基于人工神经网络的碳/陶瓷复合材料性能预测

利用人工神经网络的BP算法,建立了碳/陶瓷复合材料性能与多组分掺杂含量之间的预测模型.模型由输入层、隐含层和输出层3层神经元组成,用以模拟人脑的结构.以掺杂物的质量分数为输入参数,经石墨化后测得的复合材料的电阻率和抗折强度为输出参数.选取了30组实验数据作为学习样本,任意的7组数据作为"未知样品"对网络进行验证.结果表明,实验值和预测值相比电阻率的最大误差不超过8%,抗折强度的最大误差不超过12%.所建的网络可为碳/陶瓷复合材料设计提供理论指导.

Property prediction of carbon-ceramics composite material based on artificial neutral network

Models were established to predict the relation between components and properties of carbon-ceramics composite material based on the back propagation (BP) algorithm of the artificial neural network (ANN). The prediction models are composed of three neuron layers, i.e. input layer,hidden layer and output layer, to simulate the real structure of human brain. The volume percentages of the components are regarded as input parameters and the resistivity and antiflex strength of the composite material after graphitizing are regarded as output parameters. The selected thirty samples are considered as the data of the study and the random seven samples are predicted and assessed in the artificial neutral network of BP. On condition that the training data are precise enough,the models provide good results for the relation between components and properties of carbon-ceramics composite material. The electric resistance and benging strength error are respectively within 8% and 12% compared with experimental data. Therefore the models proposed are helpful to the design of carbon-ceramics composite material systems.