%0 Journal Article %T UHMWPE/LDPE层合板复合材料损伤声发射信号识别 %T Pattern Recognition of Damage Modes in UHMWPE/LDPE Composites Laminates by Acoustic Emission Technique %A 王旭 %A 杜增锋 %A 倪庆清 %A 刘新华 %A WANG,Xu %A DU,Zengfeng %A NI,Qingqing %A LIU,Xinhua %J 宇航材料工艺 %J Aerospace Materials & Technology %@ 1007-2330 %V 49 %N 2 %D 2019 %P 77-81 %K 复合材料;声发射;损伤机理;聚类分析;神经网络 %K Composite material;Acoustic emission;Damage mechanism;Clustering analysis;Neural network %X 为了掌握UHMWPE/LDPE复合材料的损伤机理,运用声发射技术结合聚类分析方法建立不同损伤形式的声发射信号训练样本,通过神经网络实现损伤信号的识别,并分别探讨了训练函数、传递函数、网络结构等因素对识别率的影响。研究表明,由系统聚类可提取幅度、峰值频率、持续时间为模式特征,结合K-means聚类可建立11个类别共583信号的训练样本。以混淆矩阵为识别率指标,当训练函数为traingdx、隐层/输出层传递函数为tansig/logsig、隐层神经元数量为70时,网络的识别率达97.2%,为基于声发射技术的热塑性基体复合材料损伤识别提供参考。 %X In order to master damage mechanism of UHMWPE/LDPE composite material,training sample of acoustic emission signal with different damage modes was established by clustering analysis and acoustic emission technology. Acoustic emission signal generated from different damage modes were identified by neural network. The factors affecting the recognition accuracy of network such as training function, transfer function and network architecture were discussed respectively. The results revealed pattern characteristic consisting of amplitude, peak frequency and duration can be selected by hierarchical clustering method. The training sample consisting of 583 signals with 11 classes can be established by K-means clustering method. Using the confusion matrix as the recognition accuracy index, when the training function is traingdx, the hidden layer/output layer transfer function is tansig/logsig, and the number of hidden layer neurons is 70, the recognition accuracy of the network is 97.2%. The results provide reference for the damage identification of thermoplastic matrix composites based on acoustic emission technology. %R 10.12044/j.issn.1007-2330.2019.02.015 %U http://www.yhclgy.com/yhclgy/home %1 JIS Version 3.0.0