Session: Artificial Intelligence and Machine Learning Models

Paper Number: 158461

158461 - Development and Verification of a Pipeline System Surge Screening Tool Using Deep Learning Technology 

Abstract: 

Surge events, such as fluid hammer and vapor formation or collapse, pose significant risks to the safety and economic performance of pipeline systems. These risks are amplified in complex piping and pumping networks, where intricate configurations and operating conditions create challenges in identifying potential surge issues. Surge concerns often arise during the early stages of system design, but comprehensive transient and dynamic modeling of all piping sections and subsystems at this stage is impractical due to the absence of detailed design information. Such modeling efforts are also resource-intensive, requiring significant computational and financial investments.  

A key question frequently asked during the conceptual design phase is: which areas of the system are likely to experience surge-related problems, and how can these areas be identified for further detailed analysis and design? Current industry practices often rely on simplified tools, such as the Joukowsky equation, for surge screening. While these methods provide basic estimates of pressure surges, they fail to account for the complexities of real-world systems, leading to unreliable predictions and potentially inadequate surge mitigation strategies.  

This study presents the development of a deep learning-based surge screening tool designed to overcome these limitations. The model leverages data from detailed transient hydraulic simulations to predict the magnitude and location of surge pressures caused by fluid hammer events, as well as vapor formation and collapse. A comprehensive range of parameters is considered in the training process, including pipeline length, diameter, elevation changes, pump total dynamic head (TDH) and power, valve types and closure times, flow rates, pressure control mechanisms, and network configurations. This enables the model to capture the interplay between various factors that influence surge behavior, providing a holistic approach to surge screening.  

To validate the model's accuracy and robustness, randomly generated datasets representing diverse system configurations and operating conditions were used for testing. The results demonstrate the model's ability to identify high-risk areas with greater precision than traditional methods, making it a valuable tool for engineers during the early design stages. Furthermore, the tool offers significant cost and time savings by reducing the need for exhaustive detailed modeling across the entire system.  

This presentation will cover the development, training, and validation of the deep learning-based surge screening tool, emphasizing its practical applications in real-world projects. Case studies will illustrate how the tool enhances early-stage decision-making by identifying critical areas requiring transient hydraulic modeling and informing surge mitigation strategies. The proposed approach represents a significant step forward in leveraging artificial intelligence to improve pipeline design and safety.  

Presenting Author: Phuc Do Fluor Enterprises Inc

Presenting Author Biography: Phuc Do has close to 1 year of experience in design engineering of midstream oil and gas pipelines as an associate design engineer. He is an alum of the University of Southern California, where he studied chemical engineering. He started his professional career at Fluor Corporation in the pipeline discipline. With his competency in data analyst, he hopes to utilize automation to enhance processes and systems efficiency.

Authors: