摘要
采用自行搭建的声波信号响应装置对实验中裂纹鸡蛋样本的蛋壳裂纹随机分布进行无损检测。利用端点检测等方法对声波信号(采集点在鸡蛋赤道部位)进行预处理,然后采用Hilbert Huang变换(HilbertHuang transform,HHT)和多重主成分分析(Multi-PCA)对预处理之后的声波信号进行分析,分别提取声波信号在时域和频域上的主要特征参数,用于鸡蛋蛋壳裂纹的分类检测。结果表明,当鸡蛋蛋壳裂纹大小和位置均随机分布时,由HHT和Multi-PCA提取的特征参量经由支持向量机(support vector machine,SVM)模型和人工神经网络(artificial neural network,ANN)模型均可达到较高识别率。在SVM模型中,采用径向基核函数的效果最好,检测精度高达90%;在ANN模型中,整体回归系数可达0.924 76,检测精度为86.70%。
A self-built acoustic response system was used to non-destructively detect the eggshells cracks when the eggs were artificially cracked to produce various kinds of cracks.In order to minimize the dependence on the position cracked, the knocking point was always located near the eggshell equator. The original acoustic signals were pre-processing to detect endpoints of signals and then analyzed with Hil- bert-Huang transformation (HHT) and multiply principal component analysis (MPCA).Results showed that the accuracy of detection and classification of egg with cracks is very-well via a support vector machine (SVM) and artificial neural network(ANN) models since the main features of the time domain and frequency domain were extracted from HHT and Multi-PCA when the cracks have randomly distribu- tion.The radial basis kernel function is best for SVM and the accuracy for crack classification can be up to 90%.The whole regression coefficient is 0.924 76 with the accuracy of 86.70% for ANN model.It is indicated that the analytical methods of HHT and Multi-PCA are very suitable for detecting the cracked eggshell.
出处
《华中农业大学学报》
CAS
CSCD
北大核心
2017年第4期102-109,共8页
Journal of Huazhong Agricultural University
基金
湖北省自然科学基金项目(2015CFB479)
中央高校基本科研业务费专项(2662016PY059)
关键词
无损检测
蛋壳裂纹
HILBERT
Huang变换
多重主成分分析
支持向量机
人工神经网络
non-destructive detection
eggshell cracks
Hilbert-Huang transform
multiply principal component analysis
support vector machine
artificial neural network