Scholarlink

<p>Pipelines are vital for global fluid transportation, often exposed to harsh environmental conditions that make them prone to corrosion, a leading cause of pipeline failures. However, accurately predicting the time left till failure due to corrosion remains a challenging problem. Existing methods focus on detecting failures after they occur or rely on limited predictive capabilities, often overlooking the influence of environmental factors. This study addresses this gap by proposing a machine learning-based approach to predict the time left till pipeline failure, leveraging a large dataset from the Pipeline Hazardous Materials Safety Administration, enhanced with weather data (temperature and precipitation) from the National Centers for Environmental Information. Regression and classification models were developed, with the extremely randomized trees (extra trees) algorithm achieving an R-squared of 90.35% for regression and an f1 score of 85% for classification. The SHapley Additive exPlanations (SHAP) technique was applied to identify key predictors, revealing that weather conditions significantly affect the time left till failure. Key contributions include the integration of weather data to enhance predictive accuracy, a tailored data augmentation approach to improve model robustness, a comprehensive evaluation of machine learning models, and the use of SHAP for interpretability to better understand the drivers of pipeline failure timing.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Service Life Prediction in Pipelines Using Machine Learning Techniques

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月1日

查看全文
  • Michael Tawk,
  • Samir Mustapha,
  • Fouad Trad,
  • George Saad

摘要

Pipelines are vital for global fluid transportation, often exposed to harsh environmental conditions that make them prone to corrosion, a leading cause of pipeline failures. However, accurately predicting the time left till failure due to corrosion remains a challenging problem. Existing methods focus on detecting failures after they occur or rely on limited predictive capabilities, often overlooking the influence of environmental factors. This study addresses this gap by proposing a machine learning-based approach to predict the time left till pipeline failure, leveraging a large dataset from the Pipeline Hazardous Materials Safety Administration, enhanced with weather data (temperature and precipitation) from the National Centers for Environmental Information. Regression and classification models were developed, with the extremely randomized trees (extra trees) algorithm achieving an R-squared of 90.35% for regression and an f1 score of 85% for classification. The SHapley Additive exPlanations (SHAP) technique was applied to identify key predictors, revealing that weather conditions significantly affect the time left till failure. Key contributions include the integration of weather data to enhance predictive accuracy, a tailored data augmentation approach to improve model robustness, a comprehensive evaluation of machine learning models, and the use of SHAP for interpretability to better understand the drivers of pipeline failure timing.