计算机应用与软件2012，Vol.29Issue(6)：72-75,4.
独立分量分析结合马氏距离的非监督损伤识别方法
UNSUPERVISED DAMAGE DETECTION BASED ON COMBINATION OF INDEPENDENT COMPONENT ANALYSIS AND MAHALANOBIS DISTANCE
摘要
Abstract
In structural damage detection, supervised learning methods are difficult in practice to acquire damage samples. To solve this limitation, a novel approach for estimating the structural damage based on combining independent component analysis (ICA) with Mahalanobis distance is presented in the paper. First, the ICA method is applied to extract independent source signals and mixing matrix, and the matrix is inputted as a feature parameter to Mahalanobis distance discriminant function. Then the threshold is designed according to Mahalanobis distance of healthy structural state. The compared result of Mahalanobis distances in regard to the threshold and the detection signal is used as damage judging reference. Vibration experiment is conducted on steel frame structure model in the action of impact load. Results show that the mixing matrix extracted by ICA method is an effective damage feature parameter, and the unsupervised learning method based on ICA and Mahalanobis distance can correctly detect the structural damage, therefore it provides an effective damage detection way for the structural health supervision.关键词
独立分量分析/马氏距离/混合矩阵/损伤识别Key words
Independent component analysis/Mahalanobis distance/Mixing matrix/Damage detection分类
信息技术与安全科学引用本文复制引用
曹军宏,韦灼彬,高屹..独立分量分析结合马氏距离的非监督损伤识别方法[J].计算机应用与软件,2012,29(6):72-75,4.基金项目
国家部委基金资助项目(BY208126) （BY208126）
海军工程大学自然科学基金青年资助项目. （）
