基于混沌演化和CNN-GRU的机床热误差建模

Modeling for Machine Tools Thermal Error Based on Chaotic Evolution and CNN-GRU

为充分利用温度数据中时空特征的联系,提高机床热误差的预测精度,提出了一种基于卷积神经网络(convolutional neural network,CNN)和门控循环单元(gated recurrent unit,GRU)的热误差预测方法。首先,采用混沌演化(chaotic evolution,CE)重构温度测量数据的相空间,以机床整体温度信息、气候环境和工件误差特征集作为模型输入,演化出机床更深层次的信息;其次,利用CNN提取热误差与温度信息在高维空间的联系,构造具有空间特征的时序向量;最后,通过GRU捕获其时序特征并输出热误差预测值。使用该方法对机床热误差进行预测实验,并与单纯型CNN-GRU模型、CNN模型相比,在预测精度与泛化性方面具有明显优势。

A thermal error prediction method based on convolutional neural network and gated recurrent unit was proposed to fully exploit the connection of spatio-temporal features in temperature data and improve the prediction accuracy of machine tools thermal errors.The chaotic evolution method was adopted to reconstruct the phase space of the temperature measurement data to evolve the deeper information of the machine tools,it took the overall temperature information of the machine tools,the climate environment,a nd the set of the workpiece error characteristics as the model input.The relationship between thermal error and temperature information in the high-dimensional space was extracted by the CNN,the time series vector with spatial characteristics was constructed.Time series features were captured by GRU and the thermal error prediction value was output.Compared with the monotypic CNN-GRU model and CNN model,this method has significant advantages in prediction accuracy and generality for experiments on predicting thermal errors in machine tools.