Research on Pipeline Defect Shape Classification and Recognition Based on EBWO-BP-Adaboost Model

The precise classification and identification of pipeline defect geometries constitutes a critical technological requirement for ensuring structural integrity in industrial pipeline operations. This study addresses two fundamental limitations in current approaches: the suboptimal accuracy and propensity for local optima convergence exhibited by conventional backpropagation (BP) neural networks in defect classification tasks, and the premature convergence tendency of the standard Beluga Whale Optimization (BWO) algorithm. To overcome these challenges, we present an enhanced AdaBoost ensemble prediction framework that incorporates a BP neural network optimized through our modified Beluga Whale Optimization (EBWO) algorithm. The experimental methodology involved comprehensive validation through magnetic flux leakage (MFL) detection simulations coupled with systematic defect feature extraction. The findings demonstrate that the introduced EBWO-BP-AdaBoost model attains a classification accuracy of 97.69%., which represents a significant improvement over traditional BP neural networks and standard BWO-BP models. Furthermore, the model exhibits exceptional robustness in handling complex multi-defect classification scenarios. Practical validation using operational pipeline inspection data confirms the model's effectiveness, establishing it as a novel and reliable approach for intelligent defect detection in challenging industrial environments.