Design by numbers

Swiss engineers have claimed a breakthrough in hydraulic turbine design — CFD flow simulation for a complete Francis unit. A team from Sulzer Hydro and Sulzer Innotec has successfully modelled a complete Francis turbine — from the inlet of the spiral casing to the draft tube outlet — using a 3D Navier Stokes code with stage capability.

The approach could transform the way development engineers simulate Francis turbine performance, by helping them to design new runners that match existing components more accurately, at a lower cost than by using model tests.

According to Helmut Keck, head of research and development at Sulzer Hydro in Zurich, Switzerland, the CFD calculation has now reached the point where ‘off-calculation’ rather than ‘off-test’ procurement of turbines is becoming a real possibility. The main issue now, he says, is how far this new numerical technique can be used to reduce the number of tests needed for hydraulic turbines or, for smaller units and/or replacement runners, how it can replace expensive and time-consuming model testing.

Over the past ten years, Sulzer Hydro has used a 3D-Euler code for routine flow calculations in runners. The code provides a very good prediction of the three dimensional flow effects of the accelerated flow in a water turbine. A Sebestyen (head of CFD at Sulzer Hydro) and M Sallaberger (senior CFD engineer at Sulzer Hydro) and their colleagues also appreciate the accurate prediction of cavitation and the location of the best efficiency point. The good performance of numerous new runners developed by the company are a proof of the capability of the 3D-Euler design.

The method’s limitation is that it is inviscid, which means that the flow losses and efficiency characteristics over a complete hill chart cannot be predicted. ‘Up to now, turbine flow simulations have had to consider the stationary and the rotating components of a turbine separately’, M Sallaberger says. ‘Special boundary conditions are then needed at the interface between the distributor and the runner or between the runner and the draft tube, and these are not known at the outset,’ he added.

These boundary conditions have to be determined empirically or by overlapping separate calculations, and this constitutes a possible source of error and additional effort to the designer.

Mick Casey, head of the Fluid Mechanics Laboratory at Sulzer Innotec and based in Winterthur, Switzerland, says: ‘The advantage of using the 3D Navier Stokes code, in the latest version, is the possibility of coupling rotor and stator calculations to produce a stage calculation or a complete machine calculation by means of so-called ‘circumferential mixing plane interfaces’. Even operating points far from the ‘best efficiency point’ can be calculated with astonishing accuracy, so that a complete turbine hill chart can be generated by computer.’

The progress in numerical flow analysis with CFD opens up several completely new possibilities, says Sulzer:

•In large water turbine projects, where model tests are still needed for the customer to verify performance guarantees prior to manufacturing, the numerical simulation provides an additional development tool with great potential in terms of efficiency and operating behaviour (smooth running, cavitation safety and overload reserves).

•For upgrading out-dated power stations, new runners can be designed and manu-factured directly from numerical simulation without the need for model tests, bring-ing various cost and deadline advantages.

•Analysing the losses in the whole machine, and the inter-actions of the components in a single computation, increases the relia-bility of the design during the tendering phase and also during the design phase.

•The method makes it possible to identify those turbine parts which need to be improved, in addition to the installation of a new runner.

•The method eliminates user intervention and thus possible sources of error (especially with regard to boundary conditions).

•Flow variations in the meridional plane (ie from the axis to the casing) are automatically correctly reproduced without the cumbersome overlapping technique.

Although the computer resources required increase by a factor of a thousand, the time and cost for the simulation are still low, compared with those needed for a model test in a hydraulic laboratory.

Casey claims a world first for Sulzer in applying the new CFD technique with multiple frame of reference for a complete turbine. ‘But competition in CFD is tough,’ he says.

The simulation accurately reproduces the shape of the hill-chart and several important details of the flow in the turbine components, Sulzer says. There remains, however, a certain amount of inaccuracy in the exact predictions of the efficiency level (2% in absolute efficiency level). This remaining uncertainty is attributable to the level of grid refinement and the choice of turbulence model.

Ongoing development work already shows signs of improvement in the level of accuracy and has opened up

the possibility of unsteady state calculations.

The computational grid for the Navier Stokes simulation is shown opposite. The same turbine has also been extensively tested in Sulzer Hydro’s hydraulic laboratory in Zurich, Switzerland, using both conventional five-hole probes and laser optical velocimetry. It was thus possible to compare flow predictions at the various operating points with the detailed velocity and pressure measurements and with the turbine characteristics measured on the test stand.

Peter Drtina and Miriam Sick, both from Sulzer, agree: ‘The level of accuracy achieved by the simulation surprised us. We were extremely pleased to find that the excellent agreement was not only limited to the best point, but there was still good agreement between the tests and the simulation well away from the best point, even in operating points with low efficiency and large zones of reverse flow.’

Both the position of the best efficiency point in the hill chart and the form of the hill chart itself (shape of constant efficiency contour lines, see diagrams bottom left) were well predicted, Sulzer claims. A comparison of the calculated and measured characteristic curves for the relative efficiency (diagrams top left) demonstrates the agreement at different guide vane openings.

It is only at very large values of specific circumferential speed, Ku (corresponding to very low heads) that the deviation between efficiencies in the test and simulation, increases, says Drtina. At these operating points, the flow in the draft tube is extremely chaotic (on the test rig unsteady pulsating-flow conditions were seen), and the steady calculation cannot be expected to be accurate under these circumstances, he says (see opposite).

In addition to this, the simulation provides extensive detailed information about local flow phenomena, which are extremely valuable for optimising turbine design. The diagram at the top of p21 shows the pressure distribution in a section through a part of the stay and guide vanes, the runner channel and a part of the draft tube cone. The lowest pressures are calculated to occur close to the blade suction surface (point A) and in the swirling vortex in the draft tube (point B), precisely at the points where cavitation first occurs in the test stand.

The diagram below shows the ‘circum-ferential mixing plane interface’ between the distributor and the runner. Note that there is a pressure variation in the circum-ferential direction on both sides on the interface, but the mean pressure value is the same on each side. This, Sick says, demonstrates that the mixing plane interface couples the two without distorting the natural circumferential variations in pressure that occur in both regions. The diagram below shows calculated flow streamlines in the draft tube at an operating point slightly away from the best point. It is interesting to note that an intense swirling flow has already formed in one of the two diffuser channels.

Practical applications

The Sulzer Hydro team has verified the new computational method in the turbine design process by a number of practical applications.

‘The three dimensional stage computations give us detailed information on flow phenomena we can’t even observe on the physical test stand and we get the losses of the components at different points of operation,’ Sebestyen says. However, the computational effort for the simulation of a complete turbine is still very high, so Sallaberger and his colleagues have to select the components to be investigated and only the parts of critical interaction are computed with the new method during the process of design optimisation.

After optimisation, the most promising design is put onto the numerical test rig where the losses in the complete turbine are determined at several operating points and a numerical hill chart is generated.

After the validation of the stage simulation method for Francis turbines, Sulzer is extending its range of application to other hydraulic machines, like Kaplan turbines, pumps and pump turbines.

Sulzer Hydro has worked on some interesting European projects where its CFD approach to actual contracts has been applied successfully. The following are some examples:

•Innset, Norway

Replacement of two Francis runners with runner diameters of 2.25m and an output of 44.3MW.

The target was to deliver a new runner directly from the numerical test rig (without physical model testing).

Field test results have shown that the efficiency guarantee has been achieved and the output has been increased

by 15%, with simultaneous improvement in the cavitation behaviour.

•Rosenau, Austria

Upgrading of two Kaplan units, with an increase in runner diameter from 4.3m to 4.7m (plus 9.3%).

The problem was to design a new runner which would perfectly match the existing old draft tube in spite of the large increase of runner diameter.

The CFD optimisation of the flow field in the runner and draft tube finally allowed a large improvement in the output by 28%. Such an enormous increase in capacity was verified in a physical model test.

•Geesthacht, Germany

Replacement of three twin-runners of a single-stage, multi-flow storage pump (outer diameter 3.58m). The problem in this project was matching of the runner with the narrow intake and unusual diffusor vanes.

Due to the complex nature of this case, it was decided to verify the CFD design in a model test to allow fine tuning of the impeller outlet to shift the pump characteristic if necessary. It turned out that the predicted pump characteristic was met and that fine tuning was not needed.