Application of Machine Learning in Computational Fluid Dynamics-based Design and Optimisation of Turboexpanders Used in Natural Gas Pressure Reduction Stations

A recently proposed resolution in the market of natural gas (NG) supply in urban areas considers the installation of energy-saving machinery such as turbo expanders in pressure reduction stations (PRSs) of NG distribution networks. The use of turboexpanders in these networks has successfully shown pronounced benefits over the traditional Joule-Thompson (J-T) valves, by effectively recovering the waste energy of the gas during the expansion process. On the negative side, however, turboexpanders are often exposed to off-design operations, ie mainly due to inefficient design causing an improper response to instantaneous variations of upstream pressure in a given NG distribution cycle, which may eventually compromise their advantages, if running uncontrolled. Towards addressing this very complexity, the present work is intended to introduce and examine a cost-effective, yet reliable, numerical framework that integrates machine learning (ML) with computational fluid dynamics (CFD) to improve re-design and optimisation of existing NG turboexpanders in PRS facilities, with the ultimate goal of upgrading traditional procedures frequently used for maintaining such machinery. Considering the high granularity of the proposed framework, it is anticipated that it could be conveniently extended as a robust supplemental tool for related industrial maintenance procedures dealing with NG turbomachinery and energy systems.