神经网络在制备氮化硅多孔陶瓷中的应用  

Artificial Neural Network Modeling and Analysis of Preparation of Porous Si_3N_4 Ceramics

作　　者：余娟丽[1] 王红洁[1] 张健 严友兰[1] 乔冠军[1] 金志浩[1] 

机构地区：[1]西安交通大学金属材料强度国家重点实验室,陕西西安710049 [2]先进功能复合材料国家重点实验室,北京100076

出　　处：《稀有金属材料与工程》2010年第3期464-468,共5页Rare Metal Materials and Engineering

基　　金：武器装备预研基金(9140C5602040805);新世纪优秀人才支持计划(NECT-05-0838);"973"项目(2006CB601201)

摘　　要：以凝胶注模法制备多孔氮化硅陶瓷正交试验结果作为样本,建立3层Back Propagation(BP)神经网络,并进行训练以预测陶瓷性能。通过附加试验值对建立的神经网络预测能力进行验证,证明该BP神经网络模型是有效的,能准确预测多孔氮化硅陶瓷性能。通过BP神经网络模型研究多孔氮化硅陶瓷性能的结果表明,随着固含量的增加,气孔率单调下降;固含量存在一优化值,此时陶瓷抗弯强度最大;单体含量越大,气孔率越大,而抗弯强度降低。Based on orthogonal experimental results of porous Si3N4 ceramics by gel casting preparation, a three-layer back propagation （BP） artificial neural network （BP ANN） was developed for prediction of the flexural strength and porosity. The BP ANN is composed of three neurons in the input layer, two neurons in the output layer and six neurons the hidden layer. This study demonstrates that the proposed neural network approach can predict the performances of porous Si3N4 ceramics by accuracy, and the neural network is a very useful and accurate tool for performances analysis gel casting preparation to a high degree of of porous Si3N4 ceramics. By the proposed neural network prediction and analysis, the results suggest that the porosity monotonically decreases with the increase of solid loading, flexural strength is low when solid loading was too low or too high, and flexural strength has an optimum value.

关 键 词：神经网络 多孔氮化硅陶瓷 抗弯强度 气孔率 

分 类 号：TQ174.1[化学工程—陶瓷工业]