods.