TURBO

Turbocharging CFD

Turbocharging CFD

Paul Schreier examines how CFD software suppliers are customising their products to meet the needs of specific industries – in this case, for turbomachinery engineers

Scientific Computing World: August/September 2008

Turbines, pumps, compressors and other rotating machinery have been widely used for many decades, and they are reaching impressive efficiencies. Even so, we don’t know every detail about them and there is still room for improvement. Advances in software for the analysis of flow, heat and mechanical aspects is letting us gain deeper knowledge about the internal operations of such systems and their components. The payoff is both small and great: squeezing just one more per cent of efficiency from a gas turbine on commercial aircraft can pay huge dividends, especially with rising fuel costs and increasing demands to lower emissions. But these same techniques and savings can be applied to a wide variety of rotating machinery including pumps and compressors.

Today, numerical analysis has become a standard engineering tool for turbomachinery manufacturers. Marc Tombroff, general manager at Numeca International, explains that, 20 years ago, at any major aerospace company, engineering consisted of 20 per cent simulation and 80 per cent experiments; today the ratio is just the opposite. This shift is, in part, due to the fact that while CFD and other numerical-analysis codes were traditionally the domain of specialists, software developers are packaging their tools in such a way that they become much more accessible to everyday design engineers.

Turbo workflow

To better appreciate offerings from various software suppliers, it’s useful to examine the typical turbomachinery workflow. It starts with the basic specifications of aspects such as flow rate, inlet pressure and temperature, rotational speed, pressure ratio (or head rise for pumps), the fluid medium and others such as the physical envelope, noise and constraints including min/max thicknesses and lifetime. With this data, engineers can start sizing components and see how they might interact in what many suppliers refer to as a 1D or meanline analysis. With this basic 1D sizing information, engineers can select the most appropriate components, estimate performance characteristics and make refined manipulations of the final geometry using 3D geometry codes combined with quasi-3D calculations methods. Given acceptable component 3D geometries, software can generate the appropriate meshes for subsequent CFD and FEA evaluations. The next step is to perform progressively more detailed heat/flow/structural analysis in order to optimise the design both fluid dynamically and structurally. The final step is generating the CAD/CAM files needed for building prototypes and production versions.

The detailed analysis of turbomachinery can involve a large number of physics: fluid flow, heat transfer, structural mechanics and fatigue studies, but also electromagnetic effects and sometimes even combustion or other chemical reactions. Heat transfer is especially important, given that modern gas turbine engines operate at inlet temperatures beyond components’ melting point, even using thermal-resistant super alloys.

A screen from SoftInWay’s AxStream shows a Mollier (enthalpy vs enthropy) diagram for a turbine (left); the parameter distribution between stages, here Mach numbers and efficiency (top right); and a turbine flowpath sketch with calculations including capacity, efficiency and mass flow rate (bottom right).

Bottleneck shifts to the front end

While new solvers and parallel-processing libraries have greatly improved performance during detailed 3D analysis, the result is that bottlenecks in the design process have shifted towards preprocessing, according to Will Kellar, director of Cambridge Flow Solutions.

A number of companies agree, and they have developed software targeted specifically at front-end analysis, leaving detailed 3D CFD analysis to dominant players. For example, AxStream from SoftInWay accepts basic performance data from which it does flowpath preliminary design from scratch, through-flow streamline analysis, profiling and blade design, choosing the optimal number of stages and cascade angles, and even Euler-flow CFD analysis. The user ends up with a full flowpath geometry that can be exported to other packages, such as detailed 3D multiphysics CFD for validation or directly to manufacturing. Because its solvers are based on an empirical approach different from traditional finite-element methods, it takes significantly less time, and a total design cycle for an initial geometry can be finished in few hours.

Taking more of a global view, software from Flowmaster uses performance data to create a 1D model of an entire system. With Flowmaster V7 Gas Turbine, users combine drag-and-drop customisable components from an industry-tailored catalogue with several hundred entries including valves, orifices, controllers, pumps, tanks, labyrinth seals, annular passages and even a cavity component that models the passageway between disks. Other modules that work within the V7 environment include a Swirl Solver that calculates the swirl generated by rotating components and tracks how it propagates downstream; a Secondary Air module to model the air flow through and between rotating and stationary rotor and stator blades; and a Fluid Mixing module to calculate the fluid properties of the mixture, which is useful when modelling steam injection in combustors and fuel systems of power-generation turbines using syngases.

Here Flowmaster V7 Gas Turbine models the passageway between disks using a cavity component.

This overall systems view helps, for instance, to determine if the configuration of components might result in choking downstream in the process, something that is difficult to determine when studying individual components or else would take an immense CFD model that needs a very long time to solve. The software looks at all the flows, whether liquid or gas, through all passageways and pipes from the fuel system through all rotating components and even in the combustion area. It then generates data so that a CFD package can take bigger iteration steps in converging on the final solution, thus saving design time.

Flowmaster is among the companies that have set up a dedicated team to optimise its base software for specific industries. Michael Croegaert, product marketing manager for turbines, notes that while it might take several days to model the secondary air system around one disk, his software can allow engineers to create the model for the entire secondary air system in two to three hours and then run the analysis in roughly half an hour.

Cambridge Flow Solutions takes a more selective approach to addressing preprocessing needs with BoXeR, a geometry-editing and meshing tool based on computer-graphics constructs developed for interactive 3D gaming and whose highlight is topology independence. Many software packages can bend an existing mesh a bit, but editing the geometry topology – such as getting a torus from a sphere – is much more difficult. This is what BoXeR editing enables.

The need for fast geometry editing comes not only from the time it takes to create a mesh, but engineers invariably want to make design changes and create a new mesh. As for speed advantages, director Will Kellar presents this comparison: to properly resolve all the cooling-hole jet flows in a turbine blade (internals/externals, shroud, probably some secondary flow) you need at least 10 to 20 cells across each of the 250 or so individual holes, and in total this means as many as 100 to 150 million cells for the complete geometry. With standard software running in a serial mode, geometry creation and meshing takes probably a month if you’re lucky (if it’s possible at all, he adds) and requires a machine with 100GB of RAM; in contrast, the parallel version of BoXeR needs between 15 and 30 minutes when run on 16 PCs in parallel.

The full scope of design

1D analysis is something that Ansys is adding to its well-known broad range of CFD products so as to address all the steps in the design workflow. Further, given the company’s wide spread of capabilities, it’s not easy for someone who isn’t frequently involved with modelling to know which features exist and how to access them. Thus Ansys has started to develop industry-specific ‘virtual applications’ that hide unnecessary complexity and instead provide industry-specific user interfaces. For instance, when an engineer opens Ansys Workbench in ‘turbo’ mode, the software responds with specific panels helpful for setting up problems in rotating machinery as well as menu access to specialised tools such as BladeModeler.

This turbomachinery version of Ansys is among the first and so far the most complete of its virtual industry applications, but the company is expanding the number of such tools; for example, something similar is being developed for internal combustion engines.

Turbo-machinery specific user interface for Ansys when run in Turbo mode.

For 1D analysis, BladeModeler integrates software from PCA Engineers: the Vista CCD for preliminary design of centrifugal compressors and Vista CPD for performance prediction of centrifugal compressors. ‘If you don’t get the 1D features right,’ comments Graham Cox, engineering director at PCA, ‘you can never make it up in 2D or 3D. Simply, if the compressor is too small to begin with, it’s always going to be too small for the given task.’ PCA Engineers had developed several such packages for preliminary studies in turbomachinery, and more of them will be integrated into the next release, Ansys 12. PCA’s 2D throughflow solver, Vista TF will also be offered to users in that release.

On the other hand, if the geometry already exists, BladeModeler can import the data and prepare it for simulation and analysis and, if necessary, further modification. Next comes Ansys TurboGrid, which automatically creates a hex mesh for a blade geometry. The same geometry can be used for all types of studies such as flow and structural analysis. The module then interfaces to the Ansys CFX solver for simulations, and users can also run Ansys Mechanical and Ansys Fluid Structure Interaction products, as well.

These products illustrate well how software suppliers are presenting their products in a way suitable for turbo engineers. Ansys has been in the simulation business for more than 35 years, starting in structural analysis. But through internal developments and acquisitions (best known are perhaps the acquisition of CFX, Fluent and most recently Ansoft) the firm has expanded its product line to cover most major areas of numerical analysis. Making all these diverse products and their capabilities easily accessible to engineers was one of the firm’s major goals in the past few years, and it accomplished this through the Ansys Workbench, which provides a unified datasharing and project file management across the range of Ansys products as well as supplying a common user interface, no matter which underlying analysis engine is running.

Bring on the turbo wizard

To assist turbomachinery engineers in dealing with CFD, CD-adapco has also likewise developed some special software, a ‘turbo wizard’, to help set up and run its STARCCM+ software (the successor to STAR-CD). The wizard can read blade profiling data and create a 3D geometry, even for multistage devices. For mechanical stability and to allow for internal cooling, turbo components can have extremely complex geometries and flow paths, which when imported from CAD programs often need rationalisation. For this, Star-CCM+ features a surface-fixing tool that wraps the geometry with a single surface and then automatically generates a polyhedral mesh for the finite-volume analysis. In all three domains of interest – the hot outer flow, the metal that makes up the components and internal flow channels – the automated meshing tool handles them all simultaneously, and the mesh is continuous at the boundaries on all three domains. This then allows the use of a single solver on all three domains, which considerably speeds up analysis. In addition, says Fred Mendonca, manager of vertical applications at CD-adapco, this is the first commercially available code that also uses finite volumes for stress analysis on the components.

STAR-CCM+ from CD-adapco uses a colour code to show internal streamlines through the cooling passage of a turbine blade.

With its software, Concepts NREC has a unique spin. The firm is active in the design, manufacturing and test of turbomachinery stages, and over the years it developed in-house software which is oriented exclusively to turbomachinery and has been commercially available for several decades. ‘As turbomachinery designers,’ says chief engineer Colin Osborne, ‘we know what engineers want and need in such software, and our codes are tailored to this process. And thanks to all of our engineering projects, we are constantly exposed to the latest needs of turbomachinery designers and accommodate these needs in our codes.’

The company’s software all falls under the heading of the Agile Engineering Design System, which encompasses virtually every aspect of turbomachinery design. It starts with modules for the meanline 1D analysis of various components; an analysis takes only a few seconds so you can run hundreds of cases per day. Next comes streamline curvature analysis and ‘quasi 3D CFD’ analysis that solves the flow in 2D but makes accommodations in the underlying equations for 3D effects. The AxCent code includes a geometry engine that allows quick adaptations to the physical design. Next are full 3D fluid-flow analyses with Pushbutton CFD, which is integrated within AxCent. It performs geometry setup, establishes boundary conditions, automatically generates a structured grid, and sets up CFD control parameters. While Pushbutton CFD focuses on fluid flow and heat transfer, the software can also export geometry and mesh files to general-purpose CFD packages to allow specialised analyses such as for secondary internal flow passages, reacting or unsteady flow effects. With CFD results, engineers can use the Stressprep or AxiStress modules with FEA codes to structurally analyse components for rotational, thermal and pressure loads due to aerodynamic effects; similarly, DyRoBeS performs rotordynamics analyses. The final step is to use a CAM package from Concepts NREC to generate cutting instructions for 5-axis milling machines.

FINE/Turbo from Numeca can perform the full 3D unsteady nonlinear harmonic simulation of a multistage axial compressor in a few hours.

Neural nets for optimisation

Numeca specialises in CFD software for targeted applications, one of them being turbomachinery. Its FINE (Flow Integrated Environments) software relies exclusively on hexahedra meshes, which general manager Marc Tombroff says is well established as producing the most accurate results at the fastest speeds. He adds that generating high-quality grids is a challenge and that Numeca is the only supplier with a full automated hexahedral gridding tool. The FINE/Turbo design environment includes the Autogrid tool, which automatically generates a multiblock structured hex grid in a few minutes. It then performs a CFD analysis with or without conjugate heat transfer, and CFView adds qualitative and quantitative tools for CFD analysis. Additional modules handle cooling and bleeds, nonlinear harmonics for unsteady rotating flows, automated multistage and multipoint calculations, cavitation and modelling using real gases and liquids. FINE/Turbo can perform the full 3D unsteady nonlinear harmonic simulation of a multistage axial compressor in a few hours.

The company’s FINE/Design3D takes an unusual approach to geometry optimisation for blades and channels through the use of neural networks and genetic algorithms. It first imports an existing geometry and, in a type of reverse engineering, derives the blade parameters. The engineer then establishes goals for the parameters such as efficiency or mass flow. FINE/Design3D then runs a series of simulations with a space design of experiments methodology by calling FINE/Turbo (or another CFD tool) and sets up a database. A neural network examines the results, changes one of the parameters and runs another simulation. By repeating this feedback loop, the neural network eventually zeros in on the optimal geometry, sometimes with quite innovative shapes.