Digital methods for defect classification

Summary form only given, as follows. Objective computer-based analysis methods are described for reliably distinguishing significant defects such as cracks from more benign indications arising from volumetric flaws such as slag and porosity. The methods are based firstly on colour imaging techniques which are used to display 3-D ultrasonic data obtained from defects scanned using multiple inspection techniques. The resultant images give an invaluable qualitative guide to defect type and facilitate the derivation of numerical descriptors or `features´ based on well-defined physical principles. Statistical pattern recognition methods are then used for defect classification, by analysis of the feature values obtained from scanning many known examples of each defect type