In-process identification of feed drive dynamics considering machining forces

This paper presents a new closed-loop dynamics model for ballscrew feed drives in CNC machine tools, enabling non-intrusive, in-process model calibration and motion prediction even in the presence of unmeasured machining forces. The presented model employs a Partially Linear Auto-Regressive with Exogenous input (PL-ARX) structure, where the linear component captures the servo drive and rigid-body dynamics, and the nonlinear component represents unknown machining forces. Kernel-based regression is then used to simultaneously identify the linear dynamics and machining force disturbances from internal controller signals during milling. The model is validated on two different CNC machines under experimental milling conditions. Results confirm the approach accurately identifies unbiased linear dynamics despite unmeasured disturbances and achieves precise online motion prediction. These capabilities are critical for enabling real-time feedrate optimization and model-predictive control in advanced machining systems. • Unbiased feed drive identification despite unmeasured machining forces. • Cutting force disturbance prediction using only velocity data. • Recursive Gaussian process regression for sensor-less disturbance prediction. • Experimental validation through milling operation tests.