Abstract：Stress corrosion crack (SCC) has a complex micro-structure similar with bifurcated tree branches and its non-destructive quantitative evaluation is difficult. In this paper, reconstruction schemes based on artificial neural networks (ANN) and support vector machine (SVM) are proposed based on a conductive crack model. ECT signals from a pancake coil probe scanned just over the crack are taken as the source signals for the crack reconstruction. A lot of sample ECT signals are generated using a FEM-BEM hybrid code as data sets for ANN and SVM training. Special ways for crack parameterization and signal feature extraction are introduced to improve the reconstruction precision. The numerical results show that the SVM method has a better performance than the neural network approach in the reconstruction of SCC for noise polluted signals.
[1] Sergios Theodoridis, Konstantinos Koutroumba. 模式识别[M]. 李晶皎, 王爱侠, 张广渊, 译. 3版. 北京: 电子工业出版社, 2006. [2] Chen Z, Miya K, Kurokawa M. Rapid prediction of eddy current testing signals using A-phi method and database[J]. NDT&E International, 1999, 32: 29-36. [3] Popa R C, Miya K. Approximate inverse mapping in ECT based on aperture shifting and NN regression[J]. Journal of Nondestructive Evaluation, 1998, 17(4): 209-221. [4] Cortes C, Vapnik V. Support vector network[J]. Machine Learning, 1995, 20: 273-297. [5] Yusa N, Cheng W, Chen Z, et al. Generalized neural network approach to eddy current inversion to the analysis of real cracks[J]. NDT&E International, 2002, 35: 609-614. [6] Chen C L. A rapid supervised learning neural network for function interpolation and approximation[J]. IEEE Transactions. on Neural Networks, 1996(7): 1220- 1230.
2010, Vol. 25 
