基于广义回归神经网络的交流电磁场检测裂纹量化研究

考虑到目前交流电磁场检测中裂纹量化精度和智能化水平的不足,将广义回归神经网络(GRNN)引入到交流电磁场检测技术中来,在有限元仿真试验基础上,选择了作为输入元素的交流电磁场信号特征向量,构建了一种适合于交流电磁场检测裂纹量化分析的GRNN模型,并利用归一化处理后的一些离散数据作为网络的训练和检测样本,使网络完成对整个裂纹交流电磁场范围内的主要信息的存储,从中发现输出和输入之间的内在关系,完成对未知点的预测。结果表明,与传统的线性插值方法以及BP网络相比,该方法建模简单,预测精度高,对原始数据的分布和边界条件无特别要求,推广性能强,人为调节参数少,收敛速度快,更为智能化,尤其在获得已知样本稀少的情况下仍能表现出极强的适应性,从而保证了模型的精度和推广性能,为交流电磁场检测裂纹量化提供了一种智能高效的方法。

Crack sizing for alternating current field measurement based on GRNN

Considering the deficiency of the precision and intelligence of the crack sizing method used in the alternating current field measurement(ACFM),the generalized regression neural network(GRNN) was introduced.On the basis of the finite element simulating experiment,characteristic vectors were picked up as input-elements of GRNN.Using normalized discrete data as training and testing samples,the main information was saved in the GRNN model,and the intrinsic relationship between input and output was found out.Finally,the GRNN model was used to forecast unknown points.The results indicate that compared with the traditional linear interpolation and BP neural network,the GRNN is more precise,intelligent and generalized,and it has the strong adaptability with few training and testing samples,which guarantees the precision and generalization of the model for crack-sizing forecasting in ACFM.