基于人工神经网络的电涡流逆问题解

在涡流检测中,常利用最小方差法来进行缺陷重构,这种方法要求正问题的求解效率高。提出一种新的求解正问题方法,该方法利用人工神经网络来估计涡流检测信号,大大提高缺陷重构效率。通过对理想裂纹数值验证显示,该方法可以用数值计算方法得到学习样本对人工神经网络进行训练,并且人工神经网络在学习过程中对噪声不敏感,因而可以有效抑制噪声。

The Eddy Current Inversion Based on Artificial Neural Networks

The general eddy current flaw reconstruction strategy was the least-square error method, which required the forward solver having high speed. A novel technique was presented for forward problem to rapidly predict eddy current signals by utilizing artificial neural networks, which could enhance the efficiency of flaw reconstruction. After the numerical analysis of ideal cracks, it was found that the artificial neural network training could only use the data gained by some numerical calculation method, and the artificial neural network training was not sensitive to the noise, so this technique has the ability of restrain the noise.