复合材料结构损伤的小波神经网络辨识研究

Research on Using Wavelet Neural Network to Recognize Damage in Composite Materials

将小波神经网络应用于结构健康监测,研究实现复合材料结构常见损伤的高精度辨识。剖析了小波神经网络的收敛算法,并使用了惯性系数以抑制振荡并提出了一种自适应调整学习率的算法以加快收敛。组建结构健康监测实验系统,进行数据处理和特征提取以获得不同的结构损伤模式。提出了小波神经网络初始权值的设置方法,据此删除了小波神经网络的冗余节点。将该小波神经网络应用在实验获得的各种结构损伤模式的辨识上,验证了它的高精度和快速收敛,并成功实现了复合材料结构损伤状态的辨识仿真。

This paper applied WNN to structural health monitoring and recognized five structural statuses in composites. Using the back-propagation training algorithm, the WNN similar to RBF network performed well and for faster convergence, which set the inertia coefficient, eliminated the redundancy of the hidden layer, and can adjust learning rate self-adaptively as training progresses. This paper also presented a method to set the initialized parameters of wavelet and network before training, which was rather important to train WNN. For obtaining the different damage patterns to train and simulate the wavelet neural network, a structural health monitoring system was developed. As a primary research result on the application of wavelet neural network to structural health monitoring, a wavelet neural network architecture, converging fast and approaching for high precision, is obtained and successfully recognizes the damage patterns defined by the experimental system.