摘要
研究将实测结构频率响应函数作为反向传递人工神经网络的输入数据,用来进行结构健康检测。一般情况下,把频率响应函数应用到人工神经网络的困难在于需要压缩频率响应函数的庞大数据,因为直接使用全部的频率响应函数数据使得神经网络具有大量的输入节点,从而导致网络训练收敛和计算效率方面的困难。仅仅使用部分频率响应数据,或不合适的频率窗数据点选择会引起重要信息的损失。为解决上述困难,用FORTRAN语言编写了一个简化的BP神经网络程序,把某结构的频率响应函数作为网络的输入参量。每个实测频率响应函数具有8192个数据点,神经网络采用8192-8-4结构,网络训练获得了较快的收敛速度。经过训练的网络成功识别了某结构的四种不同状态,识别误差小于10%。
This paper deals with the structural health detection using measured frequency response functions （FRFs） as input data to a back propagation（BP） artificial neural networks（ANNs）. In general, the difficulty of u- sing FRF data with ANNs is how to reduce the huge size of FRF data, because direct use of full - size FRF data will lead to the neural network having a large number of input nodes, which cause the problem of training convergence and computational efficiency. Only using partial FRF data, improper selection of data points from frequency windows may result in loss of important information. In order to circumvent the above difficulty, a simplified neural network is applied and a Back - Propagation network program in FORTRAN language is presented, using a structures frequency response function as the network input parameters. Every measured frequency response function has 8192 data points. The neural network has the 8192 -8 -4 structure, and the network training has a higher convergence speed. The network has successfully identified a structures＇s four kinds of mechanical states with errors under 10%.
出处
《计算机仿真》
CSCD
2007年第3期86-89,共4页
Computer Simulation
关键词
频响函数
神经网络
故障识别
算法
Frequency response function（FRF）
Neural network
Fault diagnosis
Algorithm
