粒子群优化小波神经网络用于碰摩声发射源定位

Localization of Acoustic Emission Source Based on Particle Swarm Optimizer Wavelet Neural Network

针对时差定位法受不同模式波速度差异及波形传播畸变等因素影响的问题,将神经网络技术应用到声发射源定位中。在通常的BP小波神经网络中,BP算法实质上是一种基于梯度下降法的局部搜索算法,易使网络陷入局部最小值而使得搜索成功概率较低。作为改进,利用粒子群算法对小波神经网络中的参数进行优化,然后再利用基于粒子群优化的小波神经网络进行声发射源定位。仿真实验结果表明,选择合适的网络结构和输入参数,粒子群优化算法可以准确定位碰摩位置,且计算更加简单有效,具有良好的应用前景和进一步研究的价值。

Due to defects of time-difference of arrival localization which influenced by speed differences of various model waveforms and waveform distortion in transmitting process, a neural network technique was introduced to calculate localization of the acoustic emission （AE） source. However, in BP wavelet neural network （WNN）, the BP algorithm is a stochastic gradient algorithm virtually, the network may get into local minimum and the result of network training is dissatisfactory. As an improved way, the particle swarm optimizer （PSO） algorithm was proposed to train the parameters of the WNN, then WNN based on PSO was used to locate the AE source. The experiment results show that PSO algorithm could achieve accurate localization of AE source by selecting reasonable network structure and input parameters. Moreover, it has good convergence and low computation complexity