PSI - Issue 71

Rakesh Kumar Sahu et al. / Procedia Structural Integrity 71 (2025) 203–209

208

4.1. Comparison of tone burst response of varying width (single and multi) From Fig. 7 it can be observed that the prediction done by the trained neural network algorithm for 12mm damage is accurately predicting 12 mm damage and for 10mm damage the prediction is 11.45 mm. From Fig. 8 it can be observed neural network algorithm for multiple damages predicts the width of damage accurately. In this case, the model predicted the width of two damages as 4.9987 mm while the actual value is 5 mm of two damages.

Fig. 5. Performance plot of two-layer feed forward neural network for single notch

Fig. 6. Performance plot of two-layer feed forward neural network for two notches

4.2 Comparison of tone burst response of varying width (single and multi) From Fig. 9 it can be observed that the predicted location of 12 mm damage is at 707 mm while the actual location of the damage is 710 mm and for 10 mm damage at 685 mm while the actual value was 675 mm. From Fig. 10, it can be observed that the prediction for damage location of multiple damages is done accurately by the trained neural network. In this case, the damage location of the first 5 mm damage predicted was 692.46 mm while the actual location was 690 mm from the reference, the damage location of the second 5mm damage predicted was 808.66mm while the actual location was 810 mm from the reference.

Fig. 7. Model predicting damage width at 200 kHz for single damage

Fig. 8. Model predicting multiple damage width at 200 kHz for two damage

Fig. 10. Model predicting multiple damage location at 200 kHz for two damage

Fig. 9. Model predicting damage location at 200 kHz for single damage