| 摘要: |
| 精确预测薄壁件铣削过程中工件的时变动力学参数是选择无颤振切削参数的基础。本文提出了基于三层神经网络的曲面薄壁件铣削时变动力学参数的预测方法。首先利用壳单元将薄壁件进行离散化，将离散单元结点处工件厚度作为输入参数，工件前三阶固有频率作为输出参数，构建三层神经网络。然后将壳单元有限元模型计算结果作为训练样本，训练得到神经网络模型。模态测试实验结果表明，有限元模型对工件固有频率预测的最大误差约为4%；与有限元模型相比，神经网络模型的最大预测误差为0.409%，因此神经网络模型的最大预测误差约为4%。同时，神经网络模型训练时间约为10 s；当预测的切削状态数量为150时，预测时间仅为0.002 s。在保证计算精度的前提下，三层神经网络模型可大幅提高计算效率。 |
| 关键词: 薄壁件 铣削颤振 动力学参数 壳单元 神经网络 |
| 基金项目: |
|
| Predicting the Dynamic Parameters of the Workpiece during Milling of Thin-walled Parts Based on Neural Network |
|
Wang Dazhen Li Qi Wang Liang
|
|
Xi’an Institute of Aerospace Power Measurement and Control Technology, Xi’an 710025
|
| Abstract: |
| Accurately predicting the time-varying dynamic parameters of workpieces during milling of thin-walled parts is the basis for selecting chatter-free cutting parameters. This paper proposes a three-layer neural network based method for predicting time-varying dynamic parameters of workpiece during milling of curved thin-walled parts. Firstly, the thin-walled part is discretized using shell elements, and the thickness of the workpiece at the discrete element nodes is taken as the input parameter, while the first three natural frequencies of the workpiece are taken as the output parameters to construct a three-layer neural network. Then, the results of finite element model built by the shell element are used as training samples to train the neural network model. Modal testing results show that the maximum error of the finite element model in predicting the natural frequencies of the workpiece is about 4%. Compared with the results of finite element model, the maximum prediction error of the neural network model is 0.409%. Therefore, the maximum prediction error of the neural network model is about 4%. At the same time, the training time of the three-layer neural network model is approximately 10 s. When the number of predicted cutting states is 150, the prediction time is only 0.002 s. The three-layer neural network model can greatly improve computational efficiency while ensuring calculation accuracy. |
| Key words: thin-walled parts milling chatter dynamic parameters shell elements neural networks |
