Compensation of thermal drift during the single-point diamond turning process based on the LSTM

Abstract

In this paper, we propose a compensation method for the nanometer level of thermal drift by adopting long-short term memory (LSTM) algorithm. The precision of a machining process is highly affected by environmental factors. Especially in case of a single-point diamond turning (SPDT), the temperature fluctuation directly causes the unexpected displacement at nanometer scale between a diamond tool and a workpiece, even in the well-controlled environment. LSTM is one of the artificial recurrent neural network algorithms, and we figure out that it is quite suitable to predict the temperature variation based on the history of thermal fluctuation trends. We monitor the temperatures at 8 spots nearby a SPDT machine, and the neural network based on LSTM algorithm is trained to construct the thermal drift model from the time series data. Results of thermal drift prediction showed that the proposed method gives an effective model upon the well-controlled laboratory environment, and by which the thermal drift can be compensated to improve the precision of SPDT process.