Magnetic flux leakage (MFL) inspection is widely used for corrosion and metal-loss detection in oil and gas pipelines, yet accurate defect characterization requires large, diverse, and physically reliable datasets. Field measurements are costly and limited, while large-scale finite element (FE) simulations are computationally expensive. To address this challenge, this study proposes a physics-guided conditional Wasserstein generative adversarial network with gradient penalty (PI-cWGAN-GP) for MFL signal augmentation. A validated FE model is first established to generate tri-axial MFL responses for pipeline defects with varying length, width, depth, and rotation angle, which are then used for generative learning. The main innovation of the proposed framework is the incorporation of MFL field gradient information into adversarial learning through a gradient-augmented discriminator. Instead of evaluating only the magnetic field intensity map, the critic jointly considers the original MFL signal and its local gradient features, enabling the model to capture defect-induced spatial leakage patterns, and geometry-dependent field variations. By embedding both defect geometry conditions and MFL gradient information into the generative process, the proposed model produces synthetic signals with improved statistical fidelity and physical consistency. The generated samples are evaluated through visual morphology comparison and quantitative metrics, including SSIM, FID, and KID. Their practical utility is further validated using a downstream dual-branch convolutional neural network for defect geometry regression and magnetic field component classification. Experimental results show that the proposed method achieves an average FID of 46.93, KID of 0.02, and SSIM of 0.98, while reducing defect-depth prediction MAE from 0.73 to 0.34 and RMSE from 1.00 to 0.52. The proposed framework supports the generation, conditioning, and evaluation of MFL signals, improving signal availability, quality, and measurement relevance for pipeline integrity assessment.
