基于神经网络正向模型和遗传算法的涡流检测自然裂纹形状重构

提出了一种基于神经网络正向模型与遗传优化算法从疲劳裂纹涡流检测(eddy current testing, ECT)信号重构裂纹形状的方法.人工制作了疲劳裂纹试样,利用一种小波分析方法对采集的疲劳裂纹ECT信号进行了去噪预处理并提取了信号特征.随后通过破坏性检测方法获得了裂纹的真实形状.在建立疲劳裂纹参数化模型基础上,利用经过处理的裂纹ECT信号和裂纹形状参数样本库对径向基函数(Radial Basis Function, RBF)神经网络进行训练.遗传算法首先创建大量表示裂纹形状参数个体的初始种群,输入经过训练的神经网络,得到对应的ECT预测信号,然后运用遗传策略进行迭代反演优化,搜索裂纹形状最优解.重构结果表明该方法具有快速、精确的优点.

Reconstruction of Natural Crack Shapes From the ECT Signals by Using an Artificial Neural Network Based Forward Model and Genetic Algorithm

This paper presents an inversion algorithm for the reconstruction the profile of fatigue crack from eddy current testing (ECT) signals.The scheme is based on forward model of artificial neural network and genetic algorithm.The fatigue crack samples are fabricated artificially and tested with a pancake coil ECT probe. A wavelet transform signal processing technique is used to reduce the noise and nondefect signals from the collected ECT signals,and then the useful characteristic is extracted.Then a destructive testing procedure is performed to obtain the true profiles of the cracks. Based on a fatigue crack parametric model, the Radial Basis Function (RBF) neural network is choosed as the forward model and trained with the preprocessed ECT signals and crack shape parameters.The genetic inversion algorithm generates the initial population of crack shape parameter and then input them into the RBF network, the predicted ECT signals are the output of the network,based on the fitnss of each individual,the genetic strategies are used to search the global optimum solution. the reconstructed results show that this method has advantages of speedy and precise.