基于神经网络方法的框架结构损伤检测的试验研究

EXPERIMENTAL STUDY ON DAMAGE DETECTION OF FRAME STRUCTURED BASED ON ARTIFICIAL NEURAL NETWORK ALGORITHM

首先建立了三层试验框架结构的有限元模型,利用未损伤状态的动态测量数据,采用神经网络方法分步对原结构的有限元模型进行了修正。然后,依据修正的有限元模型,运用神经网络方法对各种实际损伤状况进行了损伤诊断。比较了仅以三阶频率作为神经网络输入向量和三阶频率及一阶振型组合作为网络输入向量对网络训练和损伤检测结果的影响。研究表明,神经网络的输入数据越充分,网络训练的收敛速度越快;利用三阶固有频率能够对该模型结构的各种损伤进行诊断,获得满意层间刚度识别的结果。

The finite element model of an undamaged 3-story experimental frame model is firstly updated based on the measured dynamic information by using neural network method.Then,on the basis of the updated finite element(model),damage detection is conducted for various damage cases.Neural network's training and damage detection(influenced) by input vectors which include the group of the first three frequencies or the group of the first three(frequencies) as well as the first modal shape,are studied.It is demonstrated that structural damage can be successfully(identified) by using three orders of natural frequencies,and the richer neural network's input information is,the faster(neural network's) convergence speed is.