Evolution of corrosion prediction models for oil and gas pipelines: From empirical-driven to data-driven

Oil and gas pipelines are under great threat of corrosion due to the harsh service environment. It is critical to predict corrosion for the safe service of pipelines. Classical empirical-driven and mechanism-driven models have been successfully applied to predict the corrosion of oil and gas pipelines, while their complex applicability conditions and calculations become limitations. Datadriven models based machine learning (ML) are becoming the new trend owing to their efficiency and accuracy. This work systematically reviews these models including their evolution, characteristics, limitations, and performance, and highlights the application of data-driven models. In addition, a ML method database of corrosion prediction for oil and gas pipelines was created by summarizing the pre-processing, input and output parameters and performance metrics of ML models, which provide guidance for rational selection of models. Finally, conclusions and recommendations are presented and provide a broad outlook for future research paths.

Automated summary

Oil and gas pipelines face significant corrosion threats in harsh environments, making corrosion prediction crucial for their safe operation. Traditional empirical and mechanism-driven models have limitations due to their complex applicability conditions and calculations. Data-driven models based on machine learning are gaining popularity due to their efficiency and accuracy.