.png){kind=icon}
Return
Reliability constrained diffusion modeling for high-noise defect detection using ACFM
.png){kind=icon}
PRE
AI
DOI：10.1016/j.ndteint.2026.103870.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Alternating current field measurement (ACFM) has been widely employed for structural defect detection due to its non-contact nature and insensitivity to surface conditions. However, ACFM signals are often corrupted by high noise, which makes conventional approaches based on handcrafted features or fixed thresholds unable to sustain stable and reliable detection performance. To address this challenge, this paper proposes a reliability-aware defect detection method for high-noise ACFM. Specifically, a multi-scale autoencoder (MS-AE) is first employed to learn the structural prior of defect-free ACFM signals and recover the underlying normal structure from noisy inputs. The point-wise reconstruction errors are then used to construct a dynamic reliability mask that separates reliable regions from noise-dominated segments. A mask-guided denoising diffusion Transformer (MDDT) is then introduced to learn the normal samples distribution under mask constraints, thereby suppressing the influence of unreliable regions. Finally, a structure-aware defect determination mechanism is designed by integrating global reconstruction residual with local B z magnetic peak deviation, enabling robust and interpretable defect determination. Moreover, experiments on ACFM signals with multiple noise modes show that the proposed method outperforms existing approaches across multiple evaluation metrics, achieving higher detection accuracy and robustness.
Keywords:
Alternating current field measurement
Non-destructive testing
Defect detection
Generative models
Denoising diffusion model
Journal
N
IF:
4.5
Papers:
220
Citations:
0
Organization
Cited Papers
No cited papers available