Estimation of micro-crack lengths using eddy current C-scan images and neural-wavelet transform

The work reported in this paper is concerned with the development of neural network-based methods for estimating the size of cracks in the range of mum occurring around a hole on or beneath the surface of metal plate using eddy-current based C-scan images. The developed software includes wavelet transform-based feature extraction from C-scan images with known crack length and computing the energy associated with wavelet coefficient feature data. The feature data were then nonlinearly modeled using feed-forward neural network for the estimation of crack lengths. The results obtained are very promising and the method can be applied for online monitoring and estimation of micro crack sizes. The smallest crack size estimated was 200 mum within 10% estimation error. Due to limitation of resolution of the sensors, all measurements were performed in the millimeter range and images were resized again to simulate crack sizes in the micro-meter scale.