Final Report Summary - SITHYM (Simulation of Transients in Hydraulic Machines)
The main goal of the SiTHyM project was to explore alternative numerical simulation techniques able to back design engineers in making robust hydraulic machines. Hydraulic and mechanical engineering of machines used to produce hydropower is facing new challenges coming from the profound evolution of the energy production sector in Europe. The growing production from intermittent renewables (wind, solar) require balancing capacities to ensure stable electricity grids. Hydropower plays a key role in this stabilization: hydropower plants can be quickly synchronized on the grid or disconnected, they have a flexible operation range. Pump storage plants can store the energy surplus produced by less flexible or intermittent sources. Hydropower plants are thus operated more and more at off-design conditions. They also endure more and more frequent start-up and shut-down cycles. These peculiar conditions are very demanding for the machines’ structure: flow conditions poorly match the hydraulic design and lead to high solicitations and vibrations.
It is necessary for engineers to account for these new operating conditions and to create robust designs able to withstand solicitations safely. It is also important to have tools in order to assess the capacity of the installed fleet of machines to operate safely in operating modes that were not common at the time of their design.
Predictive tools need to be able to simulate for example flow features and hydraulic loadings during start-up and shut-down of units. These phases are highly transient. The control of the machine involves moving components (guide vanes, torpedos, deflectors) that are very challenging for classical CFD software based on a computational mesh. Indeed the motion of boundary conditions lead to important mesh distortion and numerical inaccuracies, so that re-meshing is required. In the SiTHyM project, it was proposed to use a numerical method that do not rely on a computational mesh to solve transient flows in hydraulic machines.
The Smoothed Particle Hydrodynamics (SPH) method is based on a mathematical formulation that is able to solve the Navier-Stokes equations on a set of disordered and moving calculation points. It is well adapted to the prediction of free surface flows with high dynamics. ANDRITZ Hydro developed in collaboration with Ecole Centrale de Lyon a variant to the standard SPH method called the SPH-ALE . It is an hybrid method that makes use of the numerical kernel of SPH and of numerical flux schemes taken from the Finite Volumes method. It is used in ANDRITZ Hydro for the hydraulic engineering of Pelton turbines. Its mesh-less paradigm was identified as a potential advantage to manage moving boundary conditions for internal flows. The purpose of SiTHyM was then to extend the application domain of SPH-ALE to internal flows, in order to get a predictive tool for the transient start-up and shut-down phases.
Internal flows is not the natural and easiest application domain for SPH and SPH-ALE. The lagrangian motion of the calculation points can lead to holes in the particle distribution in regions with flow detachment. Moreover the standard SPH numerical kernel is intrinsically isotropic, which makes directional refinement of the particle discretization difficult. Accordingly standard SPH does not appear to be computationally efficient compared to other techniques like Finite Volumes. The strategy that has been adopted in SiTHyM is to couple SPH-ALE with Finite Volumes in order to get the best of the two approaches: Finite Volumes would be used around hydraulic components in order to capture boundary layers properly and SPH particles would be used in the rest of the domain to capture the main flow.
Three main tasks have been adressed in SiTHyM: the coupling, the improvement of SPH-ALE for internal flows and the High Performance Computing.
The coupling strategy has been presented in the intermediate report. At the time of the end of SiTHyM, the extension to 2D and 3D is still in progress.
Improvements of SPH-ALE for internal flows include the development of proper inlet/outlet boundary conditions based on the Navier-Stokes Characteritic Boundary Conditions. They include a non-reflection treatment that is essential for real applications. An arbitrary particle motion has also been developed and proves to be very efficient for internal flows with moving components.
The High Performance Computing relies on an hybrid parallel strategy allowing multi-GPU computing. Speed-ups brought by one single GPU amounts to x20 to x25 compared to a single CPU core. A domain decomposition layer combined with Message Passing Interface allows the use of several GPU cards in the same virtual computer.
It was possible to simulate the start-up of a Pump Turbine in turbine mode with the pure SPH-ALE method. Despite the coarse discretization used, it was possible to measure the hydraulic loading on the various hydraulic components in the course of the runner acceleration and flow development.This type of numerical prediction is of paramount importance for mechanical engineers. The work will go on with higher fidelity simulations. In particular the coupling of SPH-ALE and FV (expected Q2 2014) will be of high value to better capture rotor-stator interactions.
It is necessary for engineers to account for these new operating conditions and to create robust designs able to withstand solicitations safely. It is also important to have tools in order to assess the capacity of the installed fleet of machines to operate safely in operating modes that were not common at the time of their design.
Predictive tools need to be able to simulate for example flow features and hydraulic loadings during start-up and shut-down of units. These phases are highly transient. The control of the machine involves moving components (guide vanes, torpedos, deflectors) that are very challenging for classical CFD software based on a computational mesh. Indeed the motion of boundary conditions lead to important mesh distortion and numerical inaccuracies, so that re-meshing is required. In the SiTHyM project, it was proposed to use a numerical method that do not rely on a computational mesh to solve transient flows in hydraulic machines.
The Smoothed Particle Hydrodynamics (SPH) method is based on a mathematical formulation that is able to solve the Navier-Stokes equations on a set of disordered and moving calculation points. It is well adapted to the prediction of free surface flows with high dynamics. ANDRITZ Hydro developed in collaboration with Ecole Centrale de Lyon a variant to the standard SPH method called the SPH-ALE . It is an hybrid method that makes use of the numerical kernel of SPH and of numerical flux schemes taken from the Finite Volumes method. It is used in ANDRITZ Hydro for the hydraulic engineering of Pelton turbines. Its mesh-less paradigm was identified as a potential advantage to manage moving boundary conditions for internal flows. The purpose of SiTHyM was then to extend the application domain of SPH-ALE to internal flows, in order to get a predictive tool for the transient start-up and shut-down phases.
Internal flows is not the natural and easiest application domain for SPH and SPH-ALE. The lagrangian motion of the calculation points can lead to holes in the particle distribution in regions with flow detachment. Moreover the standard SPH numerical kernel is intrinsically isotropic, which makes directional refinement of the particle discretization difficult. Accordingly standard SPH does not appear to be computationally efficient compared to other techniques like Finite Volumes. The strategy that has been adopted in SiTHyM is to couple SPH-ALE with Finite Volumes in order to get the best of the two approaches: Finite Volumes would be used around hydraulic components in order to capture boundary layers properly and SPH particles would be used in the rest of the domain to capture the main flow.
Three main tasks have been adressed in SiTHyM: the coupling, the improvement of SPH-ALE for internal flows and the High Performance Computing.
The coupling strategy has been presented in the intermediate report. At the time of the end of SiTHyM, the extension to 2D and 3D is still in progress.
Improvements of SPH-ALE for internal flows include the development of proper inlet/outlet boundary conditions based on the Navier-Stokes Characteritic Boundary Conditions. They include a non-reflection treatment that is essential for real applications. An arbitrary particle motion has also been developed and proves to be very efficient for internal flows with moving components.
The High Performance Computing relies on an hybrid parallel strategy allowing multi-GPU computing. Speed-ups brought by one single GPU amounts to x20 to x25 compared to a single CPU core. A domain decomposition layer combined with Message Passing Interface allows the use of several GPU cards in the same virtual computer.
It was possible to simulate the start-up of a Pump Turbine in turbine mode with the pure SPH-ALE method. Despite the coarse discretization used, it was possible to measure the hydraulic loading on the various hydraulic components in the course of the runner acceleration and flow development.This type of numerical prediction is of paramount importance for mechanical engineers. The work will go on with higher fidelity simulations. In particular the coupling of SPH-ALE and FV (expected Q2 2014) will be of high value to better capture rotor-stator interactions.