Optimization of transient operation of hydraulic machines and active flow control using machine learning
Abstract
This storage project is connected to Large compute-project SNIC 2022/3-41
The Large compute-project SNIC 2022/3-41 will provide computational resources for the following research areas:
1. Transients in hydraulic machines, such as shutdown and startup.
2. Machine learning algorithms for active flow control in hydraulic machines (DRL/PINN).
3. Optimization of novel contra-rotating pump-turbines in transient operation.
4. Turbulence: Resolved simulations of high-Re flows in complex geometries (URANS, DES, LES).
5. Cavitation: A two-phase flow phenomenon requiring extreme mesh resolution, short time steps, and long real-time simulations.
Three main applications will be investigated using computational fluid dynamics (CFD). Those are (1) transients in hydraulic machines, (2) active flow control to mitigate damaging flow structures, using machine learning, and (3) optimization of transient optimization sequences. In all three applications, it is necessary to perform highly resolved CFD simulations of the flow, using hybrid/DES/LES turbulence modelling techniques. Such simulations are resource demanding already for simple applications. At the high Reynolds numbers and in the complex geometries of the present applications, it is much more demanding. We particularly need to study long real-time events and include both mesh rotation as well as mesh deformation due to changes in operating conditions. Cavitation is a two-phase flow phenomenon that occurs under transient operation of hydraulic machines, which involves phase change as the local pressure passes the vaporization pressure. It requires highly resolved simulations both in time and space. We have, during many years, been part of the development and validation of the methods and models needed to do these kinds of simulations using the OpenFOAM open-source software, and all of the functionality to successfully do the studies are now in place.
Storage requested through the present proposal:
Due to the fact that we will be studying transients between different operating conditions, we need to save data at many time steps, in order to be able to post-process the results after finishing the simulations. I.e., this is not regular 'unsteady' simulations, but the conditions are also varying. The machine learning project involves a large number of computational cases, which both needs disk space but also a large number of files. Cavitation simulations require very fine meshes, which increases the need of disk space. We also need to install several versions of OpenFOAM and related software, which takes a lot of space and has many files.
