论文部分内容阅读
为了解决结构损伤诊断中缺乏损伤样本的问题,提出了一种基于支持向量数据描述(SVDD)的结构损伤诊断新方法。该方法只需正常状态的数据样本,而不需要任何损伤数据。首先采用小波包分解对数据预处理,以频带能量序列为特征,然后把多测量点的能量序列融合后作为特征向量,输入到SVDD分类器进行训练,实现损伤的诊断。实验表明,采用频带能量降低了数据维数,有效地浓缩了损伤信息;多测量点数据融合不仅能够使不同传感器的信息相互补充,而且减小了检测信息的不确定性;SVDD分类器较好地区分了结构正常与非正常状态,达到损伤自动诊断的目的。
In order to solve the problem of lack of damaged samples in structural damage diagnosis, a new method of structural damage diagnosis based on Support Vector Data Description (SVDD) is proposed. This method requires only normal state data samples without any damage data. Firstly, wavelet packet decomposition is used to preprocess the data, which is characterized by frequency band energy sequence. Then, the energy sequences of multiple measurement points are fused and used as eigenvectors, which are input to SVDD classifier for training to realize the damage diagnosis. Experiments show that using band energy reduces the dimensionality of data and effectively concentrates the damage information. Fusion of multiple measurement points not only makes the information of different sensors complement each other, but also reduces the uncertainty of detection information. SVDD classifier is better The area is divided into normal and abnormal structure to achieve the purpose of automatic diagnosis of injury.