Research on intelligent defect quantification models for pipeline data-driven applications

• Develops an improved PP-YOLOE-SOD model with multi-modal feature enhancement for robust defect localization in pipeline magnetic flux leakage data, achieving 91.4% [email protected] and maintaining 84.1% accuracy under strong noise (3 dB). • Establishes a defect size quantification model using Elastic Net regression, leveraging L1/L2 regularization and K-fold cross-validation to achieve low average errors (length: 2.11mm, width: 4.75mm, depth: 1.02mm) compliant with industry standards. • Proposes an integrated Localization-Feature Extraction-Quantification framework that overcomes manual interpretation bottlenecks and black-box limitations, enabling efficient, high-precision automated analysis of massive MFL data for pipeline integrity assessment. In service, pipelines are inevitably subject to structural defects, such as corrosion, pitting, and cracks, due to factors like high pressure and external loads. These defects threaten normal pipeline operation and can cause serious safety hazards and economic losses.Leveraging the principles of magnetic flux leakage detection, the distribution and severity of surface defects along the pipeline are meticulously analyzed and interpreted. This analytical approach is pivotal for ensuring the pipeline's operational safety, as it allows for the identification of potential issues such as corrosion and deformation, which can compromise the integrity of the pipeline.The large scale of actual detection data and the low efficiency of manual interpretation, coupled with the fact that,the number of defect samples is limited by equipment and cannot be fully acquired.. How to achieve efficient quantitative analysis of defects based on limited samples and engineering data has become a core challenge in the field of pipeline detection.This paper proposes an intelligent defect quantification model that combines an improved PP-YOLOE-Small Object Detection (PP-YOLOE-SOD) target detection framework tailored to the characteristics of magnetic flux leakage data with Elastic Net Regression (ENR). Through an in-depth analysis of the structure and characteristics of magnetic flux leakage data, the collected detection data is enhanced, the model framework is improved, and the improved PP-YOLOE-SOD target detection model is applied to MFL data analysis.Based on the results of target detection scanning, this method accurately locates defect areas and extracts key feature parameters. By combining ENR, it constructs a robust mapping relationship between defect size and detection signals.According to experimental results, the method demonstrates a defect identification accuracy rate of 91.4% in engineering pipeline detection, aligning with industry standards for size quantification errors.This significantly improves defect interpretation efficiency and quantification accuracy under large-scale data conditions, satisfying industry standard requirements.This not only solves the engineering problem in quantifying pipeline defects, but also offers robust technical support and theoretical guidance for the assessment of in-service pipelines.