Case Studies

Meshing Time Reduction

Honda Automobile Explains How They Save CPU Time with HEXPRESS™/Hybrid

Automotive | Grid Generation

When you gaze upon the beauty of a car’s curves, how often do you think about people who worked for you to be able to enjoy this view? Let’s not forget that designers are working hand in hand with engineers: these cars, despite their eye-catching look, have to satisfy the manufacturability, a low aerodynamic drag, structural robustness, etc.

Many of Honda cars are designed in the Honda Automobile Research & Development Center located in Tochigi, Japan. This facility houses state-of-the-art research equipments: a car-to-car crash test, a proving ground (with a road of Belgian brick!), wind tunnels, etc.

But prior to all these physical tests, Honda Engineers perform and analyze CFD computations on external aerodynamics of the car body, as well as aero-thermal computations of the underhood: radiator fans, flow around engine bay/peripherals, exhaust system, etc.

All these computations require a high level of accuracy from the software used and a fast turnaround time. The product has to be mature and robust enough to meet industrial expectations.

“We massively invested in NUMECA solutions more than 2 years ago, and are very satisfied with the level of precision they provide us. We previously used various other commercial meshing tools, but were not satisfied by the quality of the viscous layers which many times led to divergence.

Moreover, the learning curve was cumbersome. When we switched to HEXPRESS™/Hybrid, we not only solved these issues, but also divided by 3 our CPU time, while the engineering time dropped to 30 mins/mesh.

Thanks to HEXPRESS/Hybrid’s speed, the reduced turnaround time allows my team to explore more designs in a shorter amount of time. This tool is easy to get started with; my engineers are quickly able to deliver results.

The complexity of our geometries forces us to be very demanding regarding our CFD tools. We needed a mesher capable of easily handling unclean CAD with overlapping surfaces, holes, intersecting selections, etc. HEXPRESS™/Hybrid appeared to us as the perfect solution for this job.

We also greatly appreciate its Thin-Wall feature, allowing us to get a conformal interface between 2 different domains such as a pipe and the fluid inside of it. The level of details we can now capture is impressive, and improves the accuracy of our results and thus our products.

Furthermore, the service and technical support provided by NUMECA Japan meet our highest professional expectations. NUMECA experts interacted closely with us to understand our specific needs and proposed dedicated solutions, all the way from support to custom developments.”

Dr. Takiguchi,
11 Development division, Chief Engineer
Honda Automobile R&D Center

Multi-Disciplinary Optimisation

Multi-Disciplinary Optimisation of a FORD Turbocharger Compressor Design

Automotive | Turbomachinery | Grid Generation

There are several reasons why optimisation methods can at this time not be considered in routine design work:

  1. Lack of computational resources: Everybody is enjoying that those resources have increased quite a lot recently, but nevertheless design optimisation work is still seen as too costly.
  2. The need for multi-operating-points: For quite some time optimisation and inverse design techniques would be focused on one single operating point, with no guarantee or control of the off-design performance, choke flow, performance at lower speeds,… Investing in a significant effort without guaranteed ROI is clearly limiting the interest of designers working under pressure and with time constraints.
  3. The multi-disciplinary need: Especially for designs working under strong mechanical pressure, such as turbochargers, an optimisation limited to aerodynamic performance and not offering control or verification of the mechanical integrity is also not so interesting.

NUMECA’s FINE™/Design3D design optimisation software offers solutions to all 3 items mentioned above:

  1. We considerably optimised our solver speeds: The FINE™/Turbo solver converges in typically 30 minutes to 2 hours per million nodes and per core. With a dozen of cores, the aero-analysis of a new design at 3-4 operating conditions can be done in 2 hours! An impressive result compared to standard commercial solvers that need an entire day for this.
  2. Optimisation is flexible and can easily address multi-operating point problems. In the example shown below, performance is investigated and controlled from stall to choke, at different speeds.
  3. The flexibility of the optimisation allows for the analysis phase to include a CFD solver, but also a mechanical tool (or other solvers). NUMECA developed a partnership with the company Open Engineering, and we have coupled FINE™/Design3D with the FEA mechanical solver Oophelie.

The optimisation example case described below has been presented at the 2015 IGTI Turbo Expo conference (GT 2015-43631). NUMECA-USA collaborated with the FORD turbocharger research group on this project. We started from an initial compressor blade design that already presented quite high performance, with the objective of mainly decreasing the mechanical stress levels by 20%. No clear objectives were set on the aero-performance, apart from trying to maintain it if possible.

Mechanical Optimisation

The process started with an optimisation of the back plate and bore zones, which only necessitated running the mechanical analysis tool Oophelie. The blade modeler AutoBlade™ has been upgraded to allow for the parameterisation of the back plate, as shown below. Once the design parameters are selected and variation bounds applied, our optimisation process works in 2 steps:

  1. random geometries are generated and analyzed by the FEA tool (DOE process).
  2. the information resulting from this first process is used by the optimiser to find the optimum.

The whole analysis process is executed in batch mode. Once the parameters are selected, a CAD definition of the geometry is generated, which is provided to the mesh generator (see resulting mesh below; note that fillets are automatically applied to the blades, even if not included in the geometry). The resulting mesh is then resolved by the FEA tool.

Aero-Mechanical Optimisation

The mechanical optimisation process was done in one night. It was followed by a complete optimisation of the blade shape and meridional channel, involving both the CFD analysis tool FINE™/Turbo and the FEA solver Oophelie.

Several speed lines of the initial design have been analyzed. The computational domain includes not only the wheel, but also the volute and the casing treatment. We then decided to focus on the choke and stall conditions at design speed and on the near-stall conditions at a lower speed. This decision was made based on the assumption that the design performance would be indirectly controlled if the choke flow and near-stall performance are maintained or improved.

The blade camber lines have been parameterised, as well as the hub path, which was controlled by Bezier points. Also we included the camber line profiles of the splitter blades and their stream-wise and tangential positions. This led us to a total number of 19 parameters.

450 geometries have been randomly generated. We decided to first calculate them with the mechanical tool. This allowed us to eliminate 300 of them, as they were providing higher maximum stresses. We associated poor aero-performance to those (without actually running the CFD) and then applied the CFD solver only to the 150 better ones, saving a lot of computational time. After optimisation, a new blade design was selected, presenting slightly better pressure ratio performance, and with 20% lower mechanical stresses. The pressure ratio curves are shown beside.

Acoustic Analysis

Side Mirror Noise with Adaptive Spectral Reconstruction

Automotive | Acoustics

A new method called Adaptive Spectral Reconstruction (ASR) for the stochastic reconstruction of broadband aeroacoustic sources starting from steady CFD analyses is presented and applied to the evaluation of the noise radiated by a model automotive side mirror.

The new approach exploits some ideas from both SNGR and RPM, and for some aspects can be considered as a sort of mixing between the two methods since it permits to reconstruct both the frequency content of the turbulent field (as done by SNGR) and the spatial cross correlation (as done by RPM).

The turbulent field is reconstructed with a sum of convected plane waves, but two substantial differences are introduced in respect of SNGR. The first difference concerns the spatial variation of the parameters that define each wave, that depends on the wavelength of each wave, rather than being kept constant or related to the CFD correlation length. The second innovative aspect is the usage of a dedicated full hexa adaptive mesh that is refined in function of the expected local correlation length, ensuring that the mesh be enough refined to capture the relevant acoustic length scales.

The method is here applied to the evaluation of a classical side mirror model test case, and results are presented in terms of comparisons with measurements for both in plane and out of plane microphones. Visualizations of reconstructed acoustic sources are also presented.

Aerodynamic Boat Design

Ultra-Fast and Fuel Efficient Aerodynamic Boat Design by A2V with FINE™/Marine

Marine | Hydro

Advanced Aerodynamic Vessels (A2V) develops and commercializes a new generation of fast transportation vessels, using aerodynamics to improve energy efficiency. Its revolutionary shape transfers the weight of the ship from the water to the air. As the required propulsive power depends mostly on the weight carried through water, reducing this weight significantly reduces fuel consumption.

“A2V technology has changed the rules by developing and commercializing a new generation of fast and fuel efficient passenger transport vessels. They have designed and patented a wing-like catamaran geometry, which at speed provides aerodynamic lift above the water, alleviating the vessel and thus reducing its power requirements.”

Conventional Fast Vessels

Today’s conventional fast vessels’ speed depends purely on power. As a consequence increasing speed inevitably comes at the cost of much higher fuel consumption. From an economic and environmental point of view, this leads to an unsustainable cost per passenger.

Meanwhile in a conventional monohull or multihull, fuel consumption per passenger is directly linked to the size of the vessel: the bigger the vessel, the lower the fuel consumption is per passenger when the boat is fully loaded.

Thanks to the Aerodynamic Lift, Faster Means Lighter and More Efficient

A2V technology has changed the rules by developing and commercializing a new generation of fast and fuel efficient passenger transport vessels. The company markets workboats for the offshore industry, the commercial passenger maritime transport industry, and the states for their patrol, surveillance and rescue missions. How? They have designed and patented a wing-like catamaran geometry, which at speed provides aerodynamic lift above the water, alleviating the vessel and thus reducing its power requirements. Thanks to the aerodynamic support, above a critical speed, the faster A2V vessels go, the less fuel they use.

“The A2V vessel has a fuel consumption of about 9 litres per passenger per 100km at 50 knots, independent on vessel size, from 10 to 100 passengers, from 12 to 30 meters. As a comparison, present state-of-the-art crewboats typically burn more than 30 litres per passenger per 100km and travel below 40 knots”. (source:

Numerical Challenges

One of the challenges in the design of such a vessel is the modelization of the free surface deformation at high speed, to which the stepped hulls are very sensitive since the aft part of the hull is operating in the wake of the forebody. The aerodynamic with the ground effect, both in steady and unsteady conditions, has also been a challenge. In order to model the behavior in the waves with reasonable computational times, A2V carried out systematic aerodynamic analyses and used the results to build a mathematical model. This aerodynamic model was implemented in the hydrodynamic computations using FINE™/Marine dynamic libraries.

“CFD is the core of the design approach of this vessel. With NUMECA’s software FINE™/Marine, A2V relied entirely on simulation during the design of the fully instrumented 10.5m prototype.”

Accurate Predictions

Computational Fluids Dynamics is the core of the design approach of this vessel. With NUMECA’s software FINE™/Marine, A2V relied entirely on simulation during the design of the fully instrumented 10.5m prototype. The full scale measurements showed that numerical predictions were accurate, as shown in the Figure 1 where numerical predictions are continuous lines and measurements are dots. The total drag is plotted versus speed for three different cases : (1) 10 knots head wind in red, (2) no wind in green and (3) 10 knots tail wind in blue.FINE™/Marine also allowed detailed analysis of the flow to refine the design.


Lionel Huetz, CEO,
Advanced Aerodynamic Vessels, France

Wind Loads

Validation of Wind Loads on a Slender Vessel Using CFD by DAMEN & NUMECA

Marine | Hydro

DAMEN and Numeca are developing a CFD methodology to demonstrate a vessel has sufficient transversal stability to resist over-rolling in severe side winds.

With more than 80 percent of the total global trade being transported through international shipping, the world relies on a safe, secure and efficient international shipping industry. Being a truly international industry, it can only operate effectively if regulations and standards are agreed, adopted and implemented on an international basis. The IMO (International Maritime Organisation) is the United Nations specialised agency that provides the regulatory framework for this process.

One of those regulations, IMO regulation 749.18, ‘Severe wind and rolling criterion (weather criterion)’, ensures a vessel has sufficient transversal stability to resist over-rolling in severe side winds. Due to the necessarily conservative nature of the regulation (enabling it to be broadly applicable to a multitude of vessels), slender vessels like the DAMEN Fast Crew Support (FCS) 3307 have difficulties in satisfying the empirical requirements of the regulation, thus necessitating expensive experimentation in order to demonstrate the vessel’s compliance with the regulation.

In order to reduce the cost of proving compliance, DAMEN is developing, in partnership with its Computational Fluid Dynamics (CFD) code supplier NUMECA International, a CFD methodology that can be used in lieu of experimentation.

They conducted a CFD validation campaign where a wind tunnel test of the DAMEN FCS3307 vessel was numerically replicated.

As the motivation for using the CFD approach is primarily cost and time related, it is imperative that the methodology be both sufficiently accurate but also with as low as possible computational cost and total turnaround time. The methodology reflects these aims.

Read the Paper

Read the Paper

Click here to read the paper that reports the results of this validation campaign.

Read the Paper

Planing Hull Resistance

Marintek Chooses FINE™/Marine for Planing Hull Resistance Curve Prediction

Marine | Hydro

For the prediction of the resistance curve of a planing hull and validation against model tests, Marintek used FINE™/Marine and HEXPRESS™ in this user case.

The Norwegian Marine Technology Research Institute (MARINTEK) performs R&D in ocean technology for a global market, primarily in the maritime and oil and gas sectors and ocean energy. MARINTEK’s main offices and laboratories are located in Trondheim, Norway. A rational combination of physical and numerical modeling approaches has always been MARINTEK’s strategy in its research activities and commercial services. Prediction of vessel and propeller performances, design optimisation process, wake analyses and studies on propulsor-hull interaction are just a few examples of CFD applied to ship hydrodynamics.

STM was established in 1991 by the decree of the Defence Industry Executive Committee to provide system engineering, technical support, project management, technology transfer and logistics support services for Turkish Armed Forces (TAF) and Undersecretariat for Defence Industries (SSM) and also develop necessary software technologies for defence systems and establish/operate national software centers for software development, maintenance/support.

User Case

The total ship resistance of a new patrol vessel in calm water conditions was evaluated by means of CFD simulations and model tests. The hull form was developed by STM and has as main dimensions:

Length betw. perp.












The primary objective of this case study was to validate the numerical model against model tests in order to guaranty high level of accuracy to the End Client in the subsequent numerical phases of the project. CFD and model test predictions were compared for three speeds: 20, 45 and 55 knots.

CFD Simulations

The total resistance of the hull is computed by means of CFD simulations at different speeds. The simulations were performed with FINE™/Marine 4.2 at full scale, in deep water conditions, in sea water.

The vessel’s geometry is meshed using HEXPRESS™. A boundary layer grid normal to the hull surface is specified in order to reach y+ values between 30 and 80. Given the diversity of Froude regime covered in this study, new meshes were generated for each computed speed. Adaptive grid refinement was used with the free surface criterion in the proximity of the hull in the final stage of each simulation in order to increase the accuracy of the results. The final meshes were composed of between 5.0 and 7.5 millions cells.

In the simulations, the propulsion was modeled as a force applied at the center of action of the water jets. The air drag was modeled as a force applied at the center of the frontal projected area.

Model Tests

The hull model is made of foam and wood coated with paint with a hydrodynamically smooth surface finish to the linear scale of 1:16. For turbulence stimulation, fine sand grains were glued to the hull along the keel from bow to station 17.

The resistance tests were performed with the model towed by MARINTEK’s high speed rig with measurements of resistance, trim and sinkage. In the model test setup, the model is free to heave, roll and trim but fixed in all other degrees of freedom.

The effect of air drag on the projected area above water line are included in the predictions based on the projected area of the vessel.

Conversion to Total Ship Resistance

The conversion from hull model (numerical or experimental) into full scale ship is made by using the form factor method. In this method, it is assumed that the total resistance can be divided into two parts, represented by the viscous resistance and the residuary (due to vorticity, wave making and wave breaking) resistance CR. The viscous resistance is determined by multiplying the frictional resistance CF with a constant form factor k0, which is identical for the models and the ship. Further, it is assumed that the residuary resistance CR is identical for models and the ship.

When numerical or experimental results are converted to Total ship resistance RTs, the effect of the hull surface roughness is taken into account by means of empirical formula. The results are presented in terms of non-dimensional Total ship resistance CTs, in which the dimentionalization is performed using the dynamic wetted length and surface of the vessel.


The following table compares the predicted total ship resistance obtained from the model test approach and the CFD approach. For all speeds, the results agree within 0.7 %. The hydrodynamic trim angle agrees within 0.5 deg. This is a satisfying result, given that trim measurements are not corrected for scale effects and that the CFD mesh could be even further refined around the hull to address more accurately this application, which was not necessary in this study.

Model tests


VS [knots]

FN [-]

CTS [-]

Trim [deg]

CTS [-] (%)

Trim [deg] (Delta)





8.70 (+0.7%)

0.64 (-0.10)





4.38 (-0.7%)

1.32 (-0.34)





3.64 (-0.5%)

1.75 (-0.52)


The excellent agreement between the model tests and CFD predictions for the total ship resistance in calm water condition results in a good confidence level in the CFD results presented to the End Client.

Meet the Team

End User – Eloïse Croonenborghs, Research Scientist at MARINTEK, Maritime division, Trondheim, Norway
Team Expert – Sverre Anders Alterskjær, Research Scientist at MARINTEK, Maritime division, Trondheim, Norway
End Client Expert – Canan TIRYAKI, STM, Ankara, Turkey
Software Provider – NUMECA International S.A.

Propeller Cavitation

Numerical Simulations of the Cavitating and Non-Cavitating Flow around the Postdam Propeller Test Case

Marine | Hydro

Numerous studies based on experiments or computations have been carried out to investigate propeller open water characteristics. Most studies only consider the case of a propeller in straight ahead flow. However, under real conditions, a working propeller operates behind a ship usually in a complex wake, so that the propeller shows quite different hydrodynamic performance.

In this paper, the cavitating performance and open water performance of the SMP’15 propeller are numerically simulated using the flow solver ISIS-CFD. A cavitation model based on a transport equation and the k-w SST turbulence model are coupled in the flow solver. The thrust and torque coefficients are presented for the open water case. The pressure distribution on the propeller blades is also presented. For the cavitating case, the cavity surface is presented as well as the thrust and torque coefficients.

Read all about it HERE.


Boom Supersonic partners with NUMECA


Boom Supersonic partners with NUMECA, adopting its CFD solutions to advance development of the Overture supersonic passenger aircraft

New partnership gives Boom ability to build a more efficient aircraft

NUMECA International, a global leader in Computational Fluid Dynamics (CFD), multiphysics, and optimization, today announced a new partnership with Boom Supersonic (Boom), an innovative company building history’s fastest supersonic airliner.

Through this partnership, Boom is aiming to create a dramatically streamlined and highly automated workflow, both utilizing NUMECA’s expertise in creating solutions to provide quality results with the highest reliability and the fastest solution time of any code and amplifying the strengths of Boom’s world-renowned design team. The new CFD solutions adopted from NUMECA will advance the development of Overture, Boom’s Mach-2.2 commercial airliner.

“In the first pilot project we attained results up to 14x faster than with our previous design environment,” said Tim Conners, Lead Propulsion Engineer at Boom. “This gives Boom the ability to test more conditions, try more design ideas, and save millions of dollars in compute resources – yielding a more efficient aircraft in less time and for lower cost than we originally planned for.”

Boom believes that speed is less about going really fast and more about all the things it enables you to do. It is this commitment to speed—and what it allows its engineers to achieve—that drove Boom’s decision to choose NUMECA for some of its most demanding CFD simulations.

“At NUMECA, our prime focus is helping our clients, such as Boom, develop their products under real world conditions by ensuring the highest reliability of performance prediction, at a fraction of the computational cost of competitors’ solutions,” said Prof Charles Hirsch, President of NUMECA International. “In partnering with Boom, we’re excited to contribute to bringing supersonic travel to a commercial audience and assist in the development of Overture.”

About Boom Supersonic

Boom Supersonic is the Denver-based company building supersonic airliners. Founded in 2014, its vision is to remove the barriers to experiencing the planet: time, money, and hassle. Boom is backed by world-class investors including Emerson Collective, Y Combinator, Caffeinated Capital, SV Angel, and individuals including Sam Altman, Paul Graham, Ron Conway, Michael Marks, and Greg McAdoo. The company has announced pre-orders of its Overture airliner from airlines including Japan Airlines and Virgin Group. For more information, please visit

Full Engine Simulation

Fully-coupled CFD Engine Simulations

Aerospace | Automotive | Turbomachinery | HVAC

The aerospace industry, like many other industries, is under pressure to drastically reduce its environmental footprint. The Flightpath 2050 goals of the European Union state that by the year 2050 all CO2 emissions per passenger kilometer must be reduced by 75%, NOx by 90% and noise pollution by 65% (relative to the year 2000). [1] One of the ways to achieve these environmental goals is by increasing turbomachinery performance. Improved analysis in the design phase, more specifically development of more reliable predictions, advancement in accuracy, inter-disciplinarity and speed of simulation tools, can add several percentage points to engine efficiency and reduce development cost and time. [2][3]


One of the main challenges in designing an engine is the complexity in terms of geometrical details (combustion chamber features, turbine cooling holes…) and of interaction effects between the components which must be modelled with accuracy and acceptable computation time.

Full Engine Simulation Methodology

Traditionally engine design in industry has relied on tools like experimental investigation using test or flow bench set-ups, analytical models, empirical/historical data, 1D/2D codes and recently, high fidelity 3D computational fluid dynamics (CFD) for steady and unsteady flow physics modelling. Currently most of the literature work [4-6] and projects conducted at an industrial scale employ a component-by-component analysis approach, where each engine component is studied separately. Such an approach usually requires assumptions for inlet and outlet boundary conditions for each component and involves a considerable effort in coupling different component analysis tools (often with different modelling degrees of accuracy), leading to a process which is potentially error-prone from the simulation setup point of view and often resulting in significant mismatches between numerical and experimental data.

A one-way coupling approach represents one step further in increasing the accuracy of such simulations. It can be achieved, for example, by extracting outlet profiles of the flow variables from individual converged component simulations and applying them as inlet boundary condition profiles to downstream component runs. Nevertheless, the inter-component interaction is still one-way and the simulation process and results can suffer from similar drawbacks as in the totally uncoupled workflow.In the two-way coupling methodology all the components are coupled and solved simultaneously in one single simulation. This approach greatly simplifies and accelerates the simulation workflow. Since all the components are considered simultaneously, there is no need to prescribe boundary conditions between the various elements of the aero-engine. This avoids running simulations where the states at the interface between the different components have to be guessed.

The full engine CFD numerical modeling methodology can be mainly categorized in three levels:

  1. Steady-state RANS simulations: computationally low cost and usually involving a single meshed blade passage per turbomachinery row with mixing-plane interfaces between components
  2. Unsteady-state RANS time domain simulations: computationally expensive, employing several meshed blade passages per turbomachinery row, usually requiring many time steps to reach periodic flow conditions and providing solutions for a single clocking configuration per run
  3. Unsteady-state RANS frequency domain simulations: 2-3 orders of magnitude faster than time domain method computations, employing a single meshed blade passage per turbomachinery row, providing improved rotor-stator interfaces modeling and allowing arbitrarily clocked solution reconstruction in time.

NUMECA’s Approach to Full Engine CFD Simulation

A full 3D aerodynamic simulation of a complete gas turbine engine, applied to a micro turbine case has been conducted at NUMECA. The analysis was comprised of a single fully-coupled 3D CFD simulation for the flow of a KJ66 engine redesign. The injection and burning of fuel inside the combustion chamber are modeled with a simplified flamelet model. Using advanced RANS treatment with inputs from Nonlinear Harmonic (NLH) method (available as module for FINE™/Turbo), tangential non-uniformities are captured and the flow physics of the interaction between compressor, combustor and turbine are assessed.

Case Description

The selected test case is a redesign version of the KJ66 micro gas turbine (Figure 2). The Figure 2 shows the layout of the redesigned version of the KJ66 micro gas turbine used in the full engine computation. The centrifugal compressor is mounted at the engine entrance and it is composed of an impeller and a bladed diffuser row. The combustion chamber is followed by a high-pressure turbine (HPT), which drives the compressor, and a low pressure turbine (LPT), which would drive a propeller in an independent shaft. An exhaust hood is connected to the last LPT row at the engine exit.

Simulation Setup

The computational domain encompassed one blade passage for each turbomachinery blade row, a 60° sector for the combustion chamber (containing one fuel injector) and half exhaust hood. The three-dimensional mesh for the blade rows (compressor and turbine rows) was generated automatically using Autogrid5™, NUMECA’s turbomachinery dedicated full automatic hexahedral block-structured grid generator. The mesh for the combustion chamber and exhaust hood was generated with HEXPRESS/Hybrid™, NUMECA’s unstructured hex-dominant conformal body-fitted mesher for arbitrary complex geometries. The entire mesh has 19.20 million points.

As a first investigation, the steady RANS computation is performed with the Spalart-Allmaras turbulence model. Ambient total quantities are imposed at the engine inlet with specified axial velocity direction and static pressure is fixed at the outlet. The fuel injection is specified with static temperature and axial velocity of -120 m/s. Solid walls are assumed smooth and adiabatic. The convergence history is checked by following the evolution of the mass flow, the pressure ratio, and net torque between the HPT rotor and the impeller. The computation is launched in parallel on 144 processors on a computing cluster. In a second step, the computation is restarted with an improved rotor-stator connection based on the NLH method.

Combustion Model

The injection and burning of fuel inside the combustion chamber are modeled with a simplified flamelet model implemented in OpenLabs™. With this model, an additional equation is solved for the mixture fraction f, with the flame temperature as function of the composition.



Steady-state Computation

The convergence history of mass flow rate error between the engine inlet and outlet drops to less than 0.2% after 25000 iterations in approximately 24 hours of wall clock time. The net torque Mz reaches a final value of +0.21 N.m when the couple impeller-HPT rotor spins at -80000 rpm, meaning that the turbine produces enough torque to drive the compressor and both components are very close to be load balanced.

Figure 4 shows the static pressure, static temperature and absolute Mach number distributions at midspan of the compressor and turbine as well as in the solid walls of the combustion chamber and exhaust hood. The flow fields are continuous across the machine with a gradual raise of Mach number through the impeller and subsequent conversion of the kinetic energy into pressure across the diffuser. At approximately 160 kPa, air reaches the combustion chamber where the simulated combustion process takes place with a relatively small pressure loss. The maximum temperature at the combustion chamber is around 2200K at the combustor inner chamber. The hot gases from the combustion enter the HPT at approximately 983K and are expanded through the downstream blade rows, exiting the machine at 932K.

As shown in the Figures 5, the mixture fraction color contour successfully depicts the fuel stream entering the combustion chamber with f = and the combustion gases gradually reaching a value of 0.02 at the component exit. The effect of the holes in the inner chamber walls can be noticed in the magnitude of velocity color contour: they provide a flow of compressed air, acting as oxidizer for the combustion process, to mix with vaporized fuel and achieve ignition.

NLH Computation

The results from the RANS computation using a mixing plane treatment at the rotor stator interface can be compared to the results of the NLH analysis. In particular, it is interesting to note the effect of the improved connection approach with respect to the inter-components interactions.
Figures 6 and 7 show similar results on the mass flux for both simulations at the exit of the combustion chamber. A pattern linked to the periodicity of the fuel pipes and to the combustion zones can be observed. The transfer of information from the combustor outlet to the adjacent downstream HPT nozzle is different for the RANS simulation and its NLH counterpart. The mixing plane approach shows no tangential non-uniformities, while the NLH rotor-stator treatment proves to be a low cost first step in capturing tangential non-uniformities.

The Figure 8 shows the results for absolute total enthalpy and mixture fraction at the HPT and LPT inlet connections. The azimuthal averaging used in the mixing plane approach, where tangential non-uniformities are not perceived by downstream components, can be noticed for the RANS simulation. In contrast, the Fourier decomposition together with the local non-reflective boundary treatment used in the NLH method is able to show a slight improvement at the interfaces.


A successful simulation of the three-dimensional flow of a complete micro gas turbine engine has been achieved, on the basis of a fully-coupled 3D CFD simulation of a redesigned version of the KJ66 micro gas turbine. Compared to the component-by-component analysis, the fully coupled approach enables the solution of the whole engine in one single simulation. Furthermore it simplifies the simulation workflow as only the engine inlet and outlet pressures, the rotating speed and the fuel mass flow rate need to be prescribed.


Source: Mateus Teixeira, Luigi Romagnosi, Mohamed Mezine, Yannick Baux, Jan Anker, Kilian Claramunt, Charles Hirsch A Methodology for Fully-Coupled CFD Engine Simulations, Applied to a Micro Gas Turbine Engine. ASME. Turbo Expo: Power for Land, Sea and Air, Volume 2C: Turbomachinery ():V02CT42A047. doi:10.1115/GT2018-76870.

[1] Flightpath 2050: Europe’s Vision for Aviation, 2011,
[2] National Academies of Sciences, Engineering, and Medicine, 2016, “Commercial Aircraft Propulsion and Energy Systems Research: Reducing Global Carbon Emissions”, Washington, DC: The National Academies Press
[3] Mavriplis D., Darmofal D., Keyes D., Turner M., 2007, “Petaflops opportunities for the NASA Fundamental Aeronautics Program”, 18th AIAA Computational Fluid Dynamics Conf., Miami, FL, AIAA paper 2007–4084.
[4] Xiang J., Schluter J. U., Duan F., 2016, “Study of KJ-66 Micro Gas Turbine Compressor: Steady and Unsteady Reynolds-Averaged Navier–Stokes Approach”, Proceedings of the Institution of Mechanical Engineers Part G Journal of Aerospace Engineering.
[5] Gonzalez C.A., Wong K.C., Armfield S., 2008, “Computational study of a micro-turbine engine combustor using large eddy simulation and Reynolds averaged turbulence models”, ANZIAM J. 49 (EMAC2007) pp.C407–C422, C407.
[6] Turner M., 2000, “Full 3D Analysis of the GE90 Turbofan Primary Flowpath”, NASA/CR—2000-209951.

Aerodynamic Prediction

NUMECA contributes to JAXA APC-III Workshop using FINE™/Open for Aerodynamic Prediction

Aerospace | Grid Generation | Multi-Purpose

Born through the merger of three previously independent organizations on 1st October 2003, the Japan Aerospace Exploration Agency (JAXA) is the organism responsible for aviation research and technology development. In addition, Its scope comprises space and planetary study, rocket development and launch of satellites into orbit development.


Aircraft design development was originally based on experimental techniques. Technical progress in computing power and extensive research in fields such as numerical analysis or turbulence modelling has lead to complementary use of both wind tunnel testing (WTT) and Computational Fluid Dynamics (CFD), with CFD codes utilization now widespread due to continuous increase of accuracy and reliability.

Aerodynamic Prediction Challenge (APC) is a series of workshops organised by the JAXA with the aim of tracking the progress of CFD tools with respect to challenges faced in aeronautical applications. The third edition was held in Japan June 28th, 2017. NUMECA took part in APC-III as contributor to Task 1, with NASA Common Research Model (CRM) aerodynamic prediction at cruise state and high angle of attack (presence of tail wings, reflected deformation measurement data).

Task Description

The model under investigation is a 80% scaled copy of CRM at high speed (cruise state). CRM was developed by NASA to build experimental databases for the purpose of validating specific applications of CFD, with focus on the aerodynamic design of the wing. The geometry includes horizontal tail plane (HTP) at zero setting angle, which is same configuration used for NASA Drag Prediction Workshop 4 (DPW4).

A wind tunnel test (WTT) of the 80% scaled copy of the CRM was performed in the 2m × 2m transonic wind tunnel of JAXA and is therefore used as benchmark to assess numerical results provided by participants. Flow conditions correspond to Mach number M=0.847 and Reynolds number (based on wing mean aerodynamic chord) Remac=2.26 E+06.

Emphasis in the analysis of results is given to high angle of attack, particularly in the wing-fuselage junction. The prediction of detachment in flows parallel to corners formed by intersecting walls and exposed to adverse pressure gradient remains one of the main challenges of current turbulence models widely used in industry.

Deliverables requested by the organizing committee for this task included Aerodynamic coefficients (CD, CL , Cm ) with decomposition into pressure and friction, breakdown by components (wing, fuselage, HTP), and pressure coefficient distribution along nine wing sections distributed spanwise.

CFD Simulations

Two sets of meshes were used, both built and provided by JAXA to all participants. On one hand one set made of structured hexahedral cells. On the other hand, another group of hybrid tetrahedral dominant grids. Cell count was 9.15 and 29.98 million respectively. The difference in cell count is due to additional refinement for the hybrid grid at the leading and trailing edges of lifting surfaces, notably in spanwise direction. In those regions and in the intersecting walls the hybrid mesh follows structured approach.

Aeroelastic effects are taken into account using one mesh for each of the incidence angle of the range under investigation (-1.79 deg. – 5.72 deg.), which accounted for wing deformation data measured during WTT. The near wall mesh is very fine, with non dimensional wall distance of first cell in the viscous sub-layer (y+ < 1).

NUMECA FINE™/Open 6.2 CFD solver (density based, finite volume discretization, cell centred) was used to solve Reynolds averaged Navier-Stokes (RANS) equations with full multigrid approach and different turbulent formulations:

Linear eddy viscosity models (LEVM) such as Spalart-Allmaras SA-fv3, Menter SST-2003 and K-Epsilon KE-YS-1993. Experience shows that these approaches tends to predict separation too early in wing-body junctions where flow faces adverse pressure gradients.

Explicit Algebraic Reynolds Stress Models (EARSM) with a non-linear constitutive relationship between Reynolds stresses and mean strain rate, such as SBSL-EARSM and SSC-EARSM, the latter developed by NUMECA with the aim of better predicting separated flows.

Finally, the study aims at evaluating the impact of numerical artificial dissipation introduced to governing flow equations. Both scalar and matrix dissipation algorithms were used.


Several turbulence models and scalar/matrix dissipation approaches were tested on structured and unstructured grids. Globally, the transition from linear to nonlinear region is well captured with FINE™/Open. Beyond this point, the turbulence models and the dissipation models have a larger impact on the solution, which is especially noticeable in the pitching moment and its slope (which drives longitudinal stability). SSC-EARSM provides the best results. Analysis of flow over wing suction side explains those differences.

Large Side-of-Body (SOB) separation bubble is predicted by SA-fv3 and SST-2003 models at high incidence. This is not in line with experimental data and leads to different flow patterns for the whole span and lower pitching moment. SOB is not observed with KE-YS-1993. However, shock wave location appears significantly downstream. Therefore, lift is clearly overestimated. SBSL-EARSM and SSC-EARSM model improve the predicted inboard flow and the location of shock in this region.

Pressure coefficient cuts show that SSC-EARSM model, developed by NUMECA specifically for separated flows prediction, provides the best agreement with experimental data, independently of the mesh topology and numerical dissipation scheme.


The excellent agreement between the model tests and CFD predictions for the pressure distribution over the inner suction side of wing provides a good confidence level in the CFD results presented to the APC-III workshop. Moreover, NUMECA SSC-EARSM results are encouraging, with significantly improved results with respect to traditional LEVM and with affordable computational cost.

Gust Modelling

The AeroGust Project: Aeroelastic Gust Modelling

Aerospace | Wind Energy | Multi Purpose

A key element in the design of an aircraft is to make sure it can cope with the stresses arising from the impact of strong winds on the plane’s structure. Calculating the way the aircraft reacts to gusts and turbulences is essential knowledge before a plane can be built.

Currently most of the data concerning gusts is gathered during expensive wind tunnel experiments and rather late in the design process, when design options have already been narrowed down a lot. Furthermore there are very few wind tunnels in the world that are capable of accurately reproducing the conditions for a full size aircraft in flight.

The more accurate the design of the aircraft models before going to testing, the less new models need to be made for the wind tunnel. Again this does not only save a considerable amount of money and time, it also allows for entirely new kinds of design to be explored, including the use of different, more flexible materials.

That is why the AeroGust (Aeroelastic Gust Modelling) Project, funded by the European Union, has been set up. Its goal is to achieve better understanding of the aircraft/gust interactions early in the design process, by crafting new efficiencies and higher accuracy in the simulation stage.

Numeca International is proud to be one of the partners in the project, providing expert knowledge and state-of-the-art tools for Computational Fluid Dynamics (CFD) simulation. CFD is crucial in accurately predicting aerodynamic performance, in this project producing precise and detailed models of aerodynamic gust flows around the aircraft structure.

NUMECA’s FINE™/Open was used to model how the airflow interacts with the wing of an aircraft as shown in the figures on the side.

Using this virtual prototype an unlimited amount of variations in any part of the design can be tested by simply varying input, boundary, geometrical parameters etc. The objective of the AeroGust project is to generate new computer codes that will streamline the processes of the simulations and tests.

Besides aircraft design, the AeroGust project can also benefit the industry of wind turbine design: Today the distribution of wind farms is restricted by the fact that strong wind variances and gusts create large loads on the turbines. If we would be able to predict the impact of those gusts more accurately, the structures could also be placed in more challenging regions like for example the Arctic Circle and the tropics.

For more detailed information, and if you want to keep up with the latest news and events regarding the project:

Propeller Optimisation

Pipistrel reduces by 6% the Energy Consumption of an Electrical Aircraft by Optimising its Propeller with FINE™/Turbo

Aerospace | Turbomachinery | Grid Generation

Improve the energy consumption of an electric aircraft through energy recuperation.

Pipistrel used the propeller as an airborne wind turbine, by transforming the energy created by the descension of an aircraft into electric energy and storing it in a battery. The performance of the propeller design was numerically computed with FINE™/Turbo.

The aircraft consumes 6% less energy during the climb.
Net energy consumption during ascent/descent manoeuvres decreased by 19%.
A 27% increase in number of traffic pattern circuits was achieved.


The first electric aircraft was created by simply replacing the piston engine system of a normal combustion engine aircraft by an electric propulsion unit. Although environmentally friendlier, this was not an optimal design yet for an electrical aircraft.

Since the density of current state-of-the-art battery energy is still much lower than gasoline energy density, a need for optimal energy use in an electric aircraft is crucial.

Energy Recuperation as a Propeller Design Strategy

One possibility to improve the energy balance of an electric aircraft, is using its propeller as an airborne wind turbine, where the energy of a descending aircraft is recuperated into electric energy and stored in the battery.

In the case described in this article, Pipistrel designed a propeller specially adapted for exploiting in-flight power recuperation this way.

The objective was to improve the energy consumption of the Alpha Electro, Pipistrel’s electric trainer aircraft.

Climb is the most energy consuming part of the traffic pattern. A new propeller EA-002 (Figure 1) was designed to exploit the possibility of energy recuperation while preserving good performance in the climb flight phase.

Optimisation of the Propulsion System

For the optimisation of the propulsion system, a 3-way approach was performed:

  • The Aerodynamic approach: focused on the optimisation of the airfoil shapes and the chord/twist blade distribution. The performance of the propeller design was verified through CFD simulations, ran with NUMECA software AutoGrid5™ and FINE™/Turbo. (The velocity distribution over a cross section of one of the simulations is presented on the right-hand side of Figure 2)
  • The Electrical approach: includes a hardware enabled bi-directional energy flow, with the possibility to adjust the torque/angular velocity combination for maximum power recuperation at specific descent rates.
  • The Strategic approach: focused on adjusting ascent and descent rates to minimise energy consumption and maximise energy recuperation respectively.


The performance of the propulsion units was evaluated by comparing and testing 3 propellers with the following two testing methodologies:

(1) net energy consumption within 1000 ft ascent & descent manoeuvres and (2) the number of traffic pattern circuits performed with one fully charged battery.

The 3 propellers (Figure 3):

  • The AS-D propeller was primarily designed for the piston engine version of the Alpha Electro, i.e. Alpha Trainer.
  • The EA-001 propeller was developed for the Alpha Electro power train and optimised for climb and cruise phases.
  • The EA-002 propeller was also made for the Alpha Electro power train, but optimised for climb and recuperation phases.

Climb & descent manoeuvres were tested according to these parameters:

  • Climb phase manoeuvre: constant climb at 76 kts IAS for 1000 ft at 45 kW power – monitoring energy consumption, vertical speed and time to ascend.
  • Recuperation manoeuvre: throttle idle, stable descent for 1000 ft at different speed rates – monitoring recuperated energy and descent rate.

Testing for the traffic pattern circuit with one fully charged battery included manoeuvres prescribed in the Alpha Electro Pilot Operating Handbook (POH) as follows:

  • Climb/departure, crosswind and downwind cycle phases:
  • apply take-off flaps below 60 kts indicated air speed (IAS)
  • climb from 0 to 500 ft AGL by
  • applying full power (65 kW) for first 10 s of the climb
  • applying 70% (45 kW) power and climb at 76 kts IAS
  • at 500 ft AGL apply 30% (20 kW) power and maintain 500 ft AGL
  • Base, final, touch-and-go cycle phases:
  • descend from 500 ft to 0 ft AGL by reducing power to idle
  • apply landing flaps below 60 kts IAS
  • descent at 50 kts IAS to the ground

Performance comparison

The results from the climb & descent manoeuvre tests are presented in Table 1.

The AS-D and EA-001 have similar climb performance and recuperation capability. With the EA-002 installed, the aircraft consumes 6% less energy during the climb compared to the AS-D propeller and is able to recuperate 0.15kWh at the descent speed of 70kts. The propeller optimised for recuperation (EA-002) showed a 19% reduction in net energy consumption compared to the EA-001.

As can be seen from Table 2, the test pilot was able to carry out almost the same number of traffic patterns with the AS-D and the EA-001 propellers, and 27% more (compared to AS-D) with the EA-002 propeller. This resulted in better climb efficiency and higher energy recuperation of the EA-002 at a relatively low approach speed of 50 kts, which was kept constant for all three propellers.

Experimental vs CFD Simulations

A comparison was performed between CFD simulation results and flight test measurements of the EA-002 propeller. Two parameters were compared: the thrust and the propeller power coefficient.

Figure 6 and Figure 7 depict values of thrust and power coefficients versus advance ratio of the EA-002 propeller respectively. As seen in the figures, the design predictions of NUMECA’s CFD simulation results and the flight test measurements of the propeller are in good agreement.


Due to the reduction of the energy consumption, the aircraft can stay airborne longer and/or smaller battery packs may be installed for a specific flight time.

The EA-002 is set to become the first European Aviation Safety Agency certified propeller with recuperation capability for electric propulsion.


D. Eržen, M. Andrejašic, R. Lapuh, J. Tomažic and C. Gorup – Pipistrel Vertical Solutions d.o.o., Slovenia T. Kosel – Faculty of Mechanical Engineering, Slovenia

Full Journal Paper

D. Eržen, M. Andrejasic, R. Lapuh, J. Tomazic, C. Gorup, and T. Kosel, “An Optimal Propeller Design for In-Flight Power Recuperation on an Electric Aircraft”, 2018 Aviation Technology, Integration, and Operations Conference, AIAA AVIATION Forum, (AIAA 2018-3206)

Space Launch Innovation

Space Launch Innovation at Masten Space Systems with NUMECA Software

Aerospace | Grid Generation | Multi-Purpose

Until very recently, rockets that launched satellites into orbit were completely discarded after a single use -and this is still commonly done for most launches. Within the past year, first generation reusable satellite launch has been demonstrated, and the business case for reusable launch has been made.

As a prime contractor for the Defense Advanced Research Projects Agency (DARPA) Experimental Spaceplane (XS-1) program, Masten Space Systems is developing a launch vehicle with design innovations focused on next-generation reusability. DARPA’s XS-1 program aims to reduce the time to space and cost to space by orders of magnitude by achieving aircraft-like reusability and flight rate [1]. Jess Sponable, DARPA XS-1 program manager, has stated: “We think flying ten times in 10 days is something well beyond the capability of either SpaceX or Blue Origin at this time.” [2].

“Masten Space Systems heavily utilised the FINE Suite at HPC scale to design our next-generation reusable satellite launch system. We developed an aerodynamic configuration that hasn’t been done before for launch or reentry. First wind tunnel tests confirmed critical aspects of the design, and predictions compared well with measurements.”

Allan Grosvenor,
Aerodynamics Lead
Masten Space Systems

Requirements include cost per flight of less than $5 million USD transporting 3,000 lbs to a 100 nmi reference orbit inclined 90 degrees. Masten engineers needed to optimise the booster design over a wide range of flight conditions, including ascent from sea level to hypersonic speeds at the upper-stage-separation point, and then reentry and return flight.

NUMECA-USA has worked closely with Masten Aerodynamics Lead, Allan Grosvenor, and his team to develop an HPC-driven workflow that enables concept development, evolution, and optimisation, of vehicle configuration that takes into consideration aerodynamic performance, control, loads and aerothermal heating, with simultaneous trajectory optimisation.

A study of the HIFiRE-1 high speed transition experiment was one of several test cases used to validate NUMECA solver predictions (Figure 1). The HIFiRE-1 experiment (Wadhams, et. al. 2008) is a Cone-Cylinder-Flare configuration, exposed to Mach 7.2 flow (Re 1E7) conducted at the CUBRC LENS hypersonic tunnel. The numerical results are shown to capture relevant physics and to predict pressures and heat fluxes, and particularly the peak heating which is critical to thermal protection system design.

To illustrate the optimisation process we created a fictitious high speed flight vehicle test case inspired by the Colonial Viperfighter craft of the TV series Battlestar Galactica.

The ship design is controlled parametrically with NUMECA’s AutoBlade™ modeler. A set of parameters control the canopy’s length and height, the wing and vertical stabiliser angle and sweepback, and the nose radius and droop (Figure 2). Extensive sets of parameters are available for more detailed 3D shaping and variation of all portions of fuselage, lifting and control surfaces, etc.

Design necessarily required consideration of entire launch and return trajectories. Large numbers of organised studies conducted by Masten Space Systems included running sweeps through several critical flight conditions including variation of flight angles (e.g., pitch, yaw, roll) and aerodynamic control surface angles. The example shown here illustrates a subset of these studies focused on reentry.

DoDHPCMP supercomputers [3] were leveraged to conduct the extensive studies, and the Computational Pipeline (Figure 3) indicates the organised workflow that was executed to run the sophisticated design evolution process. Masten Space Systems found that only with NUMECA could they systematically produce high-quality grids (flow domain discretisation) and high accuracy solutions.

The meshing task was automated with HEXPRESS™/Hybrid, which is the new generation multi-core meshing tool of NUMECA. It has been designed to generate meshes of complex geometries, regardless of the CAD format and quality (Figure 4).

This tool produces full hexahedral and hex-dominant meshes completely in parallel and in batch mode, which has been the perfect match for Masten’s challenge. Masten Space Systems chose to utilise the option producing pure hex meshes.

The CFD simulations were all executed with NUMECA’s FINE™/Open CFD solver, and then post-processed through scripts running all in batch in a fully automatic way. Data-mining of the full 3D solutions feeds the optimisation process.

The following animations are comparisons between the baseline vehicle and the new design for reentry performance and stability.

MASTEN is lowering the barriers to space access. Their mission contributes to a shared strategic goal of extending human presence across the solar system. Their approach: reusability. Their technology development focuses at the core on entry, descent, and landing technologies (EDL) to ensure precise and safe landings on planets and other celestial bodies. Learn more


Driven by creativity, innovation, and quality, we develop software toolsets that support the world’s leading industries” Learn more

[1] DARPA pushing new effort with Experimental Spaceplane, XS-1 (
[2] DARPA experimental spaceplane program moves into next phase (
[3] Supercomputers Lower the Cost of Space Access (
[4] HIFiRE-1 Mach 7 aerothermal heating prediction (

Figure 2: Parametric design variations controlled with AutoBlade™

Figure 6: Reentry Matrix Results Comparison: Baseline (left) vs Modified (right) designs

Multi-stage Compressor Optimisation

Design Optimisation of a Multi-stage Compressor

Turbomachinery | HVAC | Automotive | Aerospace

Research-and-production Company “ENTECHMACH”, located in St. Petersburg, Russia, has been operating in the power engineering area for more than 25 years. Its main activities are design and production of centrifugal and axial compressors, steam turbines, multipliers, heat exchangers with air and water cooling systems.

Since 2 years now the company has been using the FINE™/Turbo package for the flow simulation and optimisation of stationary industrial centrifugal compressors and for research engineering. One of those developments concerns the modernisation project of a three-stage air centrifugal compressor in catalytic cracking technology.

Problem Analysis

The existing compressor suffered a number of drawbacks: low reliability of the axi-radial impellers as their covering discs were causing large stress concentration zones, failing stators due to unsuitable material choices and a number of performance issues. The performance and discharge pressure were insufficient, the surge margin was too low and the power consumption too high.

As a first step we used NUMECA’s FINE™/Turbo solver to accurately calculate a complete compressor performance map of the compressor. The result allowed us to clearly identify the main causes of these efficiency losses and low surge margin. Conclusion: The modernised compressor would need a complete replacement of the rotor and stator elements in the compressor case, bearings and lubrication oil system.


It was decided to design the new compressor with semi-opened axi-radial impellers, as they create two times less stresses compared to the closed axi-radial version with covering discs.

To make sure the newly developed impellers meet the highest possible levels of efficiency, pressure ratio and surge margin, all flow path elements were optimised with NUMECA’s FINE™/Design3D solver including the impellers, the vane diffusers and the return channels for several operating conditions.

With the use of the NLH-method, the influence of the inlet chamber and 360° inlet guide vanes on first impeller parameters was investigated, as well as the influence of the outlet chamber on the last diffuser blade row. Combined vane diffusers, which merge a long vaneless part and a low blade density part, were implemented in all stages. The numerical investigations confirmed the advantages of this solution: a wide operating range and reduction of the losses in the stator elements (as with vaneless diffusers), but at the same time providing optimal flow conditions at the inlet of the return channels, which leads to an efficiency increase (as with vane diffusers).

The main challenge to solve for the return channel optimisation was the axial flow direction at the inlet of the next impeller. Traditional return channels with 2D profiles that we analysed, could not provide the axial flow direction on all channel spans. There were highly non-uniform flows on the hub and shroud as shown in Figure 3 on the left side.

To solve this problem, we developed a non-classic return channel consisting of two parts: a 2D main profile part and a 3D outlet part. The main part was calculated in a way that would achieve the largest possible flow irregularity decrease and the 3D outlet part was designed to have an effective flow alignment on all spans. This solution suppressed the secondary flows in the return channels considerably. In Figure 3 on the right side, where 90° equals axial flow, the irregularities after the modernisation can be seen. The redesigning of the return channels eventually resulted in an increase of the operating range of the second and third impellers of more than 15% thanks to the improved impeller inflow conditions.

The new compressor performance map was calculated again for different operating conditions i.e. different angles of incidence of the Inlet Guide Vane (IGV) and inlet temperature. The results are shown in the Figure 4.


The full mesh model of the compressor was generated with IGG™ and AutoGrid5™. It included about 40 million elements and consisted of all blade rows, gaps between impellers and stator, outlet chamber and all labyrinth seal regions. The Spalart-Allmaras turbulence model for simulations was applied to verify the correspondence of mesh criterion y+ with a recommended range. Thanks to the high quality of the mesh we could apply the CPU Booster™ for our calculations, which saved us a lot of time and computing resources.


Industrial tests of the modernised compressor showed an increase of polytropic efficiency on normal mode of ~6.5% abs. and a corresponding decrease in power consumption of the same percentage. The new compressor reaches a higher pressure ratio of 4.5, compared to 4.1 before the modernisation. Furthermore the surge margin related to the normal mode significantly increased from ~6% before to 50% after. Power consumption on minimal performance is 1.5 times lower (~2.5MW).

The new, modernised compressor is in operation at client site and performing well and equally as important is proving reliable.


Vladimir Neverov, Ivan Cheglakov, Specialists on compressor machines
Aleksandr Liubimov, Head of Advanced Development and Design Department

Steady Turbomachinery Simulations

Modeling the Rotor-Stator Interface in Steady Turbomachinery Simulations

Turbomachinery | HVAC | Automotive | Aerospace


As turbomachinery flows are some of the most complex flows found in engineering applications, predicting performance is quite challenging. Separation, high velocities, rotation, high load on the blades, small gaps… These phenomena are difficult to quantify accurately and expensive to capture through experimental campaigns. So it is no surprise that this industry has been one of the first to introduce Computational Fluid Dynamics (CFD) methods in order to lower costs without loss of accuracy.

Rotor-Stator Interface Methods

The goal of turbomachinery CFD simulations is to predict flow behavior in adjacent rotating and non-rotating blade rows. This means multiple frames of reference need to be used: “moving” for the rotating parts and “stationary” for the non-rotating. The link between these rotating and stationary frames of reference is called the Rotor-Stator interface. Turbomachinery CFD solvers include different treatment methods for this interface and it is crucial to select the appropriate one, as it has an impact on the entire flow: e.g. prediction of local Mach number or prediction of global quantities like efficiency or pressure ratio.

The choice of treatment method firstly depends on the nature of the simulation: steady, transient or harmonic. In this article we are going to focus only on steady simulations, since these are the most used in a design and production level in industrial environments.

In order to pass flow information through the rotor-stator interface, a circumferential averaging process has to be performed on it. This process is known as the mixing plane approach. The rotor-stator interaction is done by exchanging circumferentially averaged flow quantities. Basically this means that the blade wake or separation phenomena occurring during the blade passage, are mixed circumferentially before entering the downstream component. As a result, velocity components and pressure are uniform in the circumferential direction. This physical approximation tends to become more acceptable as rotational speed is increasing. The mixing plane technique is by far the most used rotor/stator modeling approach in the turbomachinery industry.

To showcase the different treatment methods of this mixing plane approach, we used a 1-½ stage transonic axial compressor.

In Table 1 you can see the results of the different methods described in this table, as they appear in FINE™/Turbo.

Figure 1 shows the absolute Mach number contour, upstream of the interface between the rotor and stator.
The three top images show similar results: Differences are very subtle and essentially portray the implementation of the same approach (as explained in Table 1). The images corresponding the 1D and 2D Non Reflecting conditions however are different, especially in the wake and near the hub, showing less diffusion in these areas.

Similar observations can be made downstream of the interface (although the differences are harder to spot in Figure 2 due to the mixing).
The difference here is more pronounced in the middle of the domain, where the 1D and 2D Non-Reflecting conditions show the effect of the upstream wake on the mixing downstream.


In general, it is preferable to start with the Full Non Matching Mixing Plane or the Conservative Coupling by Pitch-wise Rowsapproach. These will most often provide a stable solution with good mass flow conservation. In case a shock is present or the solution is unstable, the 1D or 2D Non Reflecting approach could be used. However, when using this approach, attention should be paid to the mass flow conservation, as the non-reflecting approaches are not conservative by nature. Finally, the Local Conservative Coupling is only recommended for an impeller-volute case.

Figure 3 summarizes the appropriate methods to start the analysis with as a flowchart. Please note this chart is to serve as best practice rather than a definitive manual.


Accurate Impeller-volute Interaction

Accurate Impeller-volute Interaction in Turbochargers

Turbomachinery | Automotive | HVAC | Aerospace

Turbochargers are currently ubiquitous in automobiles. It is nearly impossible to find a new Diesel vehicle without turbocharging and direct injection. Gasoline cars have been also following suit for the past few years. Regulations are pushing for lower emissions and the way to achieve this has been downsizing and using turbocharging to offset the lower power density of a smaller motor.

The turbocharger, however, adds a new degree of complexity to the system. Besides its own standalone performance, a turbocharger has to perform well with the engine. A component that is crucial in the high performance of the turbocharger inside the engine system is the volute. The volute is the component that connects the turbocharger with the inlet manifold. Often, the volute is redesigned for different engines and normally there is design freedom that can lead to minimizing the pressure losses that it introduces.

It is very important to capture appropriately the interaction between the turbomachinery components (impeller and diffuser) and the volute so as to be able to predict and minimize the pressure losses. Computational Fluid Dynamics (CFD) is frequently used for simulating this behaviour.

While it is possible to obtain an accurate prediction of the pressure loss by using a steady simulation, impeller-volute interaction can cause instabilities that can impact the performance of the turbocharger. For this, an unsteady analysis is used which is quite more costly and rarely happens in the design phase but rather in the validation phase of the design cycle.

To avoid running a costly analysis and for the designer to be able to incorporate this analysis in the design phase, frequency domain simulations like the Non Linear Harmonic (NLH) Method currently implemented in the NUMECA tools are employed. NLH solves the fluid equations in the frequency domain, only needs to simulate one blade passage instead of the full wheel and the equations are solved in a steady manner thus enabling the use of convergence acceleration techniques. It allows for an unsteady solution – including the transport of information through the impeller-volute interface – at a cost similar to a steady one.

The animation shows how the pressure waves propagate from the impeller towards the volute as well as their reflection back to the impeller coming from the tongue of the volute (the area where the outlet duct meets the scroll), providing important information about how the impeller-volute interaction and interaction with the tongue.

Capturing this interaction is quite important. The reflected waves can have negative impact on the performance of the system as they can cause instabilities to the impeller such as rotating stall, increased losses and increased forcing on the blades. The phenomenon cannot be captured by a steady (mixing plane) approach due to the averaging of the flow quantities in the impeller-volute interface or by a frozen-rotor approach. Hence a costly, unsteady simulation or a more cost-efficient harmonic (NLH) simulation is the way to capture this behaviour.

This analysis can be used in any configuration, centrifugal or not, and can provide insight about the interaction between turbomachinery components and devices upstream or downstream early in the design phase while keeping the simulation cost low.

Tonal Noise Analysis

Analysis of Tonal Noise Emissions (Hochschule Trier)

Turbomachinery | HVAC | Automotive | Aerospace

NUMECA’s software tools were applied to resolve the unsteady flow field in a centrifugal compressor and to conduct an analysis of the tonal noise emissions in the surroundings of the machine. The compressor consists of four rows with an inlet guide vane, an impeller, a vaned diffuser and a return channel. 

As a first step, the unsteady flow field was computed using NUMECA’s Nonlinear Harmonic Method (NLH) in multi-rank mode. That is, interactions between successive rows can be considered over several rotor-stator interfaces. Subsequently, the Turbomachinery Wizard of FINE™/Acoustics was applied. It is a powerful tool to efficiently predict tonal and broadband noise emissions for turbomachinery applications. With only a few clicks, the user can obtain the radiated sound power, as well as the sound pressure level at external microphones, using the Green’s function approach.  

The result for the case of Hochschule Trier is given in Figure 1, showing the estimated sound pressure level (normalized with the maximum value) at external microphones. These are located on a circular arc with a radius of r = 3 m and for a polar angle between 11.25° and 90° relative to the rotation axes of the compressor. The intensity of the sound pressure level at external locations mainly depends (aside from their spatial location) on the amplitudes of the harmonic modes and the propagation directivity to the far-field. In this case, the FINE™/Acoustics wizard predicts the highest sound emissions at an angle between 30° and 45° relative to the rotation axes of the compressor.  

A first estimate of the tonal noise emissions can also be obtained with a brief post-processing in CFView. Figure 2 shows the unsteady pressure perturbation (top left) along with the resulting spectrum (top right) for two points in the centrifugal compressor – one in the vaned diffuser and one in the return channel close to the outlet. The time signal was reconstructed from the NLH-solution and contains only contributions from the first three blade passing frequencies. Since the impeller is responsible for periodic oscillations in the compressor, the amplitudes are significantly higher in its vicinity than compared to the region close to the outlet. Subsequently, the resulting sound pressure level can be calculated from the amplitudes in the spectrum. 

In summary, NUMECA’s simulation tools provide the means for an efficient analysis of tonal noise emissions in turbomachines. Both, the post-processor CFView™ and NUMECA’s FINE™/Acoustics can be applied to conduct a detailed analysis of the prevailing unsteady pressures and subsequently the resulting sound propagation. The next step will be to modify the compressor stage in order to reduce the overall noise production.  


Prediction of the unsteady Flow

Accurate and reliable Prediction of the unsteady Flow in a Radial Turbine

Turbomachinery | Grid Generation | HVAC | Automotive | Aerospace

The Institute of Thermal Turbomachinery and Machinery Laboratory (ITSM) of the University of Stuttgart, based in Germany, focuses its research on gas turbines, steam turbines and turbochargers. As part of the FVV-funded project “Blade Forces”, the NUMECA software tools were utilized. The goal was to develop a workflow for the accurate, reliable and affordable prediction of forced response vibrations in radial turbines taking into account mistuning effects. A thorough validation of this workflow was conducted experimentally and numerically.


On the numerical side, high-fidelity CFD models comprising block-structured grids were created with the automatic mesh generator AutoGrid5™ including details like the blade tip gap and fillets as well as the rotor scallop and backspace. The turbine volute was also included in the model.

Nonlinear Harmonic Approach

The unsteady flow field in the turbine giving rise to the aerodynamic excitation was simulated by means of the innovative multi-rank Nonlinear Harmonic (NLH) frequency domain approach in FINE™/Turbo. It leads to a comparable accuracy of the numerical results as the state-of-the-art time-marching models at significantly reduced computation costs. A validation with unsteady pressure measurement data showed remarkably good agreement and increased our confidence in this numerical method and in NUMECA’s solver.

Aerodynamic Damping

The aerodynamic damping in the turbine was also predicted by means of the NLH method. For this purpose, the mode shapes of the investigated resonance crossings as obtained from a prior modal analysis were applied as an elastic blade vibration with defined vibration amplitude. By utilizing the NLH method, the aerodynamic damping was computed also with high accuracy and low computation time.

Given the excellent performance of NUMECA’s software tools, we look forward to employ them in other projects and to continue our cooperation with NUMECA as an academic partner.


Nikola Kovachev, Research Engineer
Tobias Müller, Research Engineer
Prof. Tekn. Dr. Damian Vogt, Institute Director, Institute of Thermal Turbomachinery and Machinery Laboratory (ITSM), University of Stuttgart

Transient Flow

Transient Flow in a Multi-Piston Pump

Pumps | Hydro

The Grenoble Hydraulic Machinery Research and Testing Centre is a laboratory of the Grenoble Institute of Technology, investigating hydraulic applications such as hydro-power, liquid propulsion and pumping.

Scope of the project is to design a swashplate axial multi-piston pump without check valves. The pump consists of two parts: A stator, composed of two conducts to distribute the flow and a rotor, moving nine pistons in a barrel.

Main target of the study is a high fidelity numerical model and a detailed evaluation of the flow physics, in order to design the test rig and improve the pump.

„The prediction of pressure and velocity in the flow through the pump offers excellent quality of results while being very fast regarding time constraints.”

Claude Rebattet,
Head of CREMHyG laboratory

  • Therefore, a list of prerequisites exists:
  • Meshing of detailed CAD geometries
  • Complex kinematics in the model
  • Fast and time-accurate CFD solutions

Using NUMECA, CREMHyG identified potential regions of cavitation and backflow, pressure waves transportation and most importantly, critical locations for placing the experimental pressure tabs.

Find more results in the full article in our ReSolve Magazine – Issue 1, available on our website:


Full-scale CFD Simulation

Full-scale CFD Simulations of the SA Agulhas II


The Sound and Vibration Group (SVRG) at Stellenbosch University is investigating the seakeeping of the SA Agulhas II (SAAII, Figure 1), South Africa’s polar research and supply vessel, owned by the South African Department of Environment, Forestry and Fisheries. She was manufactured by STX Finland in Rauma shipyard, measures 121.3 m between perpendiculars and is 21.7 m wide. She is propelled by four Wärtsilä 3 MW diesel generators that power two Conver Team electric motors which are each connected to a shaft with a variable pitch propeller. Accommodations are available for 44 crew and 100 passengers on annual research and re-supply voyages to Antarctica, Gough Island and Marion Island (Figure 2).

This work will use the CFD capabilities of NUMECA’s FINE™/Marine to determine calm water resistance, added resistance due to waves and motion responses of the vessel, all performed in full scale.

The SVRG uses NUMECA’s FINE™/Marine software package to conduct the CFD study. It offers a seamless workflow, from mesh generation to postprocessing. The geometry and the surface mesh of the SAA II is shown in Figure 3 and Figure 4. An unstructured fully hexahedral mesh with anisotropic cell refinement is used, to ensure both a high-quality grid as well as high control over cells distribution. The inflation technique in HEXPRESS™ allows to insert the near wall cells for an accurate resolution in the boundary layer, keeping both the first cell size (important for turbulence modelling, here using a wall function) and ensuring a smooth transition into the Euler mesh.

The setup of the CFD chain is greatly simplified using the C-Wizard: this tool takes engineering data as input, e.g. drafts, speeds and sea conditions, and provides the full setup from meshing to pre-processing. However, everything can be adapted according to the user’s likings and preferences, making it an ideal tool to get started fast, but also provides full access to the flow solver and numerical details. This is of course important for an academic usage of an industrial code.

Some of the results are depicted below, for the design draft conditions and a vessel speed of 12 kn. Figure 7 shows the wetted surface for a calm water simulation, totalling to 1671 m2 and also indicating some bow spray. In this lowest form of life simulation the grid resolution is kept on the low side, the thickness of the transition zone water to air is directly linked to the mesh cell sizes, as can also be taken from Figure 7.

Accuracy and grid density were however greatly increased in the full unsteady simulations for seakeeping: shown are results for a wave length of 126 m, leading to a wave period of 9 s and an encounter period of 6.3 s, the wave height is 1.6 m. These conditions are quite common during the voyages if the SAA II. The wave patterns are given in Figure 5 and Figure 6: a correct convection of the waves through the domain can be observed, as well as non-linear interactions between the incoming waves and the bow and stern flow patterns. The vessel motions for heave and pitch are shown in Figure 8 and Figure 9, for a total simulation time of 100 s. Although these are regular waves, an out-of-phase response is already observed. This will be investigated in much more detail in the upcoming work, where appendages, irregular sea spectra will be implemented, as well as the rough sea conditions in the the SAA II is operating.

Design and Optimisation of an Inline Pump

Design and Optimisation of an Inline Pump for Marine Applications

Hydro Energy and Pumps | Turbomachinery | Optimisation


The oldest German pump manufacturer, founded in 1860, Allweiler GmbH is today the European market and technology leader in fluid handling solutions for shipbuilding, power generation and special industrial applications. Their product portfolio is unrivalled in the industry and includes centrifugal, propeller, screw spindle and progressive cavity pumps, as well as complete pump systems. They also manufacture peristaltic pumps and macerators.


Pumps in vessels face various challenges: durability and ease of maintenance are a key cost factor, while space confinements demand small overall pump dimensions, including suction and pressure ducts. An interesting configuration is the so-called inline type, where both ducts share an axis, perpendicular to the centrifugal impeller shaft. However, such twisting and bending of a flow channel normally comes with a cost in terms of efficiency and head, plus higher risk of cavitation.


The Allweiler base design featured a centrifugal impeller and a volute plus the pressure duct. For this pump an inline suction duct existed, but it was lacking especially in terms of NPSH. Hence, the challenge was to search for the best trade-off between dimensions and performance.

For the initial CFD simulations the original inline duct was neglected, instead a straight axial suction duct was used. This design’s performance was evaluated, namely head, efficiency and NPSH. The rotor was meshed with AutoGrid5™, the volute (and later the suction duct) were meshed in HEXPRESS™, leading to a total mesh size of around 4M cells. Using a 24-core machine, each operating point took less than 15min to converge with the CPU-Booster™.

As parametric modeller CAESES® from Friendship Systems was chosen. The suction duct was modelled from scratch, via sweeping of a variable cross section along two paths, also subject to optimisation. The rotor and most of the volute were fixed. The latter was cut near the tongue, and a fully parametric pressure duct was modelled: it was linked to the suction side to fulfil the inline constraint and featured independent parameters to control the curvature.

Using this parametric setup, an initial design in inline configuration was created and evaluated in CFD. It showed adequate performance compared to the axial inflow conditions, and hence the optimisation problem could be set up as follows:

  • Usage of 3 operating points (OP)
  • 10 free parameters
  • Genetic algorithms in both single- and multi-objective formulation
  • Objectives: total pressure rise in all OPs, NPSH and the total height
  • Alternatively: hydraulic efficiencies, impeller passage flow rates

For all operating points a strong correlation between height and stage performance could be observed: clear Pareto fronts for the contradicting objectives formed, indicating the feasible design space of the problem. From the various designs two were picked for closer evaluation:

  • A balanced design, with moderately decreased height but increased performance in all three OPs (head, efficiency, NPSH)
  • An extremely flat design, but also some losses in respect to the initial inline design

As always with optimisation problems, it is the customer’s demands that decide which solution fits best and is hence optimal. In this case, Allweiler chose the balanced design: it achieved the predicted performance in experiments, also regarding the important NPSH, and is now used in the inline marine pumps portfolio.

Statement from Allweiler

“The numerically optimised design featuring an inline duct showed good performance on the test bed – we just swapped the axial duct with the inline one, and the NPSH of the pump stage could still be kept.”
Manfred Britsch

„This optimization project gave us interesting insights into the skilful parameterization of the bend geometry. The use of genetic algorithms led to a geometry that would probably not have been found with conventional trial and error methods.“
Stephan Kern

Unusual Turbine Architecture

A somewhat unusual Turbine Architecture (OTH-AW)

Turbomachinery | Grid Generation

You think you have seen them all? Take a look at the animation in Figure 1 and prepare to be surprised. What you are seeing is a first impression of one of the latest projects of Prof. Dr. Andreas P. Weiß and his team at the OTH[1] Amberg-Weiden – the so called “Elektra-Turbine”. It is a redesign of a radial re-entry cantilever turbine, originally invented at the beginning of the 20th century. At the time, these turbines (20 – 300 kW) were applied on ships to directly drive pumps and fans at very low rotational speeds (equal to 3000 rpm). The general idea of the current research project is to investigate rather uncommon turbine concepts which bear the potential to be used in small, decentralized power plants.  As a first step, a 5 kW air-driven redesign has been built and investigated both experimentally and numerically in cooperation with NUMECA Ingenieurbüro.

Since this turbine is not particularly a standard design, the reader is familiarized with the flow path at first: Pressurized air (p0 approximately equals 10 bar) enters the turbine stage and is subsequently accelerated in a supersonic nozzle where it reaches a Mach number of nearly 2.7. The flow then passes the rotor twice, converting this kinetic energy into mechanical work at a ratio of 3 to 1 like in a Curtis stage The preliminary design of this geometry was developed by OTH-AW with an 1D turbine design tool, also supported by means of 3D CFD-simulations. Both steady and unsteady RANS-methods were applied using the EURANUS flow solver from NUMECA.

The simulation model incorporates one periodic pitch which extends over 120° for this case. It has been meshed with a fully structured multi-block grid by means of NUMECA’s meshing tools IGG™ and AutoGrid™. For the steady simulation, flow quantities are exchanged via a full non-matching frozen rotor interface between stationary and rotating domains. This approach however does not lead to a satisfactory agreement when compared with measurements of the global turbine efficiency. Instead, the average of an unsteady simulation has to be computed, in order to achieve a better match. However, the results from the steady CFD solution provide the means to obtain a first impression of the flow dynamics and to reveal some potential for further improvements of the geometry.

Figure 3 displays four representative flow quantities on a constant span position at 50% of the nozzle height. They stem from a steady state computation at design conditions. To begin with, the reader is referred to Figure 3 a). The flow has entered the nozzle and expands through the divergent part where it reaches a maximum Mach number of approximately 2.7. The divergent part of the nozzle terminates with a sharp edge on one side and with a minor turning on the other side (the flow is turned into it-self). Thus, the expansion stops momentarily with an oblique shock wave, labeled with shock S1 in Figure 3 b).  Downstream, the flow decelerates but remains supersonic. As it approaches the leading edge of the turbine wheel, the flow is divided into two parts, each experiencing different area progressions. The “left part”, relative to the view in Figure 3 a), accelerates again to a Mach number of 2.4, since the blade is forming another diverging nozzle together with the turbine casing. Ultimately, another shock wave is formed (shock S2) leading to a strong positive pressure gradient in the streamwise direction. This might cause the considerable amount of the flow which is not entering the blade passage but passes directly over the leading edge towards the outflow instead. The leakage jet becomes clearly visible by plotting the magnitude of the absolute velocity, as can be seen in Figure 3 d). This mass flow leakage is estimated to be approximately 9% of the inlet mass flow by the steady state solution. The remaining flow overcomes the turning in the turbine with another oblique shock annotated with S3. Except for a small region with flow separation on the suction side (induced by the show waves S2 & S3), the flow remains supersonic until the blade passage diverges and guides the flow towards the deflection channel.

In this channel, the flow suffers from viscous dissipation causing a pressure loss of approximately 20 kPa. The main reason for this is twofold. First, as a result from the complex flow phenomena upstream, the flow exits the turbine with very high velocity gradients in both the circumferential and the spanwise direction. This non-uniform flow is propagated downstream which leads to an inhomogeneous flow field in the entire channel. The second reason is the significant turning of the channel in conjunction with a high ratio of the channel width to the curvature radius. The strong curvature and the associated pressure gradient normal to the inner curve leads to a very low pressure and the flow reaches sonic conditions. As a result the flow close to the inner radius has to overcome a pressure increase from approximately 50 kPa to about 100 kPa (see Figure 3 c) ) before the flow enters the turbine wheel a second time. This leads to a relatively large area with flow separation, which can be clearly seen in Figure 3 d).

Complementary to the frozen-rotor approach, a URANS computation has been conducted to analyze the influence from highly unsteady interactions throughout in the turbine, particularly between the nozzle exit and the turbine wheel. A direct comparison reveals that the mass flow leakage is about 6 percentage points higher in the unsteady simulation. Hence, 9-15% of the incoming mass flow do not contribute to the energy conversion and are practically lost. Also, the total specific enthalpy drop appears to be overestimated by the steady simulation by nearly 11%. The influence of both effects can be directly noted in the isentropic efficiency, which is predicted to be 39.8% for the unsteady case and 47.1% for the steady simulation. The measurements show an efficiency of approximately 33% – an offset of 7 percentage points compared to the unsteady average.

Based on the results of these preliminary investigations, Prof. Weiß and his team at the OTH will continue their efforts in cooperation with NUMECA Ingenieurbüro. The objective will be to reach an isentropic efficiency of a least 50%. To be continued!

[1] Ostbayerische Technische Hochschule Amberg-Weiden, Centre of Excellence for Cogeneration Technologies


The work presented was conducted within the frame of the BTHA-project (JC 2018-56) “Low cost turbo expanders for decentralized energy applications – possibilities of 3D print manufacturing from modern plastic materials” in close collaboration with colleagues at the University Centre of Energy Efficient Buildings (UCEEB) at Technical University in Prague.  The authors want to thank the “Bayerisch-Tschechische Hochschul-Agentur (BTHA)” for their financial support.

Figure 1: A contour of on 50% span showing the absolute velocity for one timestep of an unsteady simulation

Robust Design Optimisation

Propeller Optimisation


The goal of this case is to optimise the shape of the propeller, but at the same time reduce the impact of manufacturing variability on the performances. This is of particular importance for marine applications, where even small performance drops during operation can have significant financial impact through increased operational costs. A standard (deterministic) optimisation is compared with a robust optimisation, taking these manufacturing uncertainties into account.

“The robust design optimisation reduces the standard deviation of the efficiency by 17.7% which means that the variability of the efficiency is reduced.”
Dirk Wunsch, Head of Robust Design group, NUMECA International


  • Perform an uncertainty quantification analysis to better understand the sensitivities of each geometrical feature
  • Elaborate an optimisation accounting for these uncertainties (called “robust design optimisation”)
  • Compare the robust design optimum with the deterministic one

Optimisation Results

Deterministic Design
Mean efficiency: +8,5%
Std. Dev. efficiency: +2,6%

Robust Design
Mean efficiency: +8,5%
Std. Dev. efficiency: -17,7%


Both optimisation studies, deterministic and robust lead to a performance increase of 8.5% in mean efficiency. The important difference is that the deterministic optimisation increases the standard variation of the efficiency (thus the variability of the efficiency) by 2.6%. The resulting design is more sensitive to manufacturing influence on the propeller shape.

The robust design optimisation, on the other hand, reduces the standard deviation of the efficiency by 17.7%. The resulting blade shapes vary significantly between the standard and robust optimal shapes. This underlines the importance of robust design optimisation.

In addition, uncertainty quantification studies of the propeller, allow to better understand the system behavior and to identify the most influential parameters by means of scaled sensitivity derivatives.

Conjugate Heat Transfer

Numerical Modelling of Heat Transfer in a Radial Compressor


In the following, the workflow for reliable temperature field prediction is presented within the simulation of a radial compressor, the SRV4 shown in Figure 1. The simulations are conducted with the highly efficient NUMECA tools, which allow computations at low costs.
First different levels of detail in modelling “Conjugate Heat Transfer” (CHT) are conducted by considering different domains using a fully structured approach. Second the meshing and simulation strategy comparing a structured and hybrid approach are investigated.

Levels of Detail in Modelling CHT

The investigated radial compressor (SRV4) consists of two solid bodies, the rotor and stator, as shown in Figure 2. Adjacent to the main flow channel is a cavity, which is modelled with another fluid domain. Different levels of details of the domain are generated in order to investigate their influence on the resulting temperature field:

  • (a) No CHT: Fluid domains only (main flow channel and cavity)
  • (b) Lean CHT: Fluid domains + solid blades
  • (c) Full CHT: Fluid domains + solid blades + solid hub body and casing

In order to better capture the flow pattern in the cavity under steady flow conditions, the rotor-stator interface is located between cavity (fluid, rotating) and casing (solid, non-rotating). By considering the heat transfer in the solid domain, the accuracy of the temperature prediction increases: The temperature peak at the blade tips at trailing edge is lower when using CHT in the solid part of the blades, as shown in Figure 3. Taking also into account the solid hub and casing, the heat flux inside the impeller transports the heat towards the bulb of the impeller, resulting in a higher temperature, as shown in Figure 4.

Meshing and Simulation Approaches

Two different strategies of meshing the radial compressor are used for detail level (c) with full CHT, namely a fully block-structured and a hybrid approach.

Block-Structured Approach

The block-structured grids are generated using AutoGrid5™. Instead of an HI topology, which is in general beneficial for the grid quality in the presence of splitter blades, the 4HO topology is used to obtain a matching connection inside the solid part of the blade between suction and pressure side. A matching connection in the solid domain is crucial in the presence of stronger temperature gradients, which is the case in the blade towards the trailing edge. The meridional technological effect in AutoGrid5™ is used to extrude the 4HO topology from the main flow channel to the solid impeller, resulting in a matching connection between these two parts.

The computation is performed with FINE™/Turbo and the coarse grid initialisation technique strongly accelerates the convergence. Due to the mixing plane approach (fluid-fluid and fluid-solid) only one blade passage is meshed for each domain.

Hybrid Approach

In the hybrid approach, the structured mesh from AutoGrid5™ is used for the fluid domain, but the solid mesh is generated using OMNIS™/HEXPRESS. At the fluid-solid interfaces, the grid resolution of the solid domain is set in accordance to the cell sizes of the fluid mesh to minimise interpolation errors.
The computation is performed with FINE™/OPEN using an implicit method in the solid domain of the CHT computation, which strongly improves the convergence behaviour.


Considering the complete domain, the heat flux transports the heat towards the bulb of the impeller, as visualised in the meridional view in Figure 5. This mechanism can also be seen in Figure 6, where the heat transfer coefficient is shown, indicating a heat flux solid-to-fluid and fluid-to-solid.
The temperature field of the CHT simulation can be used for structural simulations. The application of the temperature field from a CHT simulation typically has a moderate influence on maximum stresses and eigenfrequencies, but can have a large influence on the lifetime calculation in regions of strong temperature gradients.

For more information, please do not hesitate to contact us.

AutoSeal Titanic

AutoSeal – From CAD or STL to watertight Geometry in less than a Minute

Marine | Grid Generation

CFD techniques keep on improving every single day. So is the complexity of the geometry that engineers are required to model. And on top of the complexity, the CAD or the triangulation of this particular model can be done outside the CFD group, by people who have different requirements for the definition of the model. It is quite common that CFD engineers have to fill holes and remove gaps between surfaces to mesh correctly and easily. We heard these stories so many times… So we had to do something about it.

Many techniques exist for that purpose: capping the holes, removing geometry features, wrapping the geometry, etc., which are features that many CFD providers offer, as we do as well. They are often used in the absence of better. They all have their pros and cons but none of them could reach all the challenges at the same time that our users are requesting: it should work on CAD or STL files, no detail of the geometry should disappear, it must be as automatic as possible and it should be fast. That is exactly why HONDA challenged us on that particular topic, and we responded with the best and the most innovative approach currently present on the market; that we called: AutoSeal.

Directly from OMNIS™, the user defines only 3 settings: the geometry to inspect, several points to determine what is inside or outside the geometry and the minimum hole size under which the holes should be closed.

Let’s take a complex geometry: for instance, the Titanic. It contains more than 20k surfaces. If we want to create a mesh around that ship, we have to make sure it has no holes (which is very important for the Titanic) to avoid meshing inside it (or that the water goes inside the ship – pre iceberg encounter of course).

We can then define 5 inside points (4 into the chimney and 1 inside the hull) and 1 outside point external to the ship.

A minimum hole size of 0.7m can be defined, corresponding to the window size we see on the side of the ship, and we press the Apply button to launch the process on an 8 core machine. 282 closing surfaces and 50 seconds later the result is shown on the screen.

It has even found gaps between surfaces of some decks we might not have expected.

Instead of a week, engineers now only spend a few minutes to 1 hour to make the geometry ready for meshing without any simplification of the model by using AutoSeal. Surfaces are created exactly at the location where needed so that the mesh generation will go as smoothly as possible! This technology works for everything: complete cars, ships superstructures, combustion chambers, offshore platforms, or whatever challenging geometry you have, including CAD and STL, even coming from a 3D scan. Just request a free demo license to try it out by yourself! It will change your life.

AutoSeal Automotive

HONDA testifies a major Breakthrough in Meshing Speed with NUMECA’s AutoSeal and HEXPRESS™/Hybrid

Automotive | Grid Generation

You may have heard before that “80% of the time spent on CFD simulation for aerodynamics is used just for preparing the set-up of the case”. Typical challenges in the simulation preparation process are still: poor quality of the available CAD, resulting in dirty geometries containing holes and cavities, complex geometries, resulting in a large amount of manual input to create a quality mesh, and difficulty to quickly create design alternatives for iterative testing.

NUMECA’s challenge for advancing its meshing tools, and staying ahead of the CFD race, is therefore clear: achieve the fastest possible turnaround times for the highest quality of meshes. With the latest version of HEXPRESS™/Hybrid incorporating AutoSeal, we have done just that.

HONDA has been using HEXPRESS™/Hybrid for many years now for their automotive designs, and the AutoSeal technology made a real difference for them. Closing the holes and cavities of their cabin model with AutoSeal made a huge difference in speed, with the same excellent, detailed result.

Mr Akio Takamura from HONDA Automotive testifies: “A skilled engineer typically needed one full week to close all the holes of the cabin space. Thanks to AutoSeal this is now reduced to 1 hour! It’s robust too, we tested it on 10 different models with 100% success rate.”

AutoSeal automatically creates closing surfaces for dirty geometries containing holes and cavities. A huge time-saver, as this is a tedious job the engineer used to have to do manually.

To make sure using these meshes gives the same results as the CAD software models do, HONDA Automotive compared the impact in terms of aerodynamic performances.

“We compared aerodynamic performances on over 10 different models, by looking at aerodynamic coefficients and heat exchanger passing wind speed, and found little to no difference.”

“Meshing is so fast!” stated Mr. Akio Takamura. “We have observed speed up improvements for each version of HEXPRESS™/Hybrid for the past years, but this one is a major breakthrough.”

“Meshing is so fast! We have observed speed up improvements for each version of HEXPRESS™/Hybrid for the past years, but this one is a major breakthrough.”
Mr. Akio Takamura

To speed up the meshing process even more, HONDA is using another very useful advantage of HEXPRESS™/Hybrid: meshing in parallel using distributed memory. This means that meshes can be generated in parallel on several computers or cluster nodes. The results: tens of millions of cells can be created in a matter of minutes! The graph and table below give you an idea of what our mesh generator can achieve on a typical case. A 43M cell mesh for instance, running on 192 cores, can be created in less than 4 minutes. Mesh generation thus becomes a pre-processing step of the solver, which will be run on the same number of cores and partitions.

Faster turnaround times will enable our automotive customers to save millions of euros on digital prototypes every year. And it will allow the engineers to dedicate their time to what counts: optimising their designs.

“Faster turnaround times will enable our automotive customers to save millions of euros on digital prototypes every year.”

Want to know more? Would you like a demo? Send us a short note at or leave us your contact details here and we’ll be in contact very soon.

Optimisation EGR

Renault: Aerodynamic optimization of Exhaust Gas Recirculation (EGR) for car compressors

Automotive | Turbomachinery | Optimisation | Meshing

Pollution and fuel consumption are the most important features of thermal engines that must be drastically improved in the coming years. With soaring pollution in cities around the world, legislators are demanding car manufacturers to put systems on the market that are as clean and efficient as possible, whatever the driving style or conditions, from traffic jams to highload mountain trips, in hot and cold conditions.

Furthermore to reduce CO2 levels, thus diminishing the effects on global warming, fuel consumption must be significantly limited in those same real-world usage conditions. These environmental pressures are translated into legislation: the newest EU7 emission rules and new CAFE regulation will be implemented in Europe from 2023.

In order to achieve these goals, all consuming parts of cars are being meticulously analyzed to try and decrease losses to the maximum through design improvements, while taking into account inherent negative effects such as condensation issues for the compressor. It is within this framework that Renault turned to Numflo – the consulting group of NUMECA International, which is renowned for its top-level expertise in multiphysics design and analysis. The specific focus of the first study was to evaluate the impact of Low Temperature Exhaust Gas Recirculation (LT-EGR) on the efficiency of their turbo-compressors through CFD analysis. The power and reliability of NUMECA software tools were crucial for this project. At the end of the study, using the flow solutions obtained by Numflo, Renault performed condensation analyses with dedicated software. Condensation can indeed occur when ambient temperatures are low and, in the long term, can damage the blades and create icing problems.


The study focuses on the impact of 5 geometrical parameters of the LT-EGR injection on the compressor wheel efficiency:

  • Radius of the EGR injection
  • Axial distance between the EGR injection and the compressor
  • Three angles defining the orientation of the EGR around the inlet pipe.

The EGR geometry is generated using the IGG™ block structured mesher, with a script. This script automatically generates a new geometry for each new set of the 5 parameters. The inlet pipe and the compressor wheel are provided by Renault and remain unchanged during the simulation process.


On the numerical side, the meshing of the wheel consists of a high-fidelity block-structured grid, created with the automatic mesh generator AutoGrid5™. Only one periodic channel of the ,wheel is meshed. For the inlet pipe and the EGR, an automatic unstructured mesh is generated with HEXPRESS™/Hybrid software. The meshing process of the inlet pipe and EGR is automated using a dedicated script, ensuring good quality
meshes regardless of the position of the EGR. Then, both meshes are reassembled, creating the final mesh used for the CFD simulations.

Non-Linear Harmonic approach

Flow distortion introduced by the inlet pipe and EGR inside the impeller is considered by means of the innovative Non-Linear Harmonic (NLH) method in FINE™/Open. This approach solves flow perturbations in the frequency domain, allowing for high accuracy of numerical results in comparison with state-of–the-art time-marching models, at a significantly reduced computation cost. This approach allows for transmitting the 360° flow distortion generated inside the inlet pipe into the wheel, having a direct impact on its aerodynamic performances.

Simulations set-up

The fluid is air considered as a perfect gas. The following boundary conditions are used:

  • Main inlet : Total conditions and the flow direction
  • Secondary inlet : Massflow
  • Main outlet : Massflow

The first solution is started using constant values per domain.


The DoE (Design of Experiments) is generated using FINE™/Design3D with the Minamo module. A total of 26 elements is generated randomly using a « Latinized Centroidal Voronoi Tessellation » law. The analysis of the results of the database shows that higher wheel efficiency is obtained for an EGR located far from the wheel, and for an EGR with a small radius. Further flow analysis indicates that the best configurations demonstrate an important mixture between the main flow and the flow coming from the EGR, leading to less distortion at the inlet of the compressor wheel.

Based on the Minamo module, it is possible to run a deep analysis of the database to understand the influence and the relation between the free parameters and their impact on the performances. Below, an “ANOVA” graphic is providing the global sensibility of the free parameters for the considered objective.

“Higher wheel efficiency is obtained for an EGR located far from the wheel, and for an EGR with a small radius.”

A “Self organizing map” can also be used to project multidimensional data on a 2D plot. Based on an objective, it can easily allow the engineer to check if the available free parameters have (or not) the same impact on the objective. An example is provided below for the free parameters and a given objective. The highest value of the objective (green rectangle) corresponds to high values of “L” and low values of the “GAMMA”.


The condensation analyses performed by Renault, based on the Numflo CFD results, demonstrate that improvement of the wheel efficiency and improvement of the condensation issues lead to a selection of opposite values of the free parameters. In other words, if we want to increase the wheel efficiency, we inevitably also increase the condensation phenomenon.

The figures below show the impact of the two most important geometrical parameters of the EGR (distance to compressor plane and EGR injection diameter) on the efficiency losses and condensation index. The size of the bubbles is proportional to the losses and condensation level. From the left graph, it can be seen that losses are minimal for a high distance of the EGR to compressor plane and for a small radius of the EGR. However, this region corresponds to the worst condensation index. A compromise has to be performed between best efficiency and low condensation index as these 2 objectives are antagonistic.

The next step will be to perform an optimization taking into account the CFD (aerodynamic phenomenon) and the condensation aspects in a coupled manner.


Stephane Guilain, Tech. Expert in PWT Aerodynamic and Engine Air Filling DEA-MA – Advanced Engineering, Renault
Donavan Dieu, Senior Consulting Engineer, Numflo

Rotating Wheels

The World Solar Challenge - Aerodynamic simulation of rotating wheels in a solar car

Automotive | Grid Generation | External Aerodynamics

Every 2 years competitors from around the world set out on a grueling race across the Australian outback: the World Solar Challenge. The competing solar-powered cars are built by some of the brightest young minds on the planet. Student teams from all over the world, push the limits of technological innovation by engineering and building a vehicle with their own hands, powered only by the sun. The competition is designed to promote research on solar-powered cars.

The Belgian team Agoria won last year’s race with their car called BluePoint. BluePoint is the result of years of research by different teams of thesis students from KU Leuven.

One of the most important objectives of solar-powered car development is to minimize power consumption by reducing drag. To find the optimum design within the limits of the competition, the Agoria Solar Team ran many simulations using OMNIS™, and were able to make a large variety of design changes thanks to its quick and easy, yet high-quality mesh production.

Study of the influence of the rotation of the wheels on the drag

Where in previous aerodynamic simulations of the car, the wheels were not taken into account or considered as static, the thesis of Kristof Borgions and Thomas Holemans [1] discussed in this article, focuses exactly on this part of the car. The influence of the rotation of the wheels in this research was investigated using FINE™/Open with OpenLabs™.

Simulations of the solar car were performed both with rotating wheels and with stationary wheels to be able to evaluate the impact of the rotation on the total drag of the car. A particularly interesting study for comparison of the simulation results with real-life road conditions on rotating wheels and with wind tunnel tests where the wheels are stationary.

As a starting point, previous simulations of the car without wheels were considered as performed by Vandervelpen and Uten [2]. To keep computational time within reasonable limits for a master’s thesis work, the rim and tire were simplified. For example the grooves in the tire were not taken into account and the wheel arch was also simplified. The gap for the suspension was closed as the flow inside the car was neglected. In addition, only half of the car was considered in the simulation, neglecting the flow around the canopy.

Meshing and set-up

Starting from a CAD file in Parasolid format, a full hexahedral mesh was generated of around 11.5 Million cells.

OMNIS™/HEXPRESS was used to automatically group the many different surfaces from the original CAD. This drastically simplifies the further steps of the simulation process. For example by having all narrow fillet surfaces in a separate group, additional refinements can easily be made to be able to accurately capture the curvature while keeping the cell count limited. The leading as well as the trailing edge of the car can be captured accurately through an appropriate refinement.

Considering the high focus on the wheels in this work, special attention was put on the meshing of this area. For example, the space between the wheel and the arch needed to be captured accurately. As this is variable, a proximity refinement was used: the cell size was based on the distance between the two surfaces, allowing for a limited cell count.

The simulations in the thesis were performed with the FINE™/Open solver for a 3D unsteady RANS simulation. Based on previous work [2] the k-w SST turbulence model was used, as this would give the best results compared to wind tunnel measurements.

For the simulation with rotating wheels a Moving Wall boundary condition was used for the wheels, with an imposed radial velocity corresponding to the speed of the car. This was justified as the tire grooves and rim spokes weren’t taken into account. In reality stationary wheels (in a wind tunnel) would correspond to a stationary ground, but in this research a moving floor was imposed to be able to compare the results of calculations made with the same conditions.


The simulations were run on 26 cores with 160 Gb of RAM on a workstation of the ‘Applied Fluid Mechanics and (Aero)Acoustics’ research group of KU Leuven, campus Groep T Leuven. A first steady simulation was performed in 52 hours, corresponding to 4.5 CPU.h/Mpoints. For the unsteady simulations there was a remarkable difference: where the simulation with stationary wheels took 440 hours to stabilize, the simulation with rotating wheels took only 44 hours. This difference can be entirely attributed to the vortex shedding that is observed in the case of stationary wheels. Due to the rotation of the wheels, the amplitude and frequency of vortex shedding was significantly reduced.

Skin friction drag was influenced only a little by the rotation of the wheels. Pressure drag however showed to be largely impacted. The simulations also demonstrated that the front wheels had higher pressure drag than the back wheels. This can be explained by the lower stagnation pressure for the back wheel, as it is in the wake of the front wheel. Moreover the flow was approaching the back wheel under an angle coming from the vortex shedding caused by the front wheel. The pressure field showed that the pressure in the wake just downstream of the front wheel (on the left), was lower than the pressure in the wake of the back wheel (on the right).

Ultimately the rotation of the wheels reduces the drag on the wheels by approximately 40%. This has significant consequences for the car as a whole, as it reduces the CdA by approximately 10%. Quite a considerable impact.

The simulations provided a more detailed insight in the flow field structures around the wheel caused by rotation. A small recirculation zone can be noticed at the front of the wheel and the wheel arch. Here the free stream flow from upstream and the flow between the wheel and the wheel arch come together.

The computed CdA compares within 0.95% with the one measured in a wind tunnel. As the present simulations did not include the canopy, the computed results were corrected based on previous results including the canopy [2].


This work with OMNIS™/HEXPRESS and FINE™/Open with OpenLabs™ is an example of a work performed by students in the limited time available for a master thesis. It provides more insight into the flow structure around rotating wheels and its significant impact on pressure drag. Rotation of the wheels causes as much as 40% decrease in drag on the wheels, resulting in a 10% reduction of the CdA of the complete car. This work helps in particular to increase confidence in the comparison of simulation results with experimental results in road conditions and wind tunnels.


  • [1] Borgions K., Holemans T., Aerodynamic simulation of rotating wheels in a solar car. KU Leuven, Faculty of Engineering Technology. Master thesis, 2019.

[2] Vandervelpen E., Uten J., Testing of turbulence models for the aerodynamic simulations of a solar car, KU Leuven, Faculty of Engineering Technology. Master thesis, 2018.


Thomas Holemans, student, campus Groep T Leuven, KU Leuven
Kristof Borgions, student, campus Groep T Leuven, KU Leuven
Maarten Vanierschot, Assistant professor, campus Groep T Leuven, KU Leuven
Colinda Goormans-Francke, Head Academic Products and Applications, NUMECA International

Comparison of Numerical Schemes

Aerodynamic coefficients of a generic missile geometry

Aerospace | LDFSS


This study demonstrates the application of NUMECA Software on a generic missile, the ANF (Army-Navy Basic Finner). It is a typical test geometry, meaning that there is plenty of reference data available to the public, making it ideal for comparison and validation of various numeric methods. The reference data is taken from US Army Research Laboratory Report ARL-TR-6725 (November 2013), showing both free-flight test and wind tunnel data, in a huge speed regime from Ma 0.5 to Ma 4.25.

Geometry and Domain Preparation

The geometry of the ANF is quite simple and can be taken from Figure 1. The dimensions are given in calibre (1 calibre = 0.03m), meaning that the missile is 0.3m long, the various free fight models weighted 1.5894 kg. Since a low Reynolds approach was targeted, small near-to-wall cell sizes are required for such high speeds. This in turn also requires a fine triangulation of the geometry, also when keeping the very small radii at the missile nose and fin leading edges in mind (0.12mm).  For most of the simulations a large 1/8th sphere was used (R=20.0m, see Figure 3), making use of two symmetry planes for the cases at 0° angle of attack. Some configurations required an AoA of 1° however, making it necessary to increase the domain to 1/4th sphere.

Mesh Generation

Pure hexahedral numerical grids were generated to achieve a high mesh quality while keeping the cell count reasonable. Three grids of different density were created, an overview on cell count and quality is given in Table 1. Special attention was given to flow relevant features:

  • Missile cone and the tiny nose
  • Fin leading edges
  • Blunt fin trailing edges
  • Blunt missile aft body
  • A thin wall type refinement in the core wake
  • Target y+ < 5

Some impressions of the fine (15M cells) mesh are given in Figure 4 to Figure 7.

Case Setup and Pre-processing

In all simulations the CPU-Booster™ was used, both during coarse grid initialisation and on the fine grids   Some other specifics were:

      • Inlet turbulent quantities depending on free stream velocity (for 1% intensity and viscosity ratio of 1; Table 2)
      • Air prefect gas model (a comparison with real gas showed a drag difference of less than 1% for Ma 2.5)
      • SSC-EARSM turbulence model (separation sensitive corrected, anisotropic)
      • Comparison of two numerical schemes:
      • Classic Jameson-type dissipation scheme (Matrix-scheme)
      • Low-diffusive flux splitting scheme (LDFSS)

    Quantitative Results

    The reference data contains results ranging from Ma 0.5 to Ma 4.25 and were taken from wind tunnel experiments, as well as free flight tests (using numerous single-usage projectiles!).

    Three coefficients are of main interest:

        • Drag coefficient Cx0
        • Lift coefficient derivative CN,alpha0
        • Moment coefficient derivative Cm,alpha0

    While the drag coefficient can be derived from only one simulation at 0° angle of attack, the lift and moment coefficient derivatives are calculated from two simulations at AoA 0° and 1°, respectively.

    When comparing the experimental results for the drag only, significant differences can be observed: around Ma 1 wind tunnel tests (WT) indicate a quite higher drag coefficient than the free flight tests (FF), while above Ma 1.5 this is reversed. From Ma 3 on the experiments seem to converge towards comparable values (Figure 8, Figure 9). Both CFD schemes and the grid resolutions show a clear trend:

        • Matrix scheme: drag coefficients vary slightly with grid resolution, for Ma > 1 an overshoot of the 5M and an undershoot of the 10M mesh is indicated when comparing to the 15M cells mesh. Overall trend of CFD data is very good, however in general drag is a bit higher than in both experimental tests.
        • LDFSS (low-diffusive flux-splitting scheme): a clear trend in grid resolution is observed, drag values converge with increased cell count over the full operating range. Overall trend is again very good, but the new, less diffusive scheme shows substantial decreased drags for all Mach numbers, being very close to the wind tunnel data below Ma 1.5. Above that speed CFD results are close to the wind tunnel test until finally converging to 0.25 drag coefficient at Ma 5.

    The lift and moment coefficient derivative are shown in Figure 10 and Figure 11. Again, the CFD results match very well, also the sharpness of the extremum in both curves is captured nicely. For these data only the small 5M cells mesh is applied (actually this is 5M per 1/8th of a sphere as also given in Table 1, making the full domain 10M cells), so no grid study was performed here. But the two numerical schemes were applied, and for these coefficients the differences are far smaller.

    Now, who would do CFD and check numerical values only? Exactly, every CFD engineer likes qualitative results in colourful pictures! The space in this case study is limited, but a few impressions can be taken from Figure 12 to Figure 14:

        • Figure 12 shows the missile at Ma 2.5 on the fine (15M) grid, comparing the numerical schemes for the density gradient and the Mach number. Some of the flow features seem to be sharper with the LDFSS scheme (gradient), also the angle of the shocks is slightly impacted. The mesh cut on the right side shows a refinement zone around the missile, and the changes in cell sizes can be easily correlated to some changes in the flow features.
        • Figure 13 shows again Ma 2.5 and the fine mesh, now displaying an iso-surface of Ma 2.6. The patterns are very different for the two schemes, especially in the fins’ wake and the interaction with the main body wake.
        • In Figure 14 the development of a Ma 1 iso-surface with the missile velocity is given, starting from Ma 0.9 until Ma 1.7.

    Overall, the results are very good, matching all the trends in the experimental data. A great leap in accuracy can be achieved by just switching to the new LDFSS, which reduces the numerical diffusion and hence improves especially the missile drag, and is applicable at all flow speeds. Feel free to give it a try!

  • Author

    Sven Albert, NUMECA Ingenieurbüro

Optimisation for Reusable Spacecraft

Masten Space Systems: Reactive flow and heat transfer optimisation for reusable spacecraft

Aerospace | Optimisation | Multi Physics | Combustion | Heat Transfer

Masten Space Systems is a private company founded in 2004 and headquartered in Mojave, California, USA, by David Masten. Masten’s focus on reusable rocket technology is driven by the goal of enabling space transportation and reliable planetary landers for the Earth, Moon, Mars, and beyond.

One of Masten’s recent pioneering space flight vehicle projects was to design a reusable aircraft designed to fly to the edge of space, the XS-1. The XS-1 aircraft will be capable of launching a 3,000-pound spacecraft to the earth’s orbit at a cost of $5M, which is ten times less than today’s launch systems. A key objective of the program is to fly the XS-1 10 times in 10 days in order to demonstrate its “aircraft-like” operability, cost efficiency and reliability. Key anticipated characteristics of the aircraft include a physical size and dry weight typical of today’s business jets.

One of the new innovations in this effort is a full-scale additively manufactured aluminum thrust chamber assembly for the 25,000 lbf LOX/LCH4 dual expander propulsion system. The dual expander cycle  offers an ideal cycle in this size range, enabling reusability, optimal closed-loop performance, and cost effective design and production price points.

CFD for the simulation, analysis and optimisation of critical spacecraft components

Masten Space Systems is using OMNIS™/Open-DBS with OpenLabs™, NUMECA’s unstructured multi-purpose CFD solver package, for the design and analysis of various elements of reusable spacecraft and lunar vehicles for the testing of thrust chambers. In this article we talk about the numerical studies of three of their aeronautical configurations.

Simulation of plume impingement and thermal load

For the testing of thrust chambers, Masten Space Systems had to determine the temperature on the impingement plate in order to avoid reaching critically high temperatures. To do this, they simulated the impact of a hot plume at multiple times the speed of sound and a temperature hotter than the melting temperature of steel on a plate by using an inert multi-species approach. The results are shown in Figure 1, comparing the live fire test of the plume impingement in the photograph with the predicted temperature simulated by CFD. There is a clear qualitative agreement between the two.

Injector study

For the optimisation of spacecraft it is important to clearly understand the conditions of the injectors, combustion chamber and nozzle of the rocket engine. One of the critical points for example is that the walls cannot be subject to excessive thermal stress.

To analyse the combustion process and heat transfer in a generic injector configuration, a CFD study was carried out using a reactive multi-species simulation approach with a reduced chemical mechanism.

Figure 2 beside shows the injector geometry, consisting of two concentric pipes connected to a rectangular chamber. Via the inner tube Gaseous Oxygen (GOX) is injected, whereas Gaseous Methane (GCH4) is injected around it via the outer pipe. The purpose of the chamber is to create a recirculation zone, preventing the flame from blowing out. The recirculation of hot combustion products ensures a stable continuous combustion process after the engine is fired.

Two grids were generated: one coarse mesh at 1.8 million cells and one refined mesh containing 2.4 million cells with cylindrical viscous layers as shown in Figure 3.

The simulation results were in good agreement with the measurement data. Figure 4 shows the simulated temperature field and a comparison between the simulated and experimental values of the static pressure and heat flux.

Optimisation of the cooling channel of the combustion chamber and nozzle of a rocket engine

To evaluate the potential of optimising the cooling channel, a generic test model was constructed with 180 cooling channels. In the baseline design, the cooling channel has an Aspect Ratio (AT) of 4.

Figure 5 displays their total pressure, heat flux and Thermal Barrier Coating (TBC) ,temperature predictions.

These three designs all included contractions of the CH4 channel in the vicinity of the nozzle throat and then an opening up of the passage down stream. On one hand this structure appeared to accelerate the channel flow in the throat region, resulting in higher heat flux, and therefore greater cooling in that area, and maintain attached flow downstream. On the other hand, the opening up of the channel downstream appeared to reduce total pressure losses compared to the reference design.

The uniform aspect ratio baseline (ID 0) produced the lowest total pressure loss by not accelerating the channel flow anywhere, but it subsequently demonstrated the lowest cooling performance. In the search for an optimised design, Self Organising Maps (SOM) were used to cluster the different samples in a space, such that highdimension space data can be visualised in 2D maps and identify the design that leads to the best compromise between cooling efficiency and pressure drop in the cooling channels. By using SOMs the design IDs 184, 227, and 258 were singled out as the ones having the most promising characteristics.


Allan Grosvenor, Aerodynamics Lead, Masten Space Systems, USA
Jean-Charles Bonaccorsi, Technical Director, NUMECA USA
Jan Anker, Head of Combustion Modeling Group, NUMECA International

Turbocharger Compressor Optimisation

Toyota Motorsports: Optimization of a Turbocharger Compressor for Motorsport Applications

Automotive | Turbomachinery | Optimisation | Compressor | Turbocharger


TOYOTA Motorsports is a high-performance testing and development facility located in the centre of Europe, in Cologne, Germany. One focus is on chassis and engine design for automotive and motorsports. Specialising this high technology developments for motor sport engines the existing turbocharger components already show a high-performance level. Hence, further improvement by traditional means, i.e. classical trial-and error procedures are hardly compatible with the required turn-around times of a modern design environment. Numerical optimisation processes allows to explore many designs in an automatic way, thus, the developer has the opportunity to evaluate many more designs compared to what could be achieved manually.

Aside from an increasing engineering complexity, another challenge is that compressor impellers already work very close to the structure-mechanical limits of the material. Most changes in shape immediately lead to an exceeding of the acceptable stress level. An optimisation under exclusive consideration of the aerodynamic behaviour does not guarantee that the resulting optimal design is structurally feasible. Thus, simultaneous optimisation including aerodynamic and structural forces is necessary. This is referred to as multi-disciplinary optimisation coupling Computational Fluid Dynamic (CFD) with Computational Structural Mechanic (CSM) simulations.

Optimisation Targets & Workflow

In the present project, the characteristics of a centrifugal compressor for an exhaust gas turbocharger are optimised by a multidisciplinary coupled CFD-CSM optimisation. The considered compressor stage consists of a radial impeller with six main and six splitter blades with a vaneless diffuser (Figure 1). Two aero-thermodynamic objectives and one structure-mechanical as well as two aerodynamic constraints are considered:

1)     Increase of isentropic efficiency

2)     Same, or higher absolute total pressure ratio

3)     Same choke mass flow as original geometry

4)     Extension of operating range towards stall margin

5)     Maximum von Mises stress are below limit

The objectives listed above require simultaneous optimisation at four operating points: design point (target 1, 2, 5), two operating points close to the stall (target 4) and one point at the choke (target 3).

Coupled CFD-CSM Workflow with FINE™/Design3D

To satisfy the mechanical and aero-thermodynamic objectives and constraints, the CFD and CSM simulations are jointly integrated into one optimisation workflow (Figure 3). Each new design is first checked structurally by the CSM solver. Only if the von Mises stresses do not exceed the specified maximum value, this new design will be computed via CFD and included in the optimisation. Since the computational time for CFD is much higher than the one for a CSM simulation (approx. 2 min), a lot of effort can be saved if a design does not fulfil the structural constraints. However, structurally unacceptable designs are  fed into the learning database in order to drive the optimiser.


The domains for the CFD and CSM simulations are parameterised in AutoBlade™. 154 parameters define impeller, meridional channel and solid body. To avoid a huge number of structural-mechanical non-feasible designs it is decided to keep the hub shell of the base design. Therefore, the parameters, which define the hub shell of the impeller, are not modified during the optimisation.  To further reduce the number of free parameters, the thickness distribution along the camber curve is not modified. In total 33 parameters are considered as design variables for the optimisation for example shape of the hub and shroud, the camber curves as well as the meridional and tangential blade position.


For the automated optimisation workflow, it is important to have a robust setup for the mesh generation. Depending on the ranges of the modified geometry parameters, new designs can significantly differ from the original design. The mesh set up has to be robust in order to avoid mesh and simulation failures due to insufficient grid generation. Therefore, NUMECA’s robust and fully automated meshing tool OMNIS™/AutoGrid is used for the fluid domain. Mesh independence is ensured by a mesh convergence study with three different mesh resolutions (1, 2 & 3 million points). Based on this investigation, the 2 million-point structured multi-block mesh is chosen for the optimisation. To check if the high mesh quality is also ensured for the whole design space, additional to the grid convergence study the robustness test is performed. Therefor hundreds of randomly generated geometries covering the whole range of the design space are meshed automatically. In this check all geometries were successfully mesh and the orthogonality was for no design below 20°.

Optimisation & Results

In order to obtain a meaningful database for the start of the optimisation, a minimum of three times the number of samples in relation to the free parameters are needed. In the present case with 33 free parameters, hence at least 100 samples would be necessary. However, more samples were computed as it was decided to test several database generation strategies e.g. Latin Hypercube. In total four databases were generated parallel in one and a half day. For the optimisation all databases are combined. Out of 292 database designs 270 samples met all quality criteria in terms of mesh and convergence quality, so that a more than adequate database was available at the start of the optimisation.

During the optimisation each new design is added to the initial database. Thus, the database is enriched in the vicinity of the expected optimum. It is recognised during the optimisation run that not all objectives are properly achieved. Hence, the optimisation is stopped, and the priorities of the different optimisation goals were adapted. After the modification a new optimisation run is started. In this simulation run some parameter bounds were reached. Hence the optimisation is stopped and restarted after the widening of the parameter ranges a new optimisation is started. This procedure of repeated adaptions during the optimisation has proven to be useful, since – a priori – the priority of individual objectives can only be determined very vaguely.


Figure 4 summarises the results at best point (OP2) of databases and optimisation runs for the isentropic efficiency and the total pressure rise.

All designs within the red area of Figure 4 show both an increase in the total pressure and efficiency. The initial design is in the lower-left corner of this box. The two best designs D1 yellow and D1 green are compared with each other and the original design. These two designs both fulfill  all optimisation goals and show substantial improvements. By the selection of the designs not only efficiency increase and total pressure ratio are considered, but also the new operating range and the structure -mechanical results are taken into account. Figure 5 shows the geometrical differences between the selected optimal design and the original design. For the aero-thermodynamic optimisation goals the performances relative to the original design is listed with each design.

For the two selected geometries the complete speed line is simulated (Figure 6). Design D1 shows an increase in the total pressure ratio of up to 8.0% relative to the original design with a simultaneous expansion of the stall margin. In addition to the extension of the operating range, the most noteworthy improvement is the positive slope of the speed line close to stall, which, in contrast to the original design, ensures stable operation even in the immediate vicinity of the surge line. All chosen designs maintain the minimum choke mass flow. The efficiency is increased by 1.1 percent. The design with the highest pressure ratio increase also provides the highest efficiency gain.

D1 shows an efficiency increase of 1.4 % relative to the. In contrast, D2 has slightly lower overall pressure ratio and efficiency improvements but features an extended surge limit compared to D1. This is a classical conflict of multi-objective optimisation, where different goals can sometimes act in opposite directions. The final decision is up to the user.

The von Mises stresses in design D1 exceed the maximum permitted limits by approximately 3%, which is still within the tolerance limit. Figure 7 shows the increased stresses along the blade leading edge compared to the initial design. The areas of higher stresses have grown, but the peak value remains within the tolerance limits.


The presented multidisciplinary CFD-CSM centrifugal compressor optimisation is successful and all aerodynamic targets are fulfilled. The structural integrity is also ensured. The results of the optimisation are very satisfactory:

  1. Efficiency increase by up to 1.4% (percentage points)
  2. Up to 8.0% higher total pressure ratio
  3. Maintaining the chocking mass flow
  4. Extension of the surge line up to 5%
  5. Von Mises stresses are below limit
  • Author

    Kathrin Wendl, NUMECA Ingenieurbüro

Trim Optimisation

Reducing fuel usage with ship trim optimisation in FINE™/Marine

Trim | Marine | Ship Design

Ship trim optimisation has recently gained enormous momentum as it can significantly reduce fuel consumption and curb emissions. The drag on a ship changes with the trim angle for the same speed and average draft. As such, imposing an optimised trim angle during the loading process in the harbour ensures a minimal average resistance during the vessel’s journey.

The financial benefits of trim optimisation

Because of the potential for reduced fuel costs, it becomes financially interesting to create a database of optimum trim angles based on a complete trim optimization study for various cruise conditions. Besides, trim optimisation can be performed regardless of vessel type and age, although the size of the benefit depends on the vessel type. For some vessels, e.g. cruise ships, the trim angle cannot be adjusted with the same flexibility, since passenger comfort and facilities impose relatively strict constraints. The benefit is the largest for ships often sailing in part-load conditions such as ro-ro carriers and smaller containerships, with fuel savings up to 5%. For a container feeder of about 2000 TEU, sailing at 22kn, this would correspond to a savings of about 35 barrels of oil per day [source]!

The importance of dynamic trim

In the past, ships were optimised for a single speed and draft even if the ship experienced a wide range of different conditions, requiring changes to the ship speed and the draft. Now, computational fluid dynamics (CFD) is a game-changing tool to generate a matrix of optimal trim and draft conditions quickly and with great accuracy. The fact that the trim of a ship is voluntarily changed from the static position in a harbour will affect the dynamic trim when sailing, so it also has to be taken into account in the calculations.

Hundreds of simulations, one for each combination of initial trim, draft and speed, need to be run to create the optimal trim database, ruling out model testing. Additionally, the computations are performed directly at full scale. This set-up was shown to be required to accurately predict the effects of turbulence which clearly affects the optimum trim angles. Local recirculation and flow detachment can lead to different force predictions between model- and full scale as the Reynolds number, which governs these phenomena, cannot be preserved in the geometrical scaling. Compared to potential code, which doesn’t include the effects of turbulence, the resistance-increasing effects of wall-roughness (fouling) can be also included in the CFD analysis. As such, CFD can provide a much more realistic result of ship resistance during the entire lifetime.

Given the accuracy and consistency of the results, CFD also brings detailed information to complex hydrodynamic problems in a 3D space. The workflow can also be automated to a very high level, as all required actions from the software can be scripted. A naval architect or marine engineer simply needs to enter the real ship conditions, and the software sets up and runs all simulations rapidly and autonomously on a single workstation or HPC cluster.

Our approach

In practice, FINE™/Marine’s C-Wizard matrix mode creates n x m x p computations: one per {draft (n), trim (m)}-couple for the provided list of speeds (p). The displacement is kept the same for all draft-trim combinations. The Z-coordinate of the free surface also remains the same for all computations, as the ship is translated/ rotated to ensure iso-displacement conditions. The user also has the option to use open-water data for real propeller performances through an actuator disk (Figure 2). This further increases the accuracy and realism of the results keeping a low CPU cost.

The drag forces, moments and dynamic trim and sinkage computed by the flow solver are written out for each combination of (draft, trim, velocity) in the post-processing step, which also yields the displacement at each draft. As an example, Figure 3 shown an optimised trim table for a particular vessel that can be prepared based on the CFD results obtained from FINE™/Marine.

It is interesting to note that the whole project for all the hundreds of computations will be performed with one single mesh! This capability significantly reduces the total computation time due to only meshing the geometry and domain once, and it also ensures that the accuracy is just as high. The numerical uncertainty inherent to the creation of different meshes is removed simultaneously, achieved via FINE™/Marine’s unique adaptive grid refinement (Figure 4), which takes care of all the unnecessary refinement for the free-surface, during the simulation, in an anisotropic, automatic and dynamic way.


Trim optimisation is a relatively easy method for ship owners to reduce operational expenses. While in the past it may not have been possible to obtain accurate resistance predictions for a large matrix of initial drafts, trim angles and velocities, CFD is now the tool of choice to obtain a database of optimal trim angles quickly and with high accuracy. The unique features of FINE™/Marine, such as a single-mesh approach with adaptive grid refinement and the usage of real open-water data for propulsion, position the software as the ultimate CFD toolbox for matrix-based resistance applications.

Sonic Boom

Fluid dynamics investigation of the sonic boom on a supersonic aircraft concept

Aircraft | Supersonic

The return to supersonic flight is amongst the hottest topics in aviation today, as several companies (Boom Supersonic and Aerion, among others) are actively developing new supersonic commercial airliners targeted to enter in service in the coming years. In this context, a quiet flight over land is one of the major challenges to ensure the regulatory compliance of such airliners. Several research centers and companies are building demonstrators to show large sonic boom reductions with smarter designs of all the parts of the aircraft.

The nature of the sonic boom is associated with shockwaves that are longitudinal waves generated by an object that travels faster than the speed of sound. Such shockwaves caused by large supersonic aircraft are perceived by people on the ground as a sound similar to an explosion or thunder.

Exposure to aircraft noise is known to be linked to health issues, such as disturbance of sleep, developmental delays in children, obesity, and cardiovascular problems [1]. The International Council on Clean Transportation report published in 2019 warns that a network of 2,000 supersonic aircraft linking 500 city-city pairs would potentially create sonic booms as frequently as once per five minutes in some regions [2]. In view of an imminent return to commercial supersonic flight, the issue of supersonic aircraft noise is back in the spotlight of the industry and research community.

Recent research proves that it is possible to mitigate the sonic boom issue via a careful shaping of the vehicle geometry. The idea behind the low-boom supersonic aircraft concepts is the minimization of the amplitude of longitudinal sound waves also called N-waves.One of the most cutting-edge supersonic aircraft concepts, Lockheed Martin’s X-59 Quiet SuperSonic Technology (X-59 QueSST) X-plane (Figure 1), is set to take-off for its maiden flight in 2021. One of the mission goals is to provide comprehensive statistical data to regulatory authorities in order to introduce changes enabling supersonic airliners flight over land.

In preparation for the X-59 testing campaign, NASA is testing its modernized imaging system that produces high-quality air-to-air Schlieren photography. These experiments gather vital data related to the interaction of shockwaves produced by two T-38 aircraft flying in formation (Figure 2). This type of new high-quality experimental data paves a way to wider use of validated numerical simulation technologies in the development and certification of supersonic aircraft.

The use of CFD in the field of supersonic aeronautics significantly cuts the time to market and associated development costs. Sonic Boom Prediction Workshops organized by NASA aim at assessment of the sonic boom prediction methods reliability. The near-field pressure signatures prediction with CFD is perhaps its key component. Every subsequent workshop features cases with lower noise signatures, thus bringing new challenges for CFD codes. NUMECA is active in this field and was involved in the CFD investigation of a sonic boom and aerodynamic prediction of NASA Concept 25D Powered supersonic aircraft developed in the framework of recent research work [5,6].

Meshing strategy

One of the key objectives of the workshop is to assess the impact of different refinement techniques involving mesh alignment and adaptation. Despite the numerous challenges in supersonic aerodynamics, certain flow aspects can be described using fairly simple analytical expressions. One of these aspects is the Mach cone (Figure 3): shockwaves around a supersonic pointy object with a small cone angle. The cone angle of the shock waves can be determined from the cone angle of the object, and the speed of the object relative to the speed of sound, i.e. the Mach number. This flow property was employed by NUMECA to build a grid where hexahedral far-field cells are aligned with the shock wave.

In the framework of the investigation, NUMECA has built a series of 3 grids (e.g. Figure 4) for C25D aircraft in HEXPRESS™/Hybrid with equivalent settings on the surface. The only difference between grids is linked to volume mesh resolution and cell orientation in the far-field in the vicinity and below the aircraft.

Cell count, millionCell size in the far-field, meters
Orthogonal far-field mesh46.60.2
Orthogonal far-field refined mesh900.1
Aligned far-field refined mesh1050.1

Table 1: Summary of the grids generated for NASA C25D aircraft

When compared to grids generated by NASA and used by most of the workshop participants, NUMECA has managed to significantly improve mesh quality (Figure 5). This led to more accurate results. The CFD solver robustness is also significantly improved allowing the use of higher CFL numbers and hence faster convergence.

Simulation results

The operating point defined in the workshop [8], that is a supersonic flight at Mach 1.6 at an altitude of 15760 meters, is considered. The operation of the engine is taken into account by imposing the pressure at the inlet of the compressor, the so-called aerodynamic interface plane, and the total quantities at the engine exhaust.

In order to speed up the computation, only half of an airplane was studied by imposing relevant mirror boundary conditions on the symmetry plane. The dedicated supersonic inlet and outlet boundary conditions are used at the external domain boundaries.

In supersonic cases, flow quantities cannot be prescribed as boundary conditions at an outlet. Hence the pressure is prescribed at the inlet. This property of supersonic flow also allows creating much smaller domains compared to subsonic flow problems without any impact on the accuracy of the results. Therefore, a relatively small streamwise domain size is used while boundaries are aligned with shockwaves (Figure 7).

First, the prediction of the aerodynamic forces was investigated. It was found to be insensitive to volume mesh refinement and alignment settings. The result on the coarsest 46.6M NUMECA mesh compares well to the one from NASA Langley Research Center obtained on mixed-element 194M mesh that is considered as a reference, as shown in the following table.

Reference data (194.4M cells, Spalart-Allmaras)NUMECA result (46.6M cells, k-omega SST)
Lift coefficient0.05350.0537
Drag coefficient0.03290.0323

Table 2: Aerodynamic forces prediction for C25D Powered aircraft

Then, the sonic boom was investigated by studying pressure disturbance distribution in the near-field of the aircraft. This is a crucial step for computational sonic boom prediction since this distribution extracted from a CFD simulation serves as an input for dedicated solvers methods (e.g. ZEPHYRUS code by NASA) allowing to study sonic boom propagation down to the ground level. The reference data used for comparison is the average pressure distribution of the workshop entries on the finest 194.4M mesh (Figure 8).

A sensitivity analysis has revealed that the turbulence modelling technique has a rather small impact on the pressure distribution prediction at the concerned locations. All of the results presented in the article are obtained with the k-omega SST model.

Sensitivity to grid refinement (Figure 9). The comparison between 46.6M and 90M orthogonal grids reveals that closer to the airplane the N-wave i.e. pressure oscillation with the highest amplitude can be captured well with both grids. Nevertheless, the finer grid shows a clear superiority in predicting smaller amplitude pressure oscillations while this grid also becomes a must to accurately predict the pressure profile further away from the aircraft.

Sensitivity to grid alignment (Figure 10). The comparison between 90M orthogonal and 105M aligned grids reveals the crucial importance of grid alignment for the cases where the highest accuracy standards are targeted. The aligned grid allows the precise capturing of the large-amplitude N-wave sonic boom as well as all of the smaller pressure oscillations.

Sensitivity to numerical scheme (Figure 11). An analysis regarding numerical schemes has shown a clear superiority of NUMECA’s cutting-edge Low-Diffusive Flux Splitting Scheme (LDFSS) that is retained for the presented simulation results. Compared to the standard central scheme of Jameson–Schmidt–Turkel, the LDFSS ensures a considerably closer match to the reference data.

NUMECA’s tools allow an efficient visualization of shockwaves pattern around a supersonic airplane using numerical Schlieren images (Figure 12). This technique can be used for detailed studying of sonic boom formation on an airplane. In this case, it shows that the strongest shockwave is formed at the aft fuselage part. The downstream pressure oscillations that were the most challenging to predict in the framework of the mesh sensitivity analysis are coming from shockwaves interacting with the engine exhaust plume.


Sonic boom prediction is one of the main supersonic aeronautics challenges. The present study shows that NUMECA’s cutting-edge CFD technologies can be efficiently applied in a supersonic aircraft design cycle. Challenging numerical simulations such as sonic boom prediction studies can be accurately addressed in a straightforward manner. NUMECA’s Open-DBS solver with state-of-the-art numerical schemes and tailored full hexahedral meshing approach, led to accurate forces and sonic boom predictions with a lower cell count and a lower computational effort than the reference.

The investigation also underlines the importance of best practices and know-how for meshing and simulation, some of which are demonstrated in this study.


1. Aircraft noise and health effects: Recent findings. Civil Aviation Authority, report CAP 1278, 2016

2. Rutherford D, Graver B, Chen C. Noise and climate impacts of an unconstrained commercial supersonic network. International Council on Clean Transportation, working paper 2019-02, 2019



5. Wintzer M, Ordaz I. Under-Track CFD-Based Shape Optimization for a Low-Boom Demonstrator Concept. 33rd AIAA Applied Aerodynamics Conference, 2015.

6. Ordaz I, Wintzer M and Rallabhandi SK. Full-Carpet Design of a Low-Boom Demonstrator Concept. 33rd AIAA Applied Aerodynamics Conference, 2015


8. Park M. Second AIAA Sonic Boom Prediction Workshop Nearfield CFD Introduction. NASA Langley Research Center, presentation.


Artemii Sattarov is a CFD Application Engineer at the headquarters of NUMECA International. He holds BSc in Aerospace Engineering and MSc in Computational Mechanics. His double degree Erasmus Master’s studies were held at UPC BarcelonaTech and Swansea University. Prior to joining the company in 2017, Artemii was an intern at ANTONOV company, CFD research assistant at Barcelona Supercomputing Center, and student demonstrator assistant at Swansea University.

Hydro energy and Pumps

Investigation of the Motion of a Dual-Disc Check Valve using Transient CFD

Valve | FSI | AGR | Overset Grid


Check valves are commonly implemented in piping systems in order to control the fluid flow and to avoid reverse flows, since these can degrade the performance of the system and can cause structure failures. The opening of the valves is mainly caused by the hydraulic force which is the sum of viscous and pressure-based forces acting on the discs. A dual-disc check valve relies primarily on the spring force for closure. Here, the transient flow through a typical dual-disc check valve is investigated in FINE/Marine and the opening process of the valve is analysed. 

Methodology – opening period a dual-disc check valve 

To simulate the opening period of a dual-disc check valve, an overset grid approach, also known as Chimera, and an Adaptive Grid Refinement (AGR) for grid continuity are applied. The main purpose of the AGR is to ensure a better transfer of data at the interpolation regions during the computation. The kinematics of the discs are solved for one Degree of Freedom (DoF) using a FluidStructure Interaction (FSI) in order to couple the rigid body motion with the flow motion.  

Simulation set-up 

In the considered case, the fluid is water and the flow is incompressible. Different inlet velocities are simulated, leading to different mass flows, accordingly. The three-dimensional geometry is simplified such that the valve is not fully closed, since solid-solid contact is not yet available. Extensions upstream and downstream of the check valve are created, in order to place the boundary conditions far away from the check valve. The geometry and the flow are assumed to be symmetric, hence the simulation is restricted to one half of the complete domain in order to reduce computational resources. 

The grid generation is conducted in OMNIS/HEXPRESS with full unstructured hexahedral grids. The grid has a total number of cells of 1.3 million in the beginning of the computation. In the overlapping regions, the cell sizes of both domains overset grid and background grid are the same in order to ensure a good interpolation of data. 

A wall function is applied at the solid walls and the turbulence model is kwSST. A prescribed reference pressure at the outlet is given as zero-gauge pressure. To initialize the flow field, steady and unsteady simulations are firstly conducted before starting the FSI simulations. 

Aside from the two constant inlet velocities of 1 and 2 m/s, a time-dependent inlet velocity boundary condition is applied using a Dynamic Library. In this case, the inlet velocity increases linearly from 1 to 2 m/s. This can be compared to a pump start-up, for instance 

Aside from the disc’s mass, the inertia tensor and the center of gravity, a spring constant is given in order to include the moment due to the spring force in the computation for solving the kinematics of the discs 


The simulations provide a more detailed insight of the flow field inside the dual-disc check valve, such as recirculation areas in the corners of the valve and on the downstream side of the discs, due to the sharp corners and the sudden contraction and expansion of the flow.  

A correlation between mass flow, moment due to hydraulic forces and disc angle is observed. Increasing the mass flow, i.e. increasing the inlet velocity, leads to a rise of the moment caused by hydraulic forces, and consequently leading to a shorter opening period and higher disc angles (until it reaches the maximal opening angle)[1].  

If the moment due to spring force becomes greater than the moment due to hydraulic force, the disc angle oscillates (see zoom of the Figure 4) until an equilibrium between both moments is reached, consequently leading to an almost constant disc angle.  


The opening period of a typical dual-disc check valve is successfully simulated in FINE/Marine. The results show a strong correlation between mass flow, moment due to hydraulic force and disc angle. The work provides a better understanding of the flow field inside a typical dual-disc check valve. 


[1] Joaquim, A. G. Investigation of the Motion of a Dual-Disc Check Valve using Transient CFD. TU Kaiserslautern, NUMECA Ingenieurbüro. Master’s Thesis, 2020.  


Guilherme Joaquim , Student at NUMECA Ingenieurbüro

HIPER 2020

An electrified RIVA Powerboat - optimised

Marine | Optimisation | Powerboat

1. Introduction

This work was also presented at the HIPER 2020 Conference in Cortona, and the full paper is available here (page 264).
This paper investigates the hydrodynamic performance of a classic wooden powerboat, the RIVA Junior, using CFD and driven by an environmentally friendly electric motor. The original hull shape is parametrised for variation in an optimisation chain: several operating points (speeds, displacements) and various objectives are considered, aiming to model different operational profiles of an electric propulsion. The goal is to derive hull design trends for both highest speeds for a given power, but also maximised range using the limited electric energy, and of course the trade-offs in between.

2. A short description of the key ingredients

2.1 RIVA Junior and operating conditions

Object of interest is the RIVA Junior hull, a hard-chine motorboat with a single propeller and rudder. With the electric drive and its battery challenge in mind, two displacements are investigated (1m³ and 1.2m³). The idea is to model two powertrain versions (motor, controller, battery) and find the best-possible in terms of speed and range. Furthermore, a large range of speeds is considered, covering the full operating regime:

1) 5m/s: A slow full-displacement mode which might be used near the marina, when aiming for a relaxed ride only, or to save battery to come home safely.

2) 20m/s: A high-speed ride for maximum pleasure and adrenaline.

3) 12.5m/s: A moderate speed mode in between for the rational driver.

    • This leads to six operating points per design, which is quite an effort for a high-fidelity CFD simulation with motion coupling and free surface modelling. Some strategies to speed up will be given in the following chapters.

2.2 Parametric modelling and simulation ready CAD

A complex geometry like a planing hull requires a powerful modelling tool, which is CAESES® from FRIENDSHIP SYSTEMS GmbH. A total of 10 free parameters is used, which is a quite low number for the geometric variability it provides. This is a key point, as an increased number of free parameters largely increases the optimisation costs.

2.3 Meshing

The meshing is done in HEXPRESS™, which generates a fully hexahedral unstructured grid with hanging nodes. Some impressions of the mesh are given in Figure 1. The CFD-experienced reader will directly stumble over the missing discretisation along the free surface. For accurate free-surface modelling a fine layer of cells is required perpendicular to the water surface, but in our case these cells will be generated directly in the flow solver using an adaptive grid refinement strategy, see the chapters below.Mesh dependency is investigated using the base design. Since quite a few designs are foreseen in the optimisation, and on 6 operating points, each extra cell will sum up to a quite large difference in total turnaround time. Strongly dependent on the vessel speed and the design this then leads to a final average mesh size of 700k cells.

2.4 Pre-processing and AGR

FINE™/Marine is used as flow solver, a dedicated CFD system for the marine engineer. An adaptive grid refinement technique (AGR) is fully integrated, which adapts (that means refines or de-refines) the mesh during runtime, both in space and time. Several criteria are available, e.g. gradient and overset grid continuity ones, but the widest used are multi-surface ones (free surface or cavitation bubbles, see Figure 2). A dynamic CPU load balancing ensures an efficient CPU usage, so this itself is already a very dynamic system.

2.5 Optimisation strategy

The optimisation toolkit is FINE™/Design3D, which uses Cenaero’s MINAMO package, the optimisation workflow is depicted in Figure 3 and can be separated into three parts:

1) A database containing random samples is generated (the exact method used is called Latinised Centroidal Voronoi Tesselations). The goal is to gain a large diversity and little clustering of design parameters with as little samples as possible.

2) Then an online surrogate-based strategy is used: RANS-CFD simulations are still comparably expensive, which is why the discrete input-output parameters defining a specific design are approximated via a surrogate model to deliver a continuous description. On this the actual optimiser, a genetic algorithm, is applied, calling the surrogate model thousands of times.

3) Interesting candidates are then passed to the CFD chain to get an accurate evaluation of the new designs, and these are then fed back to the database. This is what online stands for in the strategy.

2.6 Efficient CFD chain challenges

The first simulations on even only slightly modified geometries already showed a huge impact on hull resistance and trim angles, but also on the solver runtimes and convergence behaviour. A planing hull is a complex and very dynamic system, and especially the database with random samples can deliver unfavourable geometries. This requires a detailed analysis of such geometries to correctly handle the outputs of interest, find quantities that clearly define such undesired behaviour and, in the end, control the optimisation process and deliver reasonable and consistent data to the optimiser. The latter is of utmost importance, since the surrogate models will of course pick up incorrect data and then mislead the optimiser.

Two extremes are depicted here:

1) Figure 4 shows a random design at 20m/s and fully in the air at the final time step. In this time step the final drag value is of course very low, which is why all the quantities of interest need to be averaged, a window of 30% of the last time steps is used for the final runs. The time-dependant drag also shows a huge oscillation due to the periodic slamming of the hull, and it is visible in the motion variables as well. While the averaged resistance might not be that bad, passenger comfort likely suffers. Hence, calculating the relative standard deviation gives a very good impression of the behaviour and can be used to drive the optimiser. For the optimised samples in this work we use an arbitrary limit of 20% as a constraint.

2) Figure 5 gives an immense bow wave and spray at 5m/s, the total resistance is more than doubled compared to the base geometry. Since the free surface is captured via the AGR algorithm this also leads to largely increased mesh sizes and hence longer simulation times. On the other hand, resistance values will be far more accurate when compared to a static mesh approach, when not using a very conservative (large refinement zone around the free surface) mesh. Hence, over the course of all designs, displacements and speeds the AGR approach surely delivers the most efficient turnaround time.

Another physical property of the hulls is the time to reach a stable hydrodynamic position, if reached at all, and this translates directly into solver convergence and time dependent drag. A convergence checking tool, directly integrated into FINE™/Marine, allows to monitor time histories and stops the process when a given tolerance for e.g. the drag is reached.

Altogether, all those dynamic systems (physics, mesh, convergence) lead to a huge variation in solving time (say 20 minutes to 2 hours per Operating Point on 14 cores), but an efficient trade-off between accuracy and time is found for all those challenges.

3. Results

3.1 Database

The database step is a quite simple one, samples are generated using the defined bounds and the sampling scheme. Depending on maturity of the base design there might already be interesting (equals better) samples in the database, and a key point to check is the accuracy of the surrogate model used later in the optimiser. MINAMO provides this by means of a leave-on-out analysis, which gives the correlation (factor and distribution). Since the genetic algorithm purely relies on the surrogate model, a good correlation is important for a successful optimisation. Also, it can indicate when enough samples are generated to continue to the optimisation step.

A total of 32 samples have been generated in the database, which is on the safe side on most cases in terms of accuracy for ten free parameters. Figure 6 shows the correlation graphs for the averaged total resistance for 5m/s and 1000kg (left) and 20m/s and 1200kg (right). The correlation coefficient for all six drag values are in the range between 0.7 and 0.9, which is very good.

As is expected with the challenges explained above, scattering of the total resistance is quite high. Figure 7 shows the averaged drag at 20m/s over 5m/s, coloured by the value at 12.5m/s. Due to the axis definition all values are negative, and hence the closer to 0 (that is in the top right side) the better the design. The blue cross gives the starting point, that is the Riva Junior base design. Aside from the extreme designs there are already some good designs, which show substantial less drag.

3.2 Optimisation

In this work only one set of multi-objective optimisation run is shown: the Pareto type optimiser uses all six averaged drag values as objectives and the relative standard deviation of the drag is specified as constraint to be less than 20. Of course, there would be room for many more sets of objectives and constraints, but the following results already show quite good results overall. Vessel motion variables are also evaluated, but except from the flying/slamming hull type these values are not excessive and hence not deemed necessary as objectives. Also, the overall physical behaviour of the hull is well represented by just the time evolution of the drag, and its standard deviation.

Figure 8 shows the total resistance values, this time without a colormap for the values 12.5m/s. Blue dots indicate database samples, orange ones are coming from the optimiser, while small dots indicate non-feasibility (here: drag deviation constraint of 20% not met). Again, the results seem very good and indicate a successful optimiser run. There are no samples in the completely un-usable region, showing a very good prediction from the surrogate model – high scattering here would point in that direction. There is also only a small amount of non-feasible designs.

Figure 9 shows again a zoom and giving all resistance values per displacements. In total 56 samples were generated during the optimisation run, which is not that much regarding the complexity of the problem. Still there are already many samples that indicate the Pareto front, and by just judging from the penalties design number 79 (marked with a star) shows a very good overall performance. At a displacement of 1m³, resistance at 5m/s is reduced from 1160N (all values shown in the plots are for half ship models) to around 1000N, which is around 13%. At 12.5m/s, the reduction is from 1760N to 1560N, translating into 11%. The highest speed of 20m/s shows the greatest improvements, total resistance is decreased from 3630N to 2340N, hence 35%. This design can also keep these improvements at 1.2m³ and is hence seen as optimal in respect to all six objectives.

These figures also show another important advantage of the Pareto-type optimisation: there are several other candidates found which are probably not as balanced as design 79, but favour objectives in respect to others: this is the strength of a true multi-objective algorithm with a wide range of very good potential designs. Looking at 1m³: design 48 is slightly worse at 5m/s compared to number 79 and can almost keep the performance at 20m/s, but it shows another 6.5% decrease at 12.5m/s, hence this could be a hull shape for a light battery configuration with great endurance at already planing conditions (water-skiing comes into mind), and still has the potential for thrilling high-speed driving.

A final word on the overall turnaround time: a 28-core machine is used for the database and optimisation, running two operating points in parallel on 14 cores. The database took only a couple of days, and the optimisation was completed in around a week (pure CPU time), which is very fast considering the complexity and the small machine used.

3.3 Data Mining

The procedure shown here produces a lot of data, and so far only new designs and their performance improvements are discussed. But, using this data and applying so called data-mining tools can greatly increase the understanding of physics and the correlation to input parameters. Figure 10 shows one of those tools, the self-organising maps (SOM). They are also based on a surrogate model, and project high-order data (here a problem with ten input parameters and six objectives) into 2D space. Each point indicates a design of the database and optimisation loop, and it is fixed in space. The colour contours display the total resistance values for all three speeds (rows) and the two draughts (columns). Note that red means low resistance and blue a high one. There is a strong correlation between all six performance values on the top left side of the SOMs, and a very favourable area at full displacement speed (5m/s) on the bottom right. The planing modes are strongly focusing on the top right side, especially for 12.5m/s, while at 20m/s the bottom part is also attractive.

These plots clearly show that trade-offs must be made when aiming for a Jack of all trades design, and number 79 (marked by a star) is very good overall. There are samples which favour either of the operating modes, but at the cost of others. Again, the designer or engineer needs to decide what he or she wants.

Another useful tool is the analysis of variance (ANOVA). It is also based on the surrogate model and allows to calculate the sensitivity of all input parameters in respect to the output, here resistance values. Figure 11 gives the ANOVA plots for all six operating conditions in this study. The rocker parameter, which impacts the keel line towards the transom, is quite dominant for all planing conditions (12.5m/s and 20m/s), although a bit dependent on the displacement: it is far more impacting on the light version than on the heavier one. Total boat length on the other hand is a strong factor for the displacement speed, while completely irrelevant for high speeds.

ANOVA can hence guide a designer in a lot of ways: for example, it can show sensitivities which were probably not known before and need to be considered much more in the next project. This can be a change of parametrisation for finer control over important features and paving the way for even better designs. And the opposite might be true as well: less important features can be neglected, which in a workflow similar to the one presented here will save CPU time, or in the real world could also save material or costs in general.

4 Conclusions

An optimisation of a small powerboat has been presented, using numerical simulations. One focus lied on the application of electric propulsion instead of a combustion engine, which brought a few extra challenges with it: battery size is a crucial parameter and was respected using two different displacements. The full operating range, from low to high speeds was considered, to allow for a complete picture of such a powerboat. A powerful combination of efficient parametric modelling using CAESES® and state-of-the-art CFD solutions from NUMECA led to a very stable, yet cost-effective design process. The results of the optimisation run were quite promising, showing some major gains in terms of total boat resistance. A few designs were discussed in more detail, showing some trade-offs between the six operating conditions. A deeper look into the results and the vast amount of data was given via data-mining tools, that can easily show trends, correlations and anti-correlations in the design space. Also, sensitivity analysis methods were applied to highlight a few of the important geometrical features of such a boat.

  • Author

    Sven Albert, NUMECA Ingenieurbüro


HiFi-TURB Project

Studying the Nature of Turbulence with Neural Concept’s Deep Learning Platform

Turbulence | Deep Learning | Data Mining | Aerospace

The most significant challenge in all areas of applied fluid mechanics is posed by the lack of understanding and thus poor prediction capability of turbulence dependent features. This leads to limited industrial confidence in CFD for many aeronautical applications such as flow detachment over an aircraft wing or shock-boundary layer interactions. Against this background, the HiFi-TURB project, which is coordinated by Numeca, sets out a highly ambitious and innovative work programme to address influential deficiencies in turbulence modelling.

The large scale availability of High-Performance Computing (HPC) opens the door to a truly novel approach to turbulence model development. Exploiting Artificial Intelligence (AI) and Machine Learning (ML) techniques applied to a database of high-fidelity, scale-resolving simulations of test cases that contain most features of separated flow regions or complex 3D flows. Figure 1 shows an example of a flow field that is used as the basis for the turbulence modelling task.

The huge amount of data generated in these simulations requires a new approach to data mining. This is where Neural Concept?brings in its tool chain based on Deep-Learning, to analyse very large amounts of data provided by 3D scale resolving simulations.

Using Neural Concept’s Geometry-based Variational Auto-Encoders (VAE), NUMECA was able to gain first insights into correlations between tens of statistically averaged flow variables. The VAE compresses the data first in a physically meaningful way into so-called ‘embeddings’ and then reconstructs the original input from the compressed data. This is done to a very high accuracy, which allows to use the ML model as a replacement, a so-called surrogate, for the original data. The advantages are a much easier handling of the data, and the possibility of exploiting data mining and analysis techniques that help to understand the physics in the data.

Figure 2 shows an example of the possible analysis. The colours of the symbols on the 2D plot correspond to the value of the ‘embedding’ and are the same in the 3D view (left) and in the 2D plot (right). Points of the same colour have the same value for all the considered physical quantities and the 3D view coloured by the embedding value, gives us one global statistical representation for several physical quantities over the investigated domain. Both plots together provide a new perspective on the flow behaviour via the machine learning model. Figure 2, shows snapshots of the views used in the Graphical User Interface.



Dirk Wunsch, Head of the Robust Design, NUMECA International

Particle Counter

Particle counters and the COVID-19 pandemic

Particles Counter | Particle Tracking | Pressure-base Solver | Flowfield Physics

The COVID-19 pandemic has brought into sharp focus the need to understand transmission mechanisms of respiratory viruses. Understanding transmission mechanisms requires study on three broad fronts: identification of virus transmission paths; establishing how the virus circulates; and, experimental validation of transmission and circulation models. Prior to the spread of COVID-19, in preparation for an anticipated influenza pandemic, the scientific community has shown that the short-range aerosol route is an important, though often neglected, virus transmission path.

Industry trade associations are collaborating to develop guidelines on how aerosol particles circulate through a building. To aid in their work, the trade associations are studying data provided by research groups who work in collaboration with particle counter manufacturers. Particle counters, typically used to measure the number of airborne particles in cleanrooms, research labs, and operating rooms, are now emerging as a technology that can help in determining behavior of aerosols.

A challenge that particle counter manufacturers are addressing is the characterization of the aerosol particles that carry the COVID-19 virus. When an infected individual sneezes, coughs or breathes, aerosol particles of water are expelled. The virus travels suspended in these particles. Since these are water particles, they evaporate over time and, therefore, particle size changes. Hence, particle counter manufacturers are challenged to measure the number of particles, especially as those particles change size.

Accurately measuring the number and size of aerosol particles carrying the COVID-19 virus is critical to the validation of any transmission or circulation model. In response, Particles Plus®, an engineering and manufacturing company located in Stoughton, Massachusetts, sought to apply computational fluid dynamics (CFD) and single-particle tracking simulations the analysis of the flow-field within a particle counter. The company selected AGS Consulting, LLC to partner with, because of their reputation for product design and optimization. They selected NUMECA to perform the CFD analysis, because of their reputation for simulation tools that accurately predict real-world product performance.

Working collaboratively with AGS Consulting, LLC and NUMECA, Particles Plus® commenced development of their next generation of particle counters.


The numerical process adopted comprised two steps: first, modeling the particle counter flow-field that constitutes the particles’ support media; second, calculating the trajectory of particles as they pass through the particle counter.

Due to the complexity of the geometry, engineers at NUMECA chose to work with FINE™/Open with OpenLabs, using the unstructured meshing capabilities of Hexpress™. First, for comparison with experimental results, flow-field simulations were run in FINE™/Open to predict pressure-drop through the particle counter. Here a pressure-based solver was used, which is faster and more accurate for incompressible flow-field simulations.

Single particle tracking was used to predict the trajectory of different sized particles through the particle counter. Particles were launched from the inlet, and trajectories calculated as they passed through the particle counter chamber.

Two assumptions were made when calculating trajectory:

  • Particle trajectory is driven by the flow-field within the particle counter, but particles do not have a significant impact on the flow-field. The particle-to-air ratio is considered low enough for this one-way coupling approach to be assumed valid.
  • Particle-to-particle interaction is neglected. Again, the particle-to-air ratio is considered low enough for this approach to be assumed valid.

Simulation Results

Horizontal and vertical cutting planes through the particle counter were defined and particle trajectories mapped onto each plane. The trajectories of Class 1, 2 and 3 particles were studied over the vertical plane, with each class of particle behaving differently:

  • The Class 1 (5.0 micrometer) particles pass through the chamber.
  • The Class 3 (0.3 micrometer) particles recirculated in the vicinity of the chamber exit. However, they do not migrate back into the main body of the chamber.
  • The Class 2 (1.0 micrometer) particles both recirculated in the vicinity of the chamber exit and, critically, do migrate back into the main body of the chamber.

When particle trajectory was studied over the horizontal cutting plane, particle trajectories were concluded to be similar, with the exception of Class 2 particles. The Class 2 particles exhibited a more intense recirculation at the exit of the chamber. The intensity of this recirculation was tentatively concluded to be driving migration of Class 2 particles back into the main body of the chamber.


Overall, AGS Consulting, LLC and NUMECA collaborated with Particles Plus® to develop an experimentally validated CFD simulation of the flow-field within a particle counter. This simulation was used to identify that large, medium and small particles behave differently as they pass through the particle counter. The two-step transfer of 1.0 micrometer particles from inlet-jet to exit-recirculation and then exit-recirculation to main-chamber recirculation, was not anticipated.

This insight into particle counter flowfield physics, and the associated physical mechanisms at play within the particle counter, have provided Particles Plus® with the basis for an on-going project aimed at identifying critical particle sizes prone to recirculation, and the optimization of particle counter geometry to minimize that recirculation.



Geoff Sheard, President, AGS Consulting, LLC

Blower Optimisation

Increasing the Performance of Industrial Blowers with NUMECA Optimisation Solutions

Optimisation | Blower | Industrial Equipment

Petrochemical and other industrial applications rely on fans for a series of gas handling processes – which can range anywhere from ambient air ventilation to harmful chloride gases above 500C. The maximum operation speed of these fans is governed and often limited by industry standards. This speed limitation forces fan manufacturers to size the diameter of the impeller up or down in order to obtain the desired pressure output. The challenge for engineers is that the resulting design can become oversized, heavy, and expensive to fabricate, test and transport.

Illinois Blower, based in Cary, Illinois, has been working with NUMECA on a design approach that solves this dilemma by achieving higher pressures and increased fan efficiency, yet maintaining impeller speed and diameter. For more than 40 years, Illinois Blower has successfully developed and built custom centrifugal fans and blowers for a variety of worldwide industrial process industries, including refinery and petrochemical power generation, pollution control, pharmaceutical, food processing, and many others.

The goal of this particular case was to increase the pressure ratio of a complete fan stage (wheel and volute) over its entire performance line. Due to manufacturing constraints, the solid body thickness around the impeller had to be maintained, and the blade shape needed to be easy to manufacture. In addition to this impeller optimisation, the engineers wanted to get a better understanding of the flow physics to help reduce pressure losses in the outlet pipe.


The first series of simulations were run from choke to stall in FINE™/Open, NUMECA’s unstructured multi-purpose CFD solver package, and the results were compared with experimental data to assure that the CFD settings were reliable.
High-speed meshing
A 7.5M-point mesh with smooth boundary layers in the volute and around the impeller blade details was generated in Hexpress™. Only one passage of the impeller had to be meshed in order to run the performance line steady calculations, and each computation took less than 1 hour on 96 cores to achieve full convergence.

Optimisation Part 1: Identifying the Main Performance Parameters

The first step of the optimisation was to identify the main factors influencing the centrifugal fan’s performance, in order to know exactly where optimisation would be most effective. Twenty user-defined parameters of the impeller blade and flow channel were carefully selected, describing the hub and shroud shapes; the blade metal angles; and the blade camber and lean. The choice of these free parameters and associated variation ranges turned out to be key for the success of the project. For each set of the parameters, a new geometry was created by FINE™/Design3D, and an unstructured mesh was automatically generated in OMNIS™/Hexpress, using a dedicated python script to save time. After parameterization, the design of experiments (DOE) database was generated by FINE™/Design3D with the Minamo module. The main optimisation algorithm of the Minamo datamining tool is based on evolutionary algorithms, accelerated by the use of surrogate models to speed up the convergence rate. A database of 70 samples was built, filling the design space with 210 CFD solutions at three operating conditions (at stall, near design point, and at choke). By applying the grid-to-grid interpolation to improve the initialization of each CFD sample, a 25% reduction of iterations (and therefore CPU time) was achieved.

A thorough analysis of the database allowed the engineering team to understand the influence of each of the free parameters and their impact on the performance. It was found that the volute was the main limiter in the optimisation of the fan performance! Ensuing work decoupled the components to further optimize the impeller separately and to design a new volute that minimizes the pressure losses observed in the sample CFD solutions.

Optimisation Part 2: Focus on the Impeller

The second part of the performance optimisation focused on the impeller blade, meridional effects, and corresponding solid walls, decoupled from the volute. Using Autogrid™ and FINE™/Turbo, the large database and optimisation can easily be run over a weekend on a desktop machine, and the process is also fully automated by python scripts. Once a new optimal impeller was obtained, the performance of the entire fan was then computed by coupling the detailed geometry of the optimal impeller with the redesigned volute.


Challenge completed. Thanks to this two-part optimisation project, Illinois Blower managed to increase fan performance by up to 44% overall, while maintaining their original design constraints. The optimisation of the impeller blade shape and flow channel led to an increase of static pressure of up to 20% at some operating points (near choke). Furthermore, thanks to a better understanding of the flow-induced pressure losses downstream of the volute scroll, a smart redesign of the volute led to an additional increase in pressure over the whole performance line, going up to 24% at some operating points (near choke).


  • Dr Edward De Jesús Rivera, Engineering Manager, Illinois Blower, Inc.
  • Fanny Besem-Cordova, Application & Consulting Engineer, NUMECA-USA

Hydrogen Aircraft

AeroDelft pushes the airline industry towards a sustainable future with liquid hydrogen aircraft


AeroDelft is a student team at the forefront of sustainable aviation. While based in Delft, over forty students from different schools and universities around the Netherlands have joined together and are working hard to push the airline industry towards a sustainable future. AeroDelft has one main goal; to prove and promote liquid hydrogen as an alternative to conventional fuels in aviation. Air travel has been on the rise ever since it became a mainstream form of transportation, and it does not show any sign of slowing down.

“To preserve the world we live in there are two options: stop flying or revolutionise the way in which we fly. AeroDelft aims to play a key role within that revolution.”

Currently, the team is working on two projects simultaneously. In 2018 Project Phoenix was founded. The team is now working on the final stretch of the Phoenix prototype, aiming to have the first successful liquid hydrogen powered flights in the summer of 2021. In addition, this year the Phoenix Full-Scale project has begun. This project will have a longer timeframe but also has a major goal. By the year 2024 AeroDelft hopes to launch a full-scale liquid hydrogen powered aeroplane, which will be able to seat two people. From there on, the sky’s the limit!

Getting the aerodynamics right for minimum energy consumption

One of the main requirements for developing hydrogen powered aircraft is to get the aerodynamics exactly right in order to minimize energy consumption. This is where AeroDelft’s partnership with NUMECA comes in. To achieve the optimum aerodynamic design for their Phoenix prototype, AeroDelft uses OMNIS™ to be able to run many simulations and test a large variety of design alternatives until achieving the optimal result. The Open-DBS solver was used for calculations on cruise conditions with an inlet velocity of 23 m/s, cruise height of 100m, pressure of 101325 Pa and temperature of 15 degrees Celsius.

Meshing and set-up

To meet the need for preliminary lift and drag estimates, OMNIS™ was used to run four simulations with increasing mesh resolution. Limited hardware use and simulation time were critical requirements. Starting from a CAD file, three fully hexahedral meshes were created with OMNIS™/HEXPRESS and a simulation was run on each of them. The first mesh was not refined within the boundary layer. In further mesh refinement, viscous layers were inserted to model the physical boundary layer with increased accuracy using a wall function. Additionally, volume refinement was applied to investigate its influence on the solution. This is visible in Figure 1.
The geometry of the aircraft CAD file included many small surfaces that were not very relevant for this analysis. Resolving these surfaces would normally drive up the cell count considerably, but with HEXPRESS these regions can easily be meshed using a much larger cell size. A good mesh quality was achieved with basic meshing settings and without any special treatment or refinement in geometrically complex areas. A closer view of the surface of the aircraft and the mirror plane are visible in Figure 2. HEXPRESS was able to compute an 8.5 million cell mesh in 22 minutes on 10 cores using an Intel® Core™ i9-7920X CPU. A simulation on the same mesh and hardware was completed in 2 hours and 13 minutes using CPU Booster technology. This is the equivalent of 2.6 core-hrs per million cells.

Simulation and Results

Simulations were run using the k-wSST turbulence model. A 1:3 scale model of the aircraft was studied in cruise conditions. Half of the geometry was simulated using a spherical domain. The first mesh dependency study provided a clear overview on the importance of mesh volume refinement for aerodynamics simulations. Figure 3 shows the Mach number on the symmetry plane for the three meshes. The meshes with volume refinement clearly show better capturing of the aircraft wake than the coarse mesh without volume refinement.
Increasing the number of viscous layers had a significant impact on the predicted lift value. Such sensitivity studies are useful in identifying refinement areas that have the highest impact on the results for the lowest increase in computing cost. Table 1 shows an overview of the results. Furthermore, the pressure coefficient on the surface of the aircraft was calculated using the finest mesh as can be seen in Figure 4. Nondimensional quantities such as the pressure coefficient are very useful in scaling studies and comparisons with experimental results.
This project is an example of the fact that it is possible for student teams to perform numerical studies with a significant impact at a reasonable computational cost. Subsequent studies could explore the sensitivity of the results to further refinement at the leading/trailing edge of the wing and the vertical/horizontal stabilizer or the wake of the aircraft.


Sam Rutten, Prototype Project Manager, AeroDelft,
Olga Lubbers, Advisor, AeroDelft
Botond Pal, Application engineer, NUMECA International.

Quadcopter Drone

Extending quadcopter drone flight time and range with OMNIS™ CFD simulation

NLH Simulation| Drone | Quadcopter | Flight Conditions | Propeller | External Aero | Optimisation

Drones have proven to be an efficient solution for a large range of applications within the military, industrial, and private consumer domains. In the past decade, their use has been soaring and their annual growth rate is anticipated to exceed 50% [1].

There are two main categories of aerial drones: rotorcraft capable of vertical take-off and landing (VTOLs) and fixed-wing vehicles. Rotorcraft drones offer important advantages over the fixed-wing systems, such as their ability to hover (maintain a constant altitude) and the fact that they are easier to control and operate. This makes them particularly suitable for unique applications like indoors operations, maneuvering in wind turbines and construction site inspections to name a few. On the other hand, multicopters also have inherent shortcomings, the most important one being their limited flight time and range. Even modern and innovative electric drones have a limited flight time of around 20-30 minutes depending on flight conditions. Only very few conventional electric multicopter drones in the high-end class, can reach flight times close to 1 hour. The application of Computer-Aided Engineering (CAE) techniques and Computational Fluid Dynamics (CFD) in particular can help to significantly improve the efficiency of drones and extend their flight time and range.

NUMECA’s fully integrated multiphysics CFD environment OMNIS™ paves the way for faster, more accurate quadcopter simulations, combining structured and unstructured meshing solutions with the fastest CFD solvers on the market within one platform. Its Nonlinear Harmonic model (NLH) has proven to be a cost-effective solution for unsteady simulations up to two orders of magnitude faster than the conventional methods. The case study presented in this article demonstrates the aerodynamic simulation and optimisation of an industrial quadcopter drone in hover mode, the most energy intensive mode of this type of drone.


The studied geometry (Figure 1) corresponds to the most widely used rotorcraft drone configuration today: the quadcopter geometry. The model used here was provided by the authors of [4]. Drone manufacturers for the private consumer sector (amateur video shooting, racing drones, drones for kids, etc) predominantly rely on this type of configuration.

The propeller blade (Figure 2) was modelled with NUMECA parametric modelers, taking into account the required thrust. Multiple sections were extracted from the original geometry and stacked together in order to build the 3D blade. An appropriate twist distribution was provided to make sure the parametrised blade resembled the original geometry as close as possible.

The setup benefited from the symmetry of this drone geometry: only one-fourth of the drone needed to be included in the computational domain, hence only one arm. The chosen domain definition represented a practical case, corresponding to “free air” simulation at a hovering altitude that is high enough to neglect any ground effect.


Due to the complexity of the drone domain, an automatic unstructured mesh was generated using Hexpress (Figure 4). Hexpress automatically refines the mesh near high curvature areas and edges, thus minimising user interaction and engineering time. This leads to a high-quality mesh, sufficiently robust to be used for an optimisation.

One blade of the propeller was meshed using multiblock structured mesh generator AutoGrid (Figure 5). AutoGrid uses a wizard-type approach for meshing various types of turbomachinery configurations with different characteristics, such as centrifugal pumps, axial compressors, etc. This approach makes it very easy and fast to generate a high-quality structured mesh with multiple grid levels. A variable tip gap was applied to the blade. Also, a matching periodic connection between two periodic faces was automatically ensured and computed. Meshing only one blade, combined with such a matching connection, led to a two-fold cell count reduction with the corresponding simulation speed-up.

Both meshes were then assembled together and a rotor-stator interface was set up between the two domains. It is worth noting that OMNIS™ allows the user to combine and run structured and unstructured meshes in the same computation, taking advantage of the intrinsic speed advantage of using structured meshes and of the robustness of the unstructured ones. This also reduces RAM and disk consumption. The approach does not require tuning any solver settings.


Open is capable of both steady and unsteady Non-Linear Harmonic (NLH) [5] simulations. The propeller was set to rotate at 5,000 RPM, whereas the drone arm is stationary. The Spalart–Allmaras model was used to predict turbulence in the flow. For the steady simulation, a mixing-plane interface was used, while for the NLH simulation, a specific treatment based on Fourier decomposition was applied. This provides the benefits of a domain scaling approach with a computational cost similar to that of the mixing plane.

The Non-Linear Harmonic method provides unsteady flow results with considerably less constraints than the domain scaling and phase-lagged methods. For this project, one harmonic per domain was added to capture the unsteady perturbation in the domain.

Comparison of the results (Figure 6 and 7) obtained in the framework of steady and unsteady simulation, revealed the presence of strong unsteady features in the flowfield. The pressure distribution on the airframe was largely impacted by the instantaneous position of the propeller. The velocity field around the drone was subject to strong periodic oscillations linked to the rotor rotation. The results comparison shows that steady simulation can provide sufficient representation of the mean flow field. However, the NLH analysis provides accurate information on the unsteadiness of the flow field, offering a large scope of valuable data for an engineer in terms of unsteady flow physics, blade and airframe loading, as well as blade tip vortex and bluff body recirculation dynamics, at a cost comparable to a steady simulation.

Drone Design Optimisation

NUMECA software offers multiple possibilities for design parametrisation and optimisation. The available optimisation methods range from single-objective optimisation to multi-objective and robust design optimisation (RDO) that takes into account operational and manufacturing uncertainties. A drone optimisation process can benefit from all these methods. The final choice of the technique depends mainly on the expected operation modes.

Figure 8 represent an example of geometry parametrisation and its variation in the framework of the optimisation study. The propeller geometry used for the optimisation study was parameterised on CAD model level. The angle of attack of the propeller at 3 spanwise sections was taken as the design variable. Each geometry was automatically re-meshed in AutoGrid. The drone arm was parameterised using 3 morphing vectors placed in the geometry. They allow the optimiser to optimise the shape of the drone using the morphing technique, while also satisfying multiple constraints that are applied to ensure feasible designs.

NUMECAS’s optimisation routines are based on gradient-free algorithms that are considered to be much more efficient than gradient-based optimisation for complex multi-component systems such as drones. The employed optimisation processes benefit from a great speed-up thanks to the use of built-in surrogate models or artificial neural networks. Underlying evolutionary and genetic algorithms ensure an optimum converged solution in terms of defined objectives, such as flight time maximisation. Practical use of such algorithms confirms that it can lead to novel, innovative, and sometimes even unexpected optimum system designs.


Simulation technologies have become a key element of drone design. The rapidly growing and highly competitive market, drives commercial drone manufacturers to improve efficiency and expand the flight envelope and range of their applications. Maximum flight time and range remain to be one of the most important issues to address for multicopter electric drones. NUMECA’s CAE suite, with OMNIS™ at its core, offers an efficient and cost-effective solution for the simulation and optimisation of drones.

The presented case study demonstrates a set of powerful capabilities, such as the combination of structured and unstructured meshing techniques and high-fidelity unsteady simulations using the Nonlinear Harmonic Method. Fully automated optimisation, based on efficient evolutionary algorithms, parameterisation and morphing, ensure a fast and robust workflow and an optimum design result for the defined objectives, such as maximising flight time and range.

Click here for a full white paper download about this case.


[1] Commercial Drone Market Size, Share & Trends Analysis Report By Application (Filming & Photography, Inspection & Maintenance), By Product (Fixed-wing, Rotary Blade Hybrid), By End Use, And Segment Forecasts, 2019 – 2025. Jun, 2019. URL: Accessed 10 May 2021

[2] Unmanned Aircraft Systems & Advanced Air Mobility. URL: Accessed 10 May 2021

[3] DJI FPV – Specs. URL: Accessed 10 May 2021

[4] The quadcopter geometry was described in the paper “Weerasinghe S.R. and Monasor M., Simulation and experimental analysis of hovering and flight of a quadrotor, 13th International Conference on Heat Transfer, Fluid Mechanics and Thermodynamics (HEFAT 2017), Spain 2017” and was kindly provided by its authors. The CAD model was generated by Miguel Monasor Pascual, Mechanical and Aerospace Engineer. Compared to the provided geometry, a new propeller geometry was generated and is retained for the study.

[5] Vilmin S., Lorrain E., Hirsch C., Swoboda M., Unsteady Flow Modeling Across The Rotor/Stator Interface Using The Nonlinear Harmonic Method, ASME Paper, GT-2006-90210, 2006

RDO Propeller

Lowering fuel consumption through Robust Design Optimization (RDO) of ship propellers

Marine | Propeller | Robust Design | Uncertainty Quantification | Optimisation

Optimising the design of ship propellers is of key importance for the marine industry, as it has a direct impact on operational costs through its influence on fuel consumption and ship performance. Even the smallest improvement in the design of the propellers can save ship operators millions of dollars across an entire fleet. The same is true, be it at a smaller scale, for the super yacht and pleasure boat industries for example.

Today propeller designs can be simulated meticulously with powerful tools that take into account a broad range of physics, from fluid flow to structural integrity to even acoustics behavior. And those designs can be optimized for optimal performance in a fully automated way.

The question that arises though, is whether we are sure that this ‘optimal’ design corresponds to the optimal design under realworld conditions, where uncertainties are an inevitable part of life. Uncertainties for example that are embedded in operating conditions such as ship speed and rotational speed of the shaft, or uncertainties in the manufacturing process, which lead to geometrical variations in the propeller shape that can have a significant impact on the final performance of the propellers.

To find a response to this question, robust optimisation should be considered. Robust Design Optimisation (RDO) takes into account a series of these uncertainties that can influence the performance of products. It allows for designs to be optimised in a ‘robust’ way, for example by making them less sensitive to inevitable variations in operating conditions or to small differences in geometries due to manufacturing variability.

To illustrate this, the below described customer case presents the optimisation under uncertainties of a ducted ship propeller. The goal of this case was to reduce the impact of manufacturing variability on the ship’s open water efficiency.


Two optimisation studies of a ducted ship propeller were performed. First, a standard (deterministic) design optimisation was used to maximise the open water efficiency of the propeller. Then, in a second step, robust design optimisation was performed to maximise the mean value of the open water efficiency and minimise standard deviation of this efficiency.

Uncertainty Quantification

A total of 12 uncertainties were identified and characterised for this case: 11 manufacturing uncertainties and 1 operational uncertainty (axial velocity). The manufacturing uncertainties were deduced from the technical norm ISO-484-2, which specifies all the manufacturing tolerances for this particular case. The manufacturing variability of every propeller has to respect these imposed manufacturing tolerances.

Our optimisation software FINE™/Design3D is equipped with a unique Uncertainty Quantification (UQ) analysis module. It enables designers to easily assess the effects of uncertainties on performance by carrying out several CFD simulations in a fully automated way. In this study the 12 defined uncertainties resulted in 35 individual CFD simulations.

Scaled sensitivities

To gain physical insight into the influence of uncertainties on performance and to reduce the computational cost, Cadence developed an intelligent post-processing tool, called “scaled sensitivities”. This tool measures the sensitivity of a performance, in this case open water efficiency, to specific uncertainties. Exploiting these sensitivity derivatives enables designers to reduce the number of uncertainties to be analysed to the minimum relevant ones, and thus minimise computational cost. For this study Figure 1 shows that open water efficiency is very sensitive to propeller chord length and ship velocity, while thickness plays only a minor role. Knowing this allows us to merge all thickness sections together for this study, forming one single uncertainty control for the thickness at 70% of the span height. This enables us to reduce the total number of uncertainties to be taken into account from 12 to 5, even though the total impact of all uncertainties is maintained in the analysis.

Deterministic vs robust optimal design

UQ analysis was performed on the standard deterministic design, in order to be able to compare its results with the robust optimal version. Figure 2 shows a characteristic Pareto plot where the standard deviation of the open water efficiency is shown over its mean value.

Standard design optimisation

The baseline design that the study started from is indicated in the graph by the red dot and the UQ results of the standard optimal design (not taking into account uncertainties) by the blue square. The plot shows that the open water efficiency has increased quite significantly by 8.5%, but that there is also an increase of 2.6% in its standard deviation. This means that this design is slightly more sensitive to the influence of manufacturing variability and axial velocity than the original design. »

Robust design optimisation

Robust optimal design 3 plotted in fig. 2 shows that the mean value of the open water efficiency increased by the same amount as for the standard optimal design, namely 8.5%. However, its variability is reduced by -17.7%. That means that this design is less sensitive to the influence of manufacturing variability and axial velocity than the original design and that it provides more stable performances.

Propeller shape

Figure 3 compares the shapes of the standard optimised design and the best robust optimised result with the original propeller shape. Even though performance is the same for both designs, their shapes are significantly different!


Robust Design Optimisation enables engineers to create designs that are less sensitive to existing and unavoidable manufacturing and operational variabilities. Comparing standard and robust design optimisation clearly showed that a comparable performance increase can be achieved with both strategies in this marine propeller case, but only the robust optimisation allows for reduction of performance variability, making it less sensitive to uncertainties originating from the manufacturing process or from operational variability.


Dirk Wunsch, Head of Robust Deisgn Group at NUMECA International

Noise Reduction Aircraft

Pipistrel Mitigates Aviation Noise Emissions of DEP Systems for Electric Aircraft

Aircraft | Noise Emissions | Pressure Distribution | NLH

Pipistrel is a world-leading designer and manufacturer of small aircrafts specializing in energy-efficient and affordable high-performance aircraft. The first company to fly an electric two-seater in 2007 and the winner of the NASA Green Flight Challenge in 2011 (with the world’s first electric four-seat airplane), Pipistrel has designed nine different experimental and serially-produced electric aircrafts, including the first type certified electric airplane, the Velis Electro. Pipistrel has also developed several propulsion systems, including batteries, power controllers and electric motors, for small and general aviation class of aircraft for NASA and Siemens, among others. With involvement in various standardization committees and research projects, Pipistrel is helping to enable the future market of hybrid-electric aviation.

The ARTEM project and Distributed Electric Propulsion

One such research project that Pipistrel is participating in is the Aircraft noise Reduction Technologies and related Environmental iMpact (ARTEM) project. The ARTEM project’s objective is to develop novel technologies to reduce aviation noise emissions for the engines and airframes of future aircrafts. Pipistrel is leading the investigation and mitigation of noise emissions of Distributed Electric Propulsion (DEP) systems. DEP systems, which are gaining popularity in the age of electric aviation, use multiple propulsion units distributed about the airframe. Since they are only connected electrically to energy sources or power-generating devices, the propulsion units can be placed, sized, and operated with greater flexibility and provide improved performance over more traditional designs. Some of the main benefits of this propulsion system are shorter take-off and landing, reduced energy consumption for increased flight ranges, and noise reduction. The technology offers numerous integration capabilities for a wide number of future aircraft concepts.

For their role in the ARTEM project, Pipistrel designed and manufactured a DEP mock-up, which serves as a validation platform for various new methods, designs, and low-noise technologies. Its aerodynamic design was based on FINE™/Open solvers using HEXPRESS™/Hybrid’s grid generator. For fast aerodynamic design, an actuator disc approximation was used, while for more refined design, steady and unsteady solvers were employed, as shown in Figure 1.

Time-Domain Navier Stokes versus Non-Linear Harmonic methods

Pipistrel and its project partners are studying noise source and propagation in the near- and far-field of the DEP design, but the main drawback of using Navier-Stokes CFD for the computation of input data for noise propagation codes (from the designer point-of-view), is its high computational cost. Time-domain (TD) Navier-Stokes simulation of the DEP’s periodic flows requires long runtimes, as several rotor revolutions need to be computed before achieving periodic state.

Figure 2 shows the evolution of the thrust of one propeller blade over one revolution by both TD and NLH methods. The periodic unsteady perturbations linked to the interaction between the propeller blades and the wing are modeled using Fourier harmonics, whose frequencies are associated to the periodicity and to the relative rotational speed between these two components. From a theoretical viewpoint, an infinite number of harmonics would lead to a perfect match with the traditional time-domain solution. From an engineering viewpoint, the NLH method has proven to satisfactorily capture the same complex unsteady flow phenomena as the TD method, using as few as three complex harmonics.

Figures 3 and 4 compare the pressure distribution on the propeller blade surface for the TD and NLH methods when the blade is aligned with the wing leading edge. For pressure distribution only negligible differences can be noticed on both sides of the blade.


FINE™/Open’s NLH method proved to be the most cost effective approach for acoustic optimization analysis of the DEP set-up. The method offered a speed-up between one to two orders of magnitude in comparison to the TD analysis, while providing very similar accuracy.

NLH technology offers a fast alternative to conventional TD methods for solving unsteady periodic flows and can easily fit into an industrial workflow.


Dr Jernej Drofelnik, Aerodynamics Design Engineer, Pipistrel Vertical Solutions