考虑电控数据的LSTM神经网络进给轴热误差建模

Thermal error modelling for machine tool feed axis based on LSTM neural network considering electro-control data

针对进给轴热误差建模中忽略电控数据和时间序列影响的问题,提出一种考虑温度变化与电控数据的长短期记忆(Long-Short Term Memory,LSTM)神经网络热误差预测模型。以三轴立式加工中心为试验对象,首先对进给轴进行热变形分析,再以温度变化、电控数据为输入样本,建立了LSTM神经网络热误差预测模型,随后通过与仅考虑温度变化的LSTM神经网络,以及同时考虑温度变化与电控数据的BP神经网络进行对比分析,试验论证表明,对数控机床进给轴进行热误差建模时,在考虑温度变化的基础上,进一步考虑电控数据可以提高模型的预测精度和鲁棒性,且在同样输入条件下,LSTM神经网络热误差预测模型相较于BP神经网络有更好的预测精度和鲁棒性。

Aiming at the problem of ignoring the influence of electro-control data and time series in the thermal error modelling of the feed axis,an LSTM neural network thermal error prediction model considering temperature change and electro-control data was proposed.It took a three-axis vertical machining centre as the test object.Firstly,the thermal deformation analysis of the feed axis was carried out.Secondly,the LSTM neural network thermal error prediction model was established with the input samples of temperature change and electro-control data.Finally,this model was compared with LSTM neural network model considering temperature change,and BP neural network model considering temperature change and electro-control data.The results showed that,in the thermal error prediction modelling,considering the temperature change and the electro-control data can improve the thermal error model′s prediction accuracy and robustness.And under the same input conditions,LSTM neural network thermal error prediction model had better prediction accuracy and robustness than the BP neural network.