Reducing thermal errors with confidence: Uncertainty-based compensation for precision machine tools

Abstract

This work presents a new probabilistic method for the compensation of thermal errors of precision machine tools. The basis of the proposed method is the Gaussian Process Regression (GPR) model combined with a threshold for the predicted standard deviation. The key advantage of the GPR model is that it not only provides point estimates for predictions but also quantifies the uncertainty associated with each prediction. Additionally, GPR combines the thermal error prediction and thermal key point selection into a single model, which considerably reduces the overall model complexity. Thermal errors are measured with the recently developed torque limit skip (TLS) thermal error measurement method for precision machine tools. When the applied threshold is exceeded, the model triggers a recalibration feedback loop using previously measured temperature and thermal error values measured with the TLS function. Results show that the self-recalibrating compensation model significantly reduces the thermal errors of the investigated machine tool.