基于LSTM循环神经网络的数控机床热误差预测方法

A thermal error prediction method for CNC machine tool based on LSTM recurrent neural network

针对传统热误差预测中忽略了机床历史累积温度状态与机床热误差之间的关联关系,提出一种基于长短期记忆(LSTM)循环神经网络的数控机床热误差预测方法。LSTM循环神经网络可以有效利用机床当前时刻和历史时刻的温升数据来表征更加符合机床热变形机制的热误差。以一台精密卧式加工中心为例,首先进行热误差实验,然后利用模糊c均值(FCM)聚类算法从20个温度点中筛选出4个关键温度点,再以其温升数据为输入热误差数据为输出建立LSTM循环神经网络热误差预测模型。最后,在不同工况下与传统热误差预测模型进行预测性能对比分析,结果表明所提热误差预测方法预测精度最高提高约52%,具有更加优越的预测精度和泛化性能。

The traditional thermal error prediction ignores the relationship between the machine tool historical cumulative temperature state and the thermal error.To solve this problem,a thermal error prediction method for CNC machine tool based on long short term memory(LSTM)recurrent neural network is proposed.The LSTM recurrent neural network can effectively use the temperature rise data both at the current and historical moments of the machine tool.In this way,the thermal error in the thermal deformation mechanism can be characterized.Taking a precision horizontal machining center as an example,the thermal error experiments are first conducted.Then,4 key temperature points are selected from 20 temperature points using the fuzzy c means(FCM)clustering algorithm.By regarding the temperature rising data of the key temperature points as input and thermal error data as output,the LSTM recurrent neural network thermal error prediction model is formulated.Finally,the prediction performance is analyzed and compared with traditional thermal error prediction models under different working conditions.Experimental results show that the proposed thermal error prediction method has a maximum increase of about 52%in prediction accuracy,which verifies its advantages of accuracy and generalization performance.