Research on an inversion method for internal detection of dual-magnetic field stress in pipelines based on the MCL-Net model

Composite defects (crack-stress coupling) in pipelines pose a major risk to oil and gas transportation. Accurately inverting stress distribution at these sites is a critical challenge for pipeline integrity management and precise safety assessment. This study develops a stress inversion model for internally detecting pipeline composite defects, using a data-driven dual-mode excitation collaborative method. This approach enables multi-physics coupling decoupling and fuses weak magnetic signals with coupled signals through data-driven fusion. The model integrates an improved Multiscale Convolutional Neural Network – Long Short-Term Memory Network (MCL-Net) and an enhanced Snow Ablation Optimisation algorithm (ISAO). Parallel multiscale convolutional kernels extract local details and long-range trends from magnetic signals, while a bidirectional LSTM – multi-head attention module captures temporal dependencies and key time-window features. Simulation results show the method achieves an R² of 0.9255, with RMSE and MAE of 0.1371 and 0.0851, respectively. In engineering experiments, it attains an R² of 0.9022, representing a 10.89% accuracy improvement over the BKA-Awarenet algorithm. The method significantly enhances stress inversion accuracy at composite defect sites, establishing a new theoretical foundation and providing effective technical support for precise pipeline safety assessment.