基于人工神经元网络预测增韧尼龙11拉伸行为

Prediction of Tensile Behavior of Toughing Nylon 11 Composite by an Artificial Neural Network Model

以马来酸酐接枝乙烯辛烯共聚物(POE-g-MAH)为相容剂,采用熔融共混法,制备了PA11/POE/POE-g-MAH三元共混物,利用ANN模型预测了三元共混物拉伸应力-应变曲线。POE-g-MAH与PA11原位生成嵌段共聚物,分散相POE在尼龙基体中的粒径在200~400 nm范围内,显著提高了材料的缺口冲击强度。当ANN模型采用3-47-1结构、隐含层和输出层传递函数分别为logsig和线性函数、优化算法为trainbr时,预测的三元共混物应力值与实验值的MSE最低,其值为2.43×10^(-7)。利用样品1、3、5、7的应力-应变曲线训练的ANN模型可以预测样品2、4、6的拉伸曲线,预测值与实验值的MSE在10^(-4)数量级,该模型的预测精度和泛化能力较好。

Maleic anhydride grafted ethylene octene copolymer(POE-g-MAH)was used as compatibilizer to prepare PA11/POE/POE-g-MAH ternary blends through melt blending.The tensile stress-strain curves of the ternary blends were predicted by an ANN model.Due to the in-situ formation of block copolymers between POE-g-MAH and PA11,the dispersed particle size of POE in the nylon matrix ranged from 200 to 400 nm,which could significantly improve the notch impact strength of the blends.When the ANN model structure was 3-47-1,with transfer functions of logsig and purelin for the hidden layer and output layer,and trainbr for the optimization algorithm,the predicted and experimental stress value of the ternary blend had a minimum MSE of 2.43×10^(-7).The ANN model trained by the stress-strain curves of samples 1,3,5 and 7 could effectively predict the tensile curves of samples 2,4 and 6.The MSE between the predicted and experimental values was on the order of 10^(-4),demonstrating good prediction accuracy and generalization ability of the ANN.