Home » Research and Development » Hydropower Technology » Fluid Mechanics – Calculation » Artificial intelligence for enhanced hydraulic turbine lifetime

project:

Jan 2023

Jun 2027

Ongoing

Artificial intelligence for enhanced hydraulic turbine lifetime

Recent advancements in artificial intelligence and machine learning enables high-dimensional controlling and decision-making. In this project, state-of-the-art artificial intelligence will be developed to detect and control undesirable and damaging flow-induced oscillations to enhance turbine lifetime. A well-developed and trained model can not only detect the presence of damaging flow structures, but it can also take optimal decisions to reduce and control such structures.

Presently, the inevitable intermittency of electrical energy resources such as solar and wind power is compensated through hydropower systems. Meaning that hydraulic turbines are not necessarily working at the steady Best Efficiency Point (BEP) condition anymore as they are used in different off-design and transient operating sequences to stabilize the electrical grid. Such operations cause flow instabilities with pressure fluctuations, load variations, and cavitation, which may deteriorate the machine and reduce its efficiency leading to entirely different engineering requirements. Thereby, a sustainable energy production system cannot be achieved unless these damaging effects are mitigated, and the hydraulic turbines are adapted to new transient operations

Projektets huvudsakliga mål är att använda och vidareutveckla avancerad artificiell intelligens för att effektivt och robust upptäcka, kontrollera och mildra flödesinducerade svängningar vid drift utanför designområdet och under övergångsförlopp, för att förlänga turbinernas livslängd. För att uppnå detta mål kommer djupa neurala nätverk att utforskas genom förstärkningsinlärning för att möjliggöra optimalt beslutsfattande för vattenkraftturbiner. Det kommer även att undersökas hur fysik-informerade neurala nätverk (Physics-Informed Neural Networks, PINNs) kan användas för att minska tiden det tar att få fram noggranna numeriska resultat.

Contact

Håkan Nilsson

Research Area Responsible

Chalmers University of Technology

Email

Contact

Saeed Salehi

Doktorand, Chalmers tekniska högskola

Email

Contact

Mohammad Sheikholeslami

PhD, Chalmers University of Technology

Email

Published results

OpenFOAM Workshop: Implementation of deep reinforcement learning in OpenFOAM for active flow control.

Meccanica: An efficient intrusive deep reinforcement learning framework for OpenFOAM.

Physics of Fluids: Modal analysis of vortex rope using dynamic mode decomposition.

Physics of Fluids: Formation and evolution of vortex breakdown consequent to post design flow increase in a Francis turbine.

W​orkshop: Towards practical applications of deep reinforcement learning in computational fluid dynamics.

Reorganization of flow field due to load rejection driven self-mitigation of high load vortex
breakdown in a Francis turbine, Faiz Azhar Masoodi, Saeed Salehi, Rahul Goyal, September 6,
2024, Physics of Fluids Vol 36

Physics-Informed Neural Networks for Modeling Linear Waves, Mohammad Sheikholeslami,
Saeed Salehi, Wengang Mao, Arash Eslamdoost, Håkan Nilsson, August 9, 2024, ASME 2024
43rd International Conference on Ocean, Offshore and Arctic Engineering