Prediction and Modeling of Magnetic Flux Leakage Signals Based on Machine Learning

点击翻译按钮获取中文摘要

The global pipeline integrity management market size was $8.65 billion US in 2020 and is projected to grow to$11.26 billion in 2028. Magnetic flux leakage (MFL) is the most widely used non-destructive testing (NDT) technique to detect cracks and corrosions in pipeline. While most MFL prediction methods can be applied to simple profile shapes, they fail to detect defects with irregular geometries and complex regions of interacting signals. We propose developing new MFL prediction approaches based on deep learning. First, we will develop a spatial temporal deep learning prediction with sparse encoding for multi-channel MFL data. This model will then be extended by incorporating physics of cracks and corrosions formation as a physic informed neural network. Second, we will develop a style transfer between MFL and laser signals as a sensor transform for MFL mapping to generated laser signal to achieve ideal performance. The anticipated research results on detecting pattern detection will provide insight to inform the next generation of pipeline defect sizing and characterization algorithms. This will improve the accuracy of inline inspection results and lead to better, cheaper, and safer maintenance for pipelines, reducing the risk of serious incidents in Canada.