基于进化神经网络的薄壁件加工变形预测与抑制方法

Optimized Neural Network Based Prediction and Control of Machining Deformation for Thin-walled Workpieces

加工变形的合理分析与精准控制可实现铣削过程的高效化和精密化。因此,首先依据实际的加工工况,建立了薄壁件铣削过程的有限元分析模型,通过薄壁T形件的铣削加工,验证了有限元仿真值与实验数据之间的吻合性。其次,利用正交实验设计方法和有限元方法获得输入样本与训练样本,通过将训练样本的预测误差作为适应度函数,采用遗传算法得到加工变形的进化神经网络模型,利用有限元仿真值比较表明进化神经网络的预测误差不超过6%。最后,以最小的加工变形为目标建立了刀具几何参数的优化模型,并依据加工变形越小染色体越健壮的原则,提出了刀具几何参数的遗传算法解算方法。在提高铣削变形计算效率的基础上,能够实现薄壁件铣削刀具的合理设计与选择。

In order to achieve the high efficiency and precision of milling process,it is very important to study the reasonable analysis and control method of machining deformation.Consequently,the finite element model is reasonably created to simulate the milling process of thin-walled workpiece.The milling of T thin-walled workpiece can validate the agreement of simulated values with the corresponding experimental data.And then,the finite element method is used to obtain the training samples in addition that the orthogonal experimental design method is adopted for the input samples.The genetic algorithm can skillfully be developed to formulize the back propagation(i.e.,BP)neutral network model by taking the prediction error of output samples as the individual fitness.It can predict the machining deformations of workpiece with less than 6%relative error.Finally,an optimal model of tool parameters is further suggested for the minimum machining deformation.In light of the principle in which the smaller the machining is deformation the stronger the chromosome is,the genetic algorithm can be employed to solve the optimal model of tool parameters.The presented method can realize the optimal design and selection of milling tool based on the improvement of the calculation efficiency.