Registration of Multimodal Pipeline In-Line Inspection Data Using Generative Adversarial Network-Based Translation

Detecting and accurately locating defects through nondestructive inspection (NDI) is critical for maintaining pipeline integrity. Employing multiple NDI techniques on the same pipeline enhances inspection reliability, especially when dealing with complex, degraded structures. However, misalignments between data acquired from different sensing modalities present a significant challenge. Achieving effective data registration that ensures spatial alignment of these multimodal measurements is difficult due to the diverse underlying physical principles and the nonlinear, multimodal nature of the resulting signals. This study proposes an automated registration method for magnetic flux leakage (MFL) and ultrasonic testing (UT) inspection data. First, a generative adversarial network (GAN) is employed to translate UT data into the MFL domain, harmonizing the multimodal data into a consistent physical space and improving cross-modality similarity. Next, the registration parameters are optimized by applying a difference metric that minimizes the gradient of the difference map between the translated and measured MFL images. Experimental results demonstrate that this GAN-based translation, combined with the gradient-guided difference metric, improves registration accuracy. The resulting alignment of the MFL and UT data enables reliable subsequent data integration.