Framework for Generating Pipeline Leak Datasets using PIPESIM

Framework for Generating Pipeline Leak Datasets using PIPESIM

No leak datasets

Pipeline leaks

Pipeline Networks

PIPESIM

RAND

Access to pipeline leak datasets is one of the most critical challenges limiting the development of robust pipeline leak detection algorithms using Machine learning. While machine learning presents an opportunity for the development of robust pipeline leak detection algorithms, its application and algorithms require large volumes of pipeline leak-related datasets for their development. This work has been able to develop a framework for the generation of pipeline leak datasets. The work utilizes the PIPESIM software for the generation of the pipeline pressure profile for the entire pipeline length. The RAND program in python is used to convert the pressure value for any selected location on the pipeline to time series pressure data. The function enables the generation of a continuous stream of data capturing the variations introduced by the inlet pressure of the pipeline. This dataset is linked with the pipeline inlet pressure such that any variation in the Inlet pressure is reflected in the pressure value at the selected sensor location on the pipeline. This results in the generation of No leak datasets from the pipeline. The leak detection datasets are generated by introducing a valve and a sink to introduce a leak on the pipeline. The pipeline pressure profile is used to generate the leak dataset using the PIPESIM and the RAND function in python. The No leak dataset enables the development of one class classification algorithms which can be used to develop leak detection algorithms.

100113

Idachaba, Francis

Tomomewo, Olusegun

Journal of Pipeline Science and Engineering

2023

journalArticle

2667-1433

10.1016/j.jpse.2023.100113

https://www.sciencedirect.com/science/article/pii/S2667143323000057