NEURAL-NETWORK CLASSIFICATION OF FLAWS DETECTED BY ULTRASONIC MEANS

A methodology for the automatic recognition of weld defects, detected by a P-scan ultrasonic system, has been developed within two stages in the present work, In the first stage, a selection of the shape parame ters defining the pulse-echo envelope reflected from a generic flaw, a nd defined in the time domain, is performed by Fischer linear discrimi nant analysis. In the second stage the classification is carried out b y a three-layered neural network trained with the backpropagation rule , where the input values are the parameters selected by the Fischer an alysis, With regard to the neural network learning process, 135 real w eld defects have been considered, The defects, distributed among the c lasses of cracks, slags of inclusion and porosity, had been previously characterized by X-ray inspection. The results obtained confirm the e ffectiveness of the approach in preserving the discriminant informatio n needed for characterization by an iterative use of Fischer analysis, and in increasing the generalization properties of the layered networ k by an interpretation of the knowledge embedded in the generated conn ections and weights. The required computation time allows in-process a pplication. Copyright (C) 1996 Elsevier Science Ltd.