摘要
小波包分解方法在基于响应信号的结构损伤检测中被证明对损伤程度高度敏感,得到广泛的应用。在小波包分解中采用的是完全二进制分叉树型分解,而实际上在分解过程中部分子信号仅含有很少的信息量,对其再进一步分解是不必要的。通过引入熵的概念,可以对分解过程中的各层子信号进行选择,仅对含有足够丰富信息的子信号进行更进一步的分解。这样做可以有效减少最终所得子信号数目,在保持灵敏度的同时降低损伤指标的维数,有助于缩减损伤识别中神经网络的规模,对于大型复杂结构的损伤检测工作具有一定的意义。
It has been proved that the wavelet packets analysis (WPA) is sensitive to damage in response signals based damage detection, so WPA is widely used. The complete binary tree is adopted in WPA, but in fact there are some sub-signals which contain only few information and further decomposition of it will be not necessary. Entropy is adopted as the criteria to choose sub-signals with rich information in each level of decomposition. Then the number of finally obtained sub-signals is reduced and the dimension of the damage index is reduced while still keeping its sensitive to damage. It will reduce the size of artificial neural network (ANN) used in damage detectian. For large and complex structure, the reduction of ANN size will make considerable time saving in training the ANN.
出处
《机械强度》
EI
CAS
CSCD
北大核心
2007年第6期873-876,共4页
Journal of Mechanical Strength
基金
国家863项目基金(2006AA04Z437)~~
关键词
小波包分解
损伤检测
最优分解
Wavelet packet analysis
Damage detection
Optimal decoinposition