Application of Neural Network and Wavelet Analysis in Monitoring Multiple Structural Damage
Chen Wenyuan, Lei Zhao, Guotang Bi · 2007
Using db5 wavelet to decompose first layer of structural response momentum of acceleration in six layers, selecting damage feature of damaged sensitivity based on response single of acceleration, thus recognizing damage time of structure and realizing the surveillance on time of structural damage. Through wavelet package decomposition of acceleration response signal of obtained structure, it can obtain eigen vector of energy in al frequency ranges and construct characteristic parameter input of BP neutral network and then construct one group of simple network output related to the input. The output can not only indicate location of structural damage and degree of structural damage, but also greatly reduce time spent on training and emulation of network. The adopted BP neural network structure is 8-7-7-6, the activation function in hidden layer is Sigmoid and the activation function in output layer is linear function training 405 group sample, realizing identification on damage location and damage degree. Analogue calculation has indicated that using wavelet analysis and BP neural network together can accurately diagnose the time, location and degree of structure, being of some feasibility.