The Thermal Error Modeling with Deep Transfer Learning

Abstract

Thermal error of CNC machine tools is one of the main factors affecting the machining accuracy. The data-driven method for thermal error modeling is an effective and efficient, but they have some flaws, such as poor accuracy, bad robustness, and etc. because of having no quite enough data set and imbalanced data set. In this paper, a new method based on transfer learning for thermal error modeling is presented for solving the issue of imbalanced data set. The dataset of monitoring the temperature field of the machine tools includes monitoring data of three kinds of operating conditions, namely stopping, idling, and machining. When the fewer idling data is used to train a model, the larger stopping data are introduced as train aids. Transfer learning is adopted to fully learn the common characteristics of the two different working conditions, which can effectively solve the problem of imbalanced dataset. The experimental results prove that our method have better performance than other methods trained only with limited idling data.