基于多传感器信息融合的SVM结构损伤诊断方法

A structure damage diagnosis method based on multi-sensor information fusion and support vector machine

为了能准确地诊断复杂结构损伤的发生、位置和程度,提出了一种聚类经验模式分解(EEMD)、小波包分解(WPT)、多传感器信息融合和SVM模式分类相结合的结构损伤诊断方法.首先对多个传感器采集的加速度振动信号进行EEMD分解,选择包含结构损伤信息丰富的固有模态函数(IMF);其次对其进行正交小波包分解,并计算小波包相对能量分布;最后把这些传感器信号的小波包相对能量融合,构成SVM分类器的输入特征向量,从而实现损伤的诊断和评价.研究结果表明:该方法在学习样本数较少的情况下仍然具有较好的适应性和分类能力;多传感器信息融合技术减小了损伤检测信息的不确定性,提高了损伤诊断准确率.

In order to make a diagnosis of damage occurrence, position and degree of the complex structures accurately, a structural damage diagnosis method was presented by means of ensemble empirical mode decomposition （EEMD）, wavelet packet decomposition, and multi-sensor feature fusion theory and support vector machine （SVM） pattern classification. Firstly, the response signals of the ASCE benchmark structure are processed by using EEMD, and the intrinsic mode function （IMF） which contains structural damage information are selected. Secondly, the selected IMF is decomposed by ortbogonal WPT, and also wavelet package energy （WPE） on decomposition frequency bands are calculated. Thirdly, the input feature vectors of SVM classifier were built by fusing wavelet packet relative energy distribution of these sensors. Finally, with the trained classifier, damage diagnosis and assessment was realized. The result indicated that it still has good adaptability and classification capability in the case of small samples and the fused feature can reduce the uncertainty of damage detection information, with the diagnosis accuracy improved.