Deep Learning-Enhanced Total Focusing Method for Guided Wave Damage Detection

Accurate defect localization in large metallic plates using ultrasonic nondestructive testing faces challenges from Lamb wave dispersive characteristics and Rayleigh criterion limitations for multiple defects. This paper proposes a super-resolution reconstruction method combining phase coherence factor (PCF) weighted Total Focusing Method (TFM) imaging with an improved deep learning network. The PCF weighting strategy effectively suppresses sidelobe interference and random noise, improving TFM image signal-to-noise ratio and contrast. A two-stage deep network architecture integrates lightweight EfficientNet-B0 defect detection with U-Net super-resolution reconstruction featuring enhanced mechanisms. For single through-hole defects, the method improves defect center spacing accuracy by 5.1% and area calculation accuracy by 92.5% compared to conventional TFM. The approach achieves super-resolution imaging even when defect spacing falls below the Rayleigh criterion resolution threshold.