Interactive Simulation

Hydro turbines are built for a very long time. Modern water turbines are a technically mature product, their peak efficiency reaches more than 95 %. But reaching this efficiency requires a vast engineering effort, especially due to the fact that hydraulic turbines usually are individual products and thus have to be designed with respect to the local conditions, such as head and discharge. That requires a tailor-made design for the different components of the turbine.

The traditional design process focuses on experiments, measurements, and model tests. This approach leads to a time and money-consuming development. In the last 15 years, CFD helped turbine designers in significantly shortening development times and saving money. By the use of modern computational resources, many problems can be detected in a very early development stage.

Turbine components are modelled in the computer, a computational mesh is generated and flow simulations are performed. Experiments are still very important and indispensable to validate CFD, but many of them can be replaced by simulations which play a prominent role to avoid serious mistakes in the design process.

There is a need for a tool that enables the designer to evaluate many geometry variations in a short time. To save manpower as well as money and to shorten development cycles, this tool must be based on numerical simulations. As engineers want to focus on turbine design and not on simulation details, this tool should allow working in an intuitive way.

This “virtual numerical test bed” is currently under development. It already works for some of the components. In this article, we will give an overview over the basic ideas.

Figure 1:
Wicked gate of bulb turbine in Ruppoldingen power station, generator housing in the background (Aare-Tessin AG, CH).
∆ 6m, P=9.5 MW

Figure 2:
Francis runner and wicked gate with streamlines in virtual testbed


Figure 3:
Francis runner and wicked gate meshes

The Virtual Water Turbine Testbed

The virtual turbine testbed is a numerical copy of a real (physical) water turbine testbed. The water flow through an entire turbine is calculated using advanced simulation techniques. The simulation aims at predicting unwanted phenomena such as vortexes, vibrations and cavitation, which are important charges for the material and have a negative result on the efficiency.

In order to treat the “virtual turbine” in the same way as in reality, the numerics must run invisible the background. That implies the generation of computational meshes, the definition of the boundary conditions according to the operating point of the machine and the computation of the flow.

The numerics should ideally run auto-matically. For verification purposes however, the user should be able to check details of the simulation.

As the analysis of geometry and simulation results must be fast, detailed, and intuitive, there is a demand for a Virtual Reality Environment (VR), in which the geometry of the turbine is displayed in conjunction with the simulation results in a realistic manner. It is possible to step inside the turbine and have a closer look into every detail of the machine.

The user can have a closer look at the pressure distribution on the surfaces or can examine the flow using streamlines, particle traces or isosurfaces. That helps in detecting zones of cavitation and vortexes.

Basically, everything the user can see and measure on a real testbed should also be possible in the computer model.

Given a specific turbine geometry, the most important target is to simulate the operating behaviour for an arbitrary operating point. Thus, especially the off-design behaviour of the turbine shall be studied. Subsequently, the geometry of the machine can be changed using the experiences given by the acquired knowledge.

Due to the existence of a virtual turbine testbed, model experiments become less significant. Especially the transferability problem from model experiments to real size prototypes is solved to a certain degree. The simulation of the machine can easily be done in real size without additional costs.

Design Process

In the computer-based design process, engineers in the first step define a machine geometry using complex CAD software. Afterwards, they use meshing tools that require a vast amount of expert knowledge to define the computational mesh.

Using the design modules developed at IHS (Institute of Fluid Mechanics and Hydraulic Machinery, University of Stuttgart), designing turbo machinery has become much easier.

The different modules are running under one consistent environment, COVISE, a visualization package developed at the HLRS. The whole design process is carried out from within the COVISE environment.

There are modules for all parts of the machine. These are the wicked gate, radial or axial runners and the draft tube. The turbine geometry is completely parameterized; a set of parameters describes the whole geometry. These parameters imply the shape of the runner blades and the guide vanes as well as the diameter of the runner or the moulding of the flow channel. The modules are flexible tools, some mouse clicks suffice to change these parameters and thus create a new virtual turbine that can serve as a starting setup for further geometry improvements using flow simulations.

The simulation process starts with the definition of the machine‘s geometry. Simultaneously, the corresponding boundary conditions that describe the operating point of the turbine are defined.

By pressing a button, an unstructured computational mesh consisting of hexahedron elements is generated.

As the modules contain automatic grid generation routines, generating a mesh does not need any user interaction. Generating a grid consisting of 100,000 elements needs about 4 seconds on modern desktop PCs.

In the next step, the mesh is decomposed for parallel processing.

Figure 4:
Process chain of iterative design

Online Simulation

“Online simulation” means that it is possible to engage into a running simulation and that an immediate validation of the simulation results is possible. Standing inside the virtual turbine, the designer can change parameters of the machine, e.g. blade profiles or the shape of the flow channel. A new computational grid is generated immediately. Within a short time range, new stable simulation results are obtained. COVISE simulation library is used to couple FENFLOSS, the Navier-Stokes based flow solver developed at IHS, with the FENFLOSS COVISE module. As the communication uses a TCP-socket connection, the simulation part can run on any computer.

To achieve an intuitive design process, it is important to have short and almost online response times from the simulation. The engineer uses his expert knowledge to decide whether the changes led to the desired result or not and tries to optimize the turbine in an iterative process.

Of course, the intention is to obtain simulation results that are as close to reality as possible.
Since the test bed shall be used for all important components (distributor, runner, draft tube), the use of massively parallel code on supercomputers is essential.

COVISE

We use the COVISE visualization system to integrate the whole process from grid generation, simulation to analysis. Each of the modules in this workflow can reside on a different computer. This allows distributing the work load among different machines. Typically, the pre- and post-processing modules run on a visualization server while the simulation runs on a remote supercomputer. The display modules either run on a visualization cluster which drives a CAVE or tiled wall or on the workstation of a user.

COVISE is a modular visualization system developed at the HLRS. The software uses a data-flow execution model, i.e. the data objects in COVISE flow through a network of modules. The modules all run as separate processes and thus can be distributed among multiple computers.

It is possible to couple COVISE environments for collaborative working.

COVISE can be used not only for off-line post-processing and visualization; it can also be used as a general distributed and collaborative integration platform. This allows integrating grid generation, simulation and post-processing modules in a seamless way to create interactive engineering applications.

The simulation process chain consists of three COVISE modules which have been developed to integrate all the aforementioned processing steps into one environment.

At first we need a module that defines the geometry and generates the grid and boundary conditions. This can be the Gate module, one of the runner modules or the Draft Tube module. Next, the Domain Decomposition module, which decomposes the grid into multiple domains for parallel simulation, and the FENFLOSS module, which couples the simulation code to COVISE. The entire COVISE dataflow network is shown in Figure 5. The simulation itself is a separate process that is coupled with the FENFLOSS COVISE module using a socket connection. It sends new data to COVISE after each global iteration. All the other modules in Figure 5 are used for data analysis and visualization, e.g. Tracer and Cutting Surface modules.

Figure 5:
Process chain in COVISE as a visual program with FENFLOSS simulation coupling

Application Example: Kiebingen Power Station

Modern CFD was used to increase the efficiency of Kiebingen power station, a small plant situated at the river Neckar in Southern Germany. The four machines, that are aged almost a century, needed some redesign. The power output per machine could be increased by 30 %, simply by substituting runner and wicked gate. All other parts were retained.

Figure 6 - Figure 9 show Kiebingen power station in real world and in VR, together with simulation data.

Figure 6: Kiebingen power station                                Figure 7: Kiebingen power station in VR

Figure 8:                                                                       Figure 9:
Inside inlet building of Kiebingen power station           Online simulation of water flow in Kiebingen wicked gate in the CAVE
in VR, wicked gate in the background

Tangible Interfaces

As a further enhancement of the described simulation techniques, we have developed an airflow simulation in urban planning using a tangible interface as an easy and effective solution for interaction.

The user can control the simulation and visualization through manipulation of physical objects. These movable objects are equipped with black and white markers which identify the object. A high resolution IEEE1394 Camera is positioned above a physical model of the simulation domain. The camera picture is captured and analyzed by a modified version of ARToolKit (developed by Hit Labs, University of Washington). It returns the position and orientation in six degrees of freedom of all markers which are completely visible.

Figure 10:
Tangible Interface of Stuttgart train station urban planning example

Tangible interfaces can be used as controls for digital parameters, data sets, computing resources, and other digital content. There are many advantages of this approach. First of all, the parameters that can be changed using the tangible objects are represented in a very clear way. The interaction is simple due to the fact that it is so close to reality. In addition to that, tangible interfaces can be used on desktop applications as well as in immersive environments.

Users can explore the solution collaboratively by sharing multiple screens or immersive environments distributed in a room, building or anywhere around the globe. Results of the simulation are available in real-time.

A VRML model of the objects which are simulated serves as a visual reference in the virtual environment and it defines the tangible interface. A new VRML node, ARSensor, acts as interface between ARToolKit and the VRML model. It takes the position and orientation of one of the markers from ARToolKit and transforms it into the VRML coordinate system. This transformation can then be routed to the transform node of the tracked object which thus follows the movement of the marker and the physical object respectively.

Figure 11:
Grid around urban planning buildings.

An automatic grid generator receives the position of the objects and generates a hexahedronal computational mesh around the objects which is then decomposed for parallel processing. Figure 11 shows the decomposed grid with each partition coloured different.

By using tangible interfaces, a much more natural mode of operation is possible than with a traditional user interface where it is usually necessary to type in coordinates to move objects. The natural perception of the model in comparison to a simple monitor image is another important advantage.

Urban Planning Example

In an interactive airflow simulation of downtown Stuttgart, we simulate the impact of architecture on the fresh air supply of the city.

Stuttgart is located in a basin. In Figure 12 the topography of the area is shown. The red sphere represents the downtown district around Stuttgart main station and the castles.

Figure 12:
Topography of Stuttgart (z-scaled)

The rail-tracks will be removed and put underground within the next years. This will reveal a huge site for new buildings right at the heart of Stuttgart. As there are hills almost all around the city, the area is an important corridor for the city’s fresh air supply (see blue arrows) and thus is of significant importance for the Stuttgart locale climate.

The challenging task is to find a construction plan which bears that in mind.

The dataset covers an area of 250 x 300 m of the area behind the Stuttgart railway station. The boundary condition is a wind profile with low speed on the ground and higher speed towards the top of the grid. The wind direction can be changed interactively. Six buildings can be placed in that area by either changing parameters of the grid generation module or by moving the physical buildings in the model.

We have built a model of the construction site (Figure 10). The airflow around these buildings is simulated using our described approach. By moving the modelled buildings around, many different layouts can be tested in a very short time while at the same time the user gets a vivid impression of the architecture in the virtual environment. Cutting planes and particle traces can be positioned by moving a marker in the model or in the virtual environment with the 3D mouse.

Figure 13:
Screenshots of the VR-Display, cutting-surfaces, particle-traces and isosurfaces

This is just one example for using tangible interfaces in conjunction with online simulations. Some other possible application scenarios for this simple grid generator are the layout of clean rooms, where a laminar airflow is required or the allocation of supercomputer racks in a computing room where it is important to design and dimension the air-condtion.

• Martin Becker
Höchstleistungsrechenzentrum Stuttgart (HLRS), Germany

  top