HAL will be down for maintenance from Friday, June 10 at 4pm through Monday, June 13 at 9am. More information
Skip to Main content Skip to Navigation

Application of Artificial Neural Networks to the study of hydraulic turbine draft tubes

Abstract : The draft tube of a hydraulic turbine are divergent shaped equipment located downstream the runner and responsible for efficiently converting the residual kinetic energy leaving the runner into pressure, thus increasing the effective head and performance of the machine. As the head losses inside a draft tube can represent an important portion of the total energy losses, especially in the case of low head bulb turbines, it is essential to accurately predict the flow behaviour inside the draft tube if more efficient and competitive hydraulic machines are to be designed. In this context, numerical simulations are less expensive than experiments and give access to a more detailed description of the flow inside the draft tube. Compared to conventional RANS simulations using two-equation linear eddy-viscosity models, LES is more capable of capturing the complicated flow dynamics inside the draft tube (i.e. highly turbulent flow, with a wide range of vortical motions, swirling and subjected to an adverse pressure gradient). However, as the downstream flow behaviour is highly influenced by the inlet conditions and since comprehensive experimental measurements at the inlet of the draft tube are expensive, rarely available and often limited to mean flow quantities measured at a single radial traverse, the real challenge for performing these simulations consists in determining proper mean and fluctuating inlet boundary conditions. Thus, a new approach based on data-driven techniques, such as machine learning, is proposed. Its goal is to use any known information about the downstream flow along with a previously generated database to reconstruct the unknown or incomplete inlet boundary conditions for a numerical simulation. Thanks to an artificial upstream extension added to the original domain, the proposed approach gives more space and time for the simple synthetic fluctuations being injected in LES to develop before reaching the important portion of the computational domain. First, the simple yet challenging case of the swirling flow inside the ERCOFTAC conical diffuser is investigated. Compared to the experimental data and previous numerical works that used ad hoc inlet boundary conditions, the obtained results are very good. Then, in the case of the more complex flow inside a bulb turbine draft tube, the proposed approach yields better results in both RANS and LES in comparison to the initial simulations using basic inlet boundary conditions. Finally, a detailed analysis of the flow topology and head losses inside the draft tube demonstrates the impact of proper inlet boundary conditions for the design of more efficient draft tubes.
Complete list of metadata

Contributor : Abes Star :  Contact
Submitted on : Monday, January 3, 2022 - 2:39:10 PM
Last modification on : Saturday, March 26, 2022 - 4:17:02 AM
Long-term archiving on: : Monday, April 4, 2022 - 7:20:19 PM


Version validated by the jury (STAR)


  • HAL Id : tel-03508014, version 1




Pedro Henrique Dias de Véras Sousa. Application of Artificial Neural Networks to the study of hydraulic turbine draft tubes. Chemical and Process Engineering. Université Grenoble Alpes [2020-..], 2021. English. ⟨NNT : 2021GRALI079⟩. ⟨tel-03508014⟩



Record views


Files downloads