Mechanism-informed friction-dynamics coupling GRU neural network for real-time cutting force prediction

Monitoring cutting force can provide the most reliable reference information for sensing machining state and supporting adaptive machining. The indirect monitoring approach based on real-time cutting force prediction using Computer Numerical Control (CNC) inherent servo signals has great potential to achieve long-term low-cost and accurate cutting force monitoring since no additional sensors are required. However, due to the complexity of the nonlinear dynamic relationship between the cutting force and servo signals, accurate real-time prediction of cutting force remains an open problem to be investigated in depth. To overcome this challenge, the excitationresponse relationship of the machine tool feed system disturbed by cutting force is analyzed first, and then the real-time cutting force prediction problem based on CNC inherent servo signals is defined as a nonlinear dynamic system modeling problem with adaptive time delay. On this basis, a novel mechanism-informed Friction-Dynamics coupling Gated Recurrent Unit (FD-GRU) neural network is proposed. The proposed FD-GRU neural network has two sub GRU neural networks for nonlinear dynamic friction estimation and cutting force inverse estimation respectively, and the mechanism relationship among all related variables is structurally preserved. In addition, a Kalman filter for estimating acceleration is embedded in the part of cutting force inverse estimation to provide more effective input information. Comparative verification was carried out through a set of hole milling experiments containing different combinations of cutting parameters. Compared with the purely data-driven ordinary GRU neural network, the proposed FD-GRU neural network can improve the prediction accuracy and generalization performance by nearly 30% and have better training stability. More importantly, it also can learn the clear mechanism meaning, which provides a strong explanation for its excellent performance.