A data-driven model for thermal error prediction considering thermoelasticity with gated recurrent unit attention

Abstract

It is of great significance to reduce the thermal error of machine tools. However, there is a time lag between different temperature measurement points due to the thermoelastic effect, which causes inaccurate prediction when using only the current temperature. In this paper, the GRU time series neural network with an attention mechanism is employed to establish the thermal error model of the screw, which uses historical data and sets different weights for them. In addition, since the thermal error is closely related to the working condition, the electronic control data that reflect the working condition in the CNC system are considered. Compared with several state-of-art methods, such as RNN and LSTM, the prediction results demonstrate the superiority of the proposed method. The actual machining indicates that the compensation rate exceeds 75% and can reduce the thermal error from 20 µm to 5 µm.