摘要
采用基于灵敏度分析的BP神经网络模型,将丁苯橡胶(SBR)复合材料的8种力学性能数据经过主成分分析(PCA)降维后作为神经网络的输入向量,耐磨性能数据作为输出向量,对SBR复合材料的耐磨性能进行预测,并计算各输入向量的灵敏度矩阵,从而分析输入量对耐磨性能的影响程度。结果表明:通过PCA降维处理,可以消除神经网络输入向量之间的共线性,简化网络,提高网络的预测性能;预测误差在允许范围内,说明BP网络适用于橡胶材料的耐磨性能预测;灵敏度分析显示定伸应力、拉断伸长率和拉断永久变形对SBR橡胶复合材料的耐磨性能影响最大。
A back propagation（BP） neural network model based on sensitivity analysis was established to predict abrasion of SBR composites.The data of eight kinds of mechanical properties of SBR composites were dimensionally reduced through principal component analysis（PCA）,and the PCA data were utilized as the input vectors while abrasion as the output vector of the BP network.Meanwhile,the sensitivity matrix of the input vector was calculated in order to analyze the influence of mechanical properties on abrasion.The results demonstrated that the co-linearity between the network input vectors could be eliminated by PCA and the network was simplified at the same time.The prediction error was within the allowable range,indicating that the BP network was suitable for SBR abrasion prediction.Sensitivity analysis indicated that the abrasion resistance of SBR was remarkably influenced by modulus,elongation at break and permanent deformation at break.
出处
《橡胶工业》
CAS
北大核心
2014年第2期69-73,共5页
China Rubber Industry
基金
新世纪优秀人才支持计划项目(NCET-10-0202)
关键词
丁苯橡胶
神经网络
主成分分析
耐磨性能
灵敏度分析
SBR
neural network
principal component analysis
abrasion resistance
sensitivity analysis
