Based on Lifting Scheme Wavelet Packet Transform and Artificial Neural Network Wing-box Multi-damage Detection
|School||Zhejiang University of Technology|
|Course||Mechanical Design and Theory|
|Keywords||Structural health monitoring technology Lifting scheme wavelet transform Lifting wavelet packet transform Damage characteristic vectors Neural network|
Structural health monitoring system is now widely used in various industries such as aerospace, civil engineering, etc. With the development of modern industrialization, people have increasingly high demand for reliability to various structures, and damage detection in service of large complex structure is essential.This thesis proposes structural multi-damage detection method by using lifting scheme wavelet packet transform and artificial neural network. The main research showed as follows: Firstly, this thesis discusses the main principle, construction method of lifting scheme and the method of border processing. And this thesis also introduces the lifting scheme wavelet packet transform algorithm. Numerical example proves that lifting scheme wavelet packet can denote the energy distribution of the signal best, so it can be used to the signal feature extraction.Secondly, dynamic models of wing-box with intact and several kinds of damage levels are built to gain their dynamic responses through instantaneous analysis. The structural vibration response signal will change after damage. The characteristic vector that contains the information of damage (such as the wavelet packet node energy change index) obtained from the structural vibration response signal with lifting scheme wavelet packet analysis can be used to detect the structural damage. The numerical analysis results of wing-box structure show that: the sensors in different location have different capability in identifying the structure damage feature, and single sensor may not contain all the damage information.Finally, a damage detection method based on multi-sensor data fusion and artificial neural network is presented. The characteristic vector from different sensors are fused, and then used as input variable for back-propagation neural network to detect the damage. Numerical example of wing-box shows that the position and degree of damage are identified, which proves the proposed method is effective. Multi-sensor data fusion of structural damage identification methods based on data fusion and lifting scheme wavelet analysis can accurately identify the location and degree of structural damage, and the actual measurement error is more robust.