At present, pipeline transportation is a very common mode in the long-distance transportation of liquid materials such as oil. Compared with others transportation ways, pipeline transportation is not only high-throughput and continuous, it is also reliable and of low energy consumption. Although the probability of pipeline leakage is low, the result of leakage is catastrophic, including economic losses, personal safety, and environmental pollution. Once the liquid pipeline leakage occurs, risk assessment and emergency repair measures are urgent. The key to these is to determine the leakage parameters of the pipeline. The determination of the leakage location is conductive to timely maintenance. The leakage coefficient and time affect the leakage volume and leakage range. The current leakage detection methods are quite mature and diverse. However, many popular methods are based on physical mechanism and hydraulic calculations with assumptions (unchanged wave propagation speed, constant pipeline friction coefficient, etc) resulting in poor accuracy and high false alarm rate. Therefore, from the perspective of data driven, the relationship among leakage parameters and the operational flow rate and pressure is mined in this paper. First, due to the limited leakage accidents for one pipeline, a large amount of leakage data is generated through experimental simulation. For every second, the pressure and flow rate in upstream as well as in downstream should be recorded, making it hard for common deep learning algorithms to cope with such a high-dimensional complex dataset. To overcome the dimensionality problem, conditional generative adversarial network (CGAN) is then introduced to treat the dimensional data as labels when the leakage parameters model is trained. It consists of two key components, generative network and discriminative network, which are two powerful neural networks. These two networks explore the distribution of the leakage parameters through continuous adversarial training. After training the leakage data, the leakage parameters can be estimated based on the detected data (upstream and downstream pressure and flow rate) when the leakage occurs. Finally, four cases of pipeline leakage are tested to demonstrate our superiority over two traditional algorithms, i.e. neural network and support vector machine. The computational results show that the errors of four cases are all less than 10% by our approach while more than 12% by the comparison algorithms. Thus, this method has a potential prospect in the estimation of leakage parameters, and can effectively guide site management when the leakage occurs.


A Method of Leakage Parameters Estimation for Liquid Pipelines Based on Conditional Generative Adversarial Network

Division / Sponsors: 
Pipeline Systems Division

Paper Proceedings: https://asmedigitalcollection.asme.org/IPC/IPC2020/volume/84447

IPC2020 - Virtual Conference

Supporting the need among researchers worldwide for additional online and multimedia content, ASME now offers the ASME Conference Video Collection. This platform will collect and provide long-term access to conference presentation videos, encompassing all technology areas within ASME's technical conference and events program. About 30,000-40,000 engineers across all mechanical engineering and related disciplines participate in ASME conferences each year. This includes authors of approximately 6,000 to 8,000 technical papers published by ASME, the majority of whom also submit video presentations for our virtual conferences. The video collection will be available as an add-on to existing product packages or as a separate subscription.