Neural network-based inverse analysis for defect identification with laserultrasonics

This paper describes an application of the neural network-based inverse ana lysis method to the identification of a surface defect hidden in a solid, u sing laser ultrasonics. The inverse analysis method consists of three subpr ocesses. First, sample data of identification parameters versus dynamic res ponses of displacements at several monitoring points on the surface are cal culated using the dynamic finite-element method. Second, the back-propagati on neural network is trained using the sample data. Finally, the well-train ed network is utilized for defect identification. Fundamental performance o f the method is examined quantitatively and in detail, through both numeric al simulations and laser ultrasonics experiments. Locations and depths of v ertical defects are successfully estimated within 12.5% and 4.1% errors rel ative to the specimen thickness, respectively.