基于MLP神经网络的双金属复合管弯曲壁厚减薄预测研究OA

To address the issue of the excessive wall thickness reduction during the bending process of the bimetallic composite tubes,which affects the forming accuracy and performance,a high-precision prediction model for wall thickness reduction rate has been established to provide a data-driven defect prediction solution.Taking copper/titanium(Cu/Ti)bimetallic composite bent pipes as the research object,a simulation scheme was designed with three process parameters of cross-sectional hollow coefficient,bending degree,and bending angle as variables,and experimental verification was carried out.A database source was established,followed by the establishment of a multi layer perceptron(MLP)neural network model.The influence of training sample size(N=100～808)and hidden layer node number(Node=10～300)on the performance of the machine model was analyzed.Results indicate that as the sample size N increases,the model performance shows a significant improvement under all node numbers.There is an optimal solution for the number of hidden layer nodes(N=808,Node=256),at which point the model performance indicators reach their optimum and its prediction accuracy reaches its highest(MAPE=6.694 6%).This indicates that the MLP prediction model constructed can accurately and efficiently map the complex nonlinear relationship between the bending process parameters of bimetallic composite tubes and the wall thickness reduction rate.The influence of sample size and the number of hidden layer nodes on the model performance is discovered,thus verifying the feasibility of machine learning methods replacing or assisting traditional finite element analysis in the field.