摘要
针对结构健康监测系统中的传感器数量多、数据信息复杂的特点,从模式识别和局部控制、全局参与的思想出发,提出了多传感器信息融合方法对结构损伤进行识别。首先应用小波包变换对结构振动测试数据进行特征提取,通过不同传感器特征向量的合成完成数据层融合;然后建立三个耦合神经网络分别实现结构损伤的确认、定位及定量,并完成决策层的信息融合;最后进行了36个损伤工况的结构模型实验研究,验证了所提出的方法是可行的和有效的。从实验验证的结果来看,对损伤率在7.5%以上的结构,损伤识别精度较高;对于损伤确认和损伤定位,识别精度较高,而对于损伤程度识别有一定偏差。
For the reason that there are too many sensors and data structural health monitoring system and based on the ideas of pattern classification and local decision referring global information, a method of multi-sensor information fusion is proposed to conduct damage identification in the paper. Firstly, Wavelet packet transform is introduced to extract features of vibration measured data and information fusion of data layer is conducted by assembling feature vectors of different sensors. Secondly, three coupling neural networks are constituted to realize damage validity, damage localization and damage quantification and to complete information fusion of decision layer. Finally, an experiment of three-story frame structure is conducted to prove the validity and feasibility of the proposed method by simulating 36 damage conditions. As a result, there is a good precision to identify damage of rigidity reductions of 7.5% or more. A good result in damage validity and damage location is obtained, and the error in damage quantification is small.
出处
《振动工程学报》
EI
CSCD
北大核心
2005年第2期155-160,共6页
Journal of Vibration Engineering
关键词
结构振动
损伤识别
信息融合
小波包
耦合神经网络
Data processing
Identification (control systems)
Neural networks
Sensor data fusion
Sensors
Structural frames
Structures (built objects)
Vibrations (mechanical)
Wavelet transforms
