A Combined Computational Fluid Dynamics Machine Learning (CFD-ML) Framework for Numerical Design and Optimisation of Turbo Expanders Used in Natural Gas Liquefaction Units

With the destructive greenhouse gas effects on the rise, there has been a soaring demand for effective strategies that can aid in optimising energy consumption on a global scale. Out of the most attentive strategies in the present market of sustainability and net-zero emission is the application of turbo expanders in the liquefaction process of natural gas (NG). Although the use of turbo expanders has been promising in terms of harvesting energy in liquefied natural gas (LNG) production units compared to traditional equipment such as the Joule-Thomson valve, their complex operation – which is mainly associated with the complex behaviour of NG flows at low-temperature conditions – has challenged not only the related maintenance and condition monitoring procedures, but also the numerical tools regularly utilised for design and optimisation of such machinery. Aiming to address this deficiency, the current work brings together computational fluid dynamics (CFDs) and machine learning (ML) to develop a numerical framework able to effectively predict the performance of low-temperature gaseous NG turbo expanders through fast-response robust surrogate models, while providing an insight into the physical aspects of NG flow inside the turbo expander. The proposed framework can be ultimately extended as a supplemental tool for evaluating the efficiency of relevant industrial turbomachinery, helping to improve their design, implementation and maintenance procedures.