Conditional Denoising Diffusion Model-Based Defect Reconstruction for Alternating Current Field Measurement

Defect detection is critical for safely operating equipment in harsh environments. Defect reconstruction, as a representative defect detection method, visualizes defects morphology to support equipment maintenance decisions. Existing defect reconstruction methods usually rely on simple mathematical calculations and physical principles, and the reconstructed defects deviate significantly from the real defect patterns. To enhance the accuracy of defect reconstruction, we propose a defect reconstruction algorithm based on the conditional denoising diffusion model (CDDM) for alternating current field measurement (ACFM). The diffusion model exhibits strong generative capabilities, particularly when provided with suitable conditions, enabling it to produce results that meet specific requirements according to the given conditions. Specifically, we first utilize the ACFM technique to obtain the induced magnetic field Bz signal around the crack and reconstruct the defect according to the dual gradient fusion algorithm (DGFA). We then utilize the defects reconstructed image by DGFA as a condition to guide the CDDM to reconstruct the defects morphology. The experimental results show that the CDDM can accurately reconstruct defects length and angle and that the model is capable of reconstructing cracks of smaller lengths (as small as 1 mm). Compared to existing defect reconstruction algorithms, our model achieves the best performance. The average errors in the reconstructed defect crack length and angle based on CDDM are 0.428 mm and 0.417°, respectively.