A Gaussian process regression-based surrogate model of the varying workpiece dynamics for chatter prediction in milling of thin-walled structures

Abstract

Since the dynamics of thin-walled structures instantaneously varies during the milling process, accurate and efficient prediction of the in-process workpiece (IPW) dynamics is critical for the prediction of chatter stability of milling of thin-walled structures. This article presents a surrogate model of the IPW dynamics of thin-walled structures by combining Gaussian process regression (GPR) with proper orthogonal decomposition (POD) when IPW dynamics at a large number of cutting positions has to be predicted. The GPR method is used to learn the mapping between a set of the known IPW dynamics and the corresponding cutting positions. POD is used to reduce the order of the matrix assembled by the mode shape vectors at different cutting positions, before the GPR model of the IPW mode shape is established. The computation time of the proposed model is mainly composed of the time taken for predicting a known set of IPW dynamics and the time taken for training GPR models. Simulation shows that the proposed model requires less computation time. Moreover, the accuracy of the proposed model is comparable to that of the existing methods. Comparison between the predicted stability lobe diagram and the experimental results shows that IPW dynamics predicted by the proposed model is accurate enough for predicting the stability of milling of thin-walled structures.