Leaks and ruptures are the most important possible risks for operational oil and gas pipelines. Due to their hazardous effects on the environment, much research has been conducted to prevent and detect possible ruptures on a pipeline to protect people and the environment by enhancing safe operation. Any improvement in leak detection technologies to increase the accuracy and sensitivity while eliminating false leak and rupture alerts will protect the environment and assure hazard-free operation.
Data mining algorithms are widely used in many industries, including the energy industry. They have already been implemented as computational leak detection methodologies. To increase confidence and improve accuracy and sensitivity, different algorithms may be introduced to detect ruptures.
In our study, a 36" crude oil pipeline with two pump stations was configured in a pipeline simulator. The pipeline parameters of flow, pressure, and temperature were computed for several leak and rupture cases, and data science algorithms such as Logistic Regression, Neural Network, and Multivariate Adaptive Regression Splines were used as classifiers to detect the leaks and ruptures.
Multivariate Adaptive Regression Splines (MARS) is an important statistical learning tool for both classification and regression. MARS is nonparametric, adaptive, and effective in high dimensional problems with a proven record for fitting nonlinear multivariate functions. The contribution from the basis functions together with interaction effects between the predictors are used to determine the response variable: MARS produces a resultant model as an explicit formula.
MARS proves itself as a comparative classifier to the already known logistic regression and neural network methods as a new leak and rupture detection computation data science technique for pipeline operators.
