Accurate assessment of corrosion defects on pipelines is crucial for fitness-for-service determination and failure pressure prediction. Since the 1960s, pipeline defect assessment has evolved into a three-level framework. Artificial intelligence (AI) provides data-driven tools to enhance pipeline integrity management. By leveraging machine learning methods, AI can extract complex patterns from multi-source data, enabling rapid analysis beyond traditional methods. In corrosion defect assessment, AI has been widely used to determine failure pressure, defect growth rate, and failure probability. This talk introduces the latest progress made in this area. A stacking ensemble model is developed by combined seven base learners with multiple meta-learners, predicting burst pressure of pipelines containing corrosion defect, achieving a higher accuracy than individual models, and an improved generalization. With frequent scarcity of defect data for a given pipeline, a CopulaGAN-based data augmentation approach is developed to predict the failure pressure. The pipe wall thickness, defect depth, and outer pipe diameter are identified as key factors affecting pipeline failure. Furthermore, a TabPFN-SHAPIQ hybrid model is used to predict the failure pressure of hydrogen-blending natural gas pipelines with cracks. The model provides interpretable insights into feature interactions, such as the synergy between crack length and the hydrogen content.
