ULTRASONIC CHARACTERIZATION OF DEFECTS IN STEEL USING MULTI-SAFT IMAGING AND NEURAL NETWORKS

Due to the recent improvements in nondestructive inspection (NDI) techniques, even small or weak inhomogeneities can be detected and discrimination between critical and non-critical defects becomes more crucial. In practice, welded regions of steel components are of special interest. Characterizing a detected inhomogeneity may be done by using cross-sectional images, which are reconstructed from ultrasonic B-scan data. However, in this application high-resolution images are very hard to obtain, because the geometry of most objects severely limits the amount of ultrasonic information which can be collected. In this paper it is demonstrated that optimized data acquisition may yield ultrasonic B-scan data containing responses of multiple wave paths, which can be used to provide multiple focused images. Furthermore it is shown that the application of neural networks may give additional information on the shape of the defect, provided that correct pre-processing of the ultrasonic B-scan data is applied.