Forecasting pipeline safety and remaining life with machine learning methods and SHAP interaction values

In-line inspection (ILI) is a common and useful way to provide comprehensive threat assessments and to detect defect dimensions. However, the ILI is very expensive and most of its costs are unnecessary since very few pipelines contain defects. Many previous studies have used machine learning (ML) methods to predict the safety of pipelines based on the features extracted from ILI signal results, which is very laborious and difficult to apply in industry. In this paper, the ILI results were predicted using multiple ML models through three prediction cases with only pipeline attributes and environmental features as input variables. In the first case, six ML methods were compared to predict whether there are defects in a pipeline section. The Catboost (CAT) method showed the best performance in all evaluation metrics with the highest certain prediction ratio (94%). The second case was using three ML methods to predict a defect depth, defect length, and defect width. The prediction accuracy of the defect depth was much higher than that of the defect length and defect width. A CAT model was the optimal model in this case due to its best performance in all the predictions. In the third prediction case, three ML methods were used to predict a defect length growth rate and defect depth growth rate based on different ILI results from different years. CAT was also the best method in this case with the highest accuracy. The predicted defect length growth rate and defect depth growth rate were used to calculate the remaining lifetime of pipelines based on the maximum allowable defect depth and failure pressure. The remaining lifetime of pipelines with different thickness, land use, and soil types was analyzed with changes in the specified minimum yield strength of a pipeline, a pipeline year, and maximum operating pressure. The accurate predictions in the three cases and the correlation analysis between input features and outputs helped company save a lot of costs and provided valuable information and suggestions in the further progress.