Research on compensation and segmentation methods for ultrasonic phased array testing of aluminium alloy components

Depth-dependent echo attenuation, speckle noise, and weak boundary continuity complicate three-dimensional defect segmentation in ultrasonic phased-array data of aluminium alloy components. This study proposes a depth-attenuation-adaptive multi-scale curvature enhancement and segmentation method. An equivalent attenuation coefficient is estimated from the depth-wise mean amplitude profile, and bounded exponential gain compensation enhances deep weak echoes while limiting noise amplification. Multi-scale curvature responses are obtained by local quadratic surface fitting and adaptively fused according to the local signal-to-noise ratio. Finally, a depth-aware and seed-referenced velocity-field region-growing strategy is used, in which curvature and gradient variations regulate local propagation speed and high-confidence seeds determine the adaptive growth criterion. Experiments on a 6061-T6 aluminium alloy block containing eight flat-bottom holes at different depths showed that the method achieved mask-conditioned gains of 9.82 dB in ΔSNR, 10.18 in ΔCNR, and 7.74 in ΔLC, while maintaining accurate geometric localisation. A supplementary experiment on a V2068 aluminium alloy valve block containing a 1-mm machined reflector demonstrated the method’s applicability under a more complex structural background. These results support the effectiveness of the workflow for the specimens investigated, while validation on natural defects and independent datasets remains necessary.