CAJ | 학술논문

以铝板为研究对象,首先基于超声仿真技术,设计了"单发多收"超声阵列探头,对位于铝板内部不同深度处的21类缺陷分别进行建模和仿真分析,并获取相应的时域信号;其次,基于主成分分析(principal component analysis,PCA),对各缺陷的幅频初始特征进行提取,获得各缺陷的特征向量;最后,采用概率神经网络(probabilistic neural network,PNN)对各缺陷进行定位分析. 研究分别以21×9和21×4个缺陷为训练样本和测试样本, 分析结果表明: 缺陷定位的平均正确率为82.14%,96.43%, 100%. 研究论证了采用超声阵列探头,并结合主成分分析和概率神经网络,进行缺陷定位的有效性.
이려판위연구대상,수선기우초성방진기술,설계료"단발다수"초성진렬탐두,대위우려판내부불동심도처적21류결함분별진행건모화방진분석,병획취상응적시역신호;기차,기우주성분분석(principal component analysis,PCA),대각결함적폭빈초시특정진행제취,획득각결함적특정향량;최후,채용개솔신경망락(probabilistic neural network,PNN)대각결함진행정위분석. 연구분별이21×9화21×4개결함위훈련양본화측시양본, 분석결과표명: 결함정위적평균정학솔위82.14%,96.43%, 100%. 연구론증료채용초성진렬탐두,병결합주성분분석화개솔신경망락,진행결함정위적유효성.
Taking aluminum plate as the research object, based on ultrasound simulation techno logy, this study firstly designs the array probe of single-launch-multiple-reception model, which simulates 21 kinds of defects located at different depths inside the aluminum plate to obtain the corresponding time domain signals. Secondly, it extracts initial features of amplitude frequency of all defects to gain each defect feature vector according to principal component analysis. Finally, the probabilistic neural network is used to make location analysis for each defect. The training samples of defects and the test samples of defects are explored. Results show that the average accuracy for locating defects is 82.14%, 96.43% and 100% respectively. The research demonstrates effectiveness of the defect localization by using ultrasonic array probe combined with principal component analysis and probabilistic neural network.