Defect characterization by an adaptive learning classifier in ultrasonic inspection of ferritic materials

The purpose of the work presented here is to determine the efficiency and reliability of the discrimination method ''Adapatative Learning Network''. This process is used to identify the nature of defects detected during Ultrasonic Testing (UT), by analysing some of the echo parameters calculated in temporal and frequency domains. The survey which dealt with several hundreds of echoes, revealed that the machined artificial reflectors, well defined geometrically, were identified with a probability exceeding 95 %. The fatigue cracks, producing less characteristic echoes, are identified with a slightly lower probability (90 % - 95 %)