基于PSO的结构损伤检测应用研究

建筑结构损伤前后固有频率的变化包含了结构损伤位置和程度的信息,在此理论基础上,构造了BP神经网络的输入参数.针对BP梯度下降算法导致的收敛速度慢和易陷入局部最小的缺点,引入粒子群演化(PSO)算法来优化神经网络各层间的连接权值.首先通过有限元法提取结构固有频率的变化,结合PSO对神经网络进行训练,然后分别对结构的损伤位置和损伤程度进行识别.计算分析结果表明,PSO的引入,相较于单纯的BP算法,该方法在结构损伤检测中取得更优的识别效果.

Research and Application on Damage Detection of Structure Based on PSO Algorithm

The location of structural damage and the information about the extent of damages are manifested in the change of natural frequency of the concrete structure damaged before and after.On the basis of this theory,a BP neural networks model based on particle swarm optimization(PSO) algorithm is constructed.Through the introduction of PSO algorithm,the connection weights of the model can be optimized,overcoming the shortcomings of the inefficiency in terms of convergence and the great possibility being stuck in a local minimum which results from the usage of FEM.First of all,the calculation of the changes of the structure natural frequency is integrated with the PSO algorithm for training the nerve network,which is followed by the location of the damage and identification of the extent of the structural damage.The computation indicates that the accuracy and convergence velocity processed by this method is much better than that only adopted by BP algorithm,which can enhance the recognition of the damages in the structure damage examination.