F.A.Q.

Content

1. What are the required computer specifications?
2. Are the designs manufacturable?
3. Can the software only be used for electronic cooling systems?
4. How are the designs created on Diabatix Coldstream?
5. How can the resolution of the mesh affect the quality of the simulation results?
6. How can the resolution of the mesh affect the design?
7. Why does my design case take so long to complete?
8. Can I run supersonic cases on Diabatix Coldstream?
9. Can Diabatix Coldstream handle external aerodynamic optimization problems?
10. How is turbulence taken into account on Diabatix Coldstream?
11. What turbulence model should I choose?
12. Is it possible to start from an existing design and modify it?
13. How many objectives can be used in a single design optimization?
14. Can I restart an optimization from the same point after stopping it?
15. How does the fanInlet/fanOutlet boundary patch work? Can I set my own fan?
16. The mean outlet temperature is too low, can I trust the simulation results?

---

What are the required computer specifications?

Just an internet connection. You will be able to access the platform and run your optimizations via your internet browser. The software platform runs on secure HPC resources that we provide. We do not expect our clients to have the computing power available in-house. For postprocessing, a discrete GPU is recommended. This will make the 3D visualization of your results easier. You can both store the results on the platform and download them to your storage. Be aware of the fact that you need to have enough storage space if you would like to download the results.

Are the designs manufacturable?

Yes, the designs are manufacturable. Manufacturing constraints can be included in the case setup. The software supports various manufacturing techniques, and constraints depend on the technique. It is important to note that the manufacturing constraints are defined mathematically in our software, and as engineers, we recommend critically appraising the resulting design, since minor errors may occur. You can set constraints for various manufacturing techniques like CNC milling, sheet metal forming, 3D printing, die casting, stamping, injection molding, and much more.

Can the software only be used for electronic cooling systems?

No, the software is not restricted to just electronics or even thermal problems. It can design components for any application that requires thermal management or flow distribution. Anything that produces heat or that involves flow is within the software’s scope. In summary, all thermal and flow problems can be solved. Industries that are using our software include but are not limited to automotive, laser technology, aerospace, home appliances, medical imaging, etc.

How are the designs created on Diabatix Coldstream?

The designs are created using a topology optimization approach. In topology optimization, contrary to shape optimization, there is no need for an initial design input. This allows for a wider search of the design space, thus increasing the probability of finding better-performing and more innovative designs.

How can the resolution of the mesh affect the quality of the simulation results?

In CFD analysis, the accuracy of the simulation is closely related to the resolution of the mesh. Broadly speaking, smaller mesh elements (higher resolution) can capture smaller physical phenomena, thus increasing the accuracy of the simulation. On the other hand, simulations with a higher resolution will require more computational resources (cores, memory, time).

In Diabatix Coldstream, once a base resolution is chosen, the mesh will be automatically generated, taking into account not only physical phenomena but also geometric inputs to provide the best possible ratio of quality to computational cost.

How can the resolution of the mesh affect the design?

It is easy to extend the logic of the previous FAQ to the design creation. A better mesh generates better data that is used to generate more accurate designs.

It should also be noted that the resolution of the mesh will determine the minimum allowable feature size, meaning that the smallest quantity of material that can be added or removed depends on the size of the cells in the design region.

To meet the user input, Diabatix Coldstream will automatically define the resolution of the design space based on the minimum feature criteria entered.

Why does my design case take so long to complete?

The duration of the design run depends on the inputs of the case. If the design region is very big and one selects the smallest feature size, Coldstream will automatically generate a mesh with a very high resolution. Coldstream does this to make sure the user input is followed. It is very important to have a good feature size to design region ratio. You can contact the support engineers to give you more guidance on which feature size to select for your design case.

This ratio in combination with the amount of credits selected, can result in a very demanding CFD problem with the resources available. This translates to a longer running time. In conclusion, assigning more credits to a design case will decrease the computational time.

Can I run supersonic cases on Diabatix Coldstream?

Diabatix Coldstream is based on an incompressible flow solver, therefore, at the moment supersonic flows are not allowed on the platform. Liquids are generally accepted as being incompressible, but this is not the case for gases. A good rule of thumb to take into account when setting a forced or mixed convection case is that the Mach Number should be < 0.3 ( ~110 m/s for air at standard sea level conditions). If Mach > 0.3, the compressibility effects cannot be ignored and the results of the simulation will not be accurate.

Can Diabatix Coldstream handle external aerodynamic optimization problems?

Diabatix Coldstream can do external aerodynamic optimizations, however this is not its main forte. Our Diabatix sales and support engineers can advise you on how to set up each specific case.

A possible approach can be the goal to reduce the drag forces, create a “wind tunnel” and set the objective to "powerLossMinimization" between the inlet and outlet. Since Diabatix Coldstream uses a topology optimization approach, the design region has to be smartly chosen to allow not only addition but also removal of material from the object being optimized.

How is turbulence taken into account on Diabatix Coldstream?

Turbulence is the chaotic movement of the fluid. This chaotic movement happens on different length scales throughout the entire fluid domain, in turbulence, we speak about "eddies" when talking about this chaotic movement. There are thus eddies of different length scales. Turbulence models, as the name suggests, try to model the turbulent behavior of fluids in motion while maintaining the computational cost at an acceptable level. An oversimplified explanation is that turbulence models mimic the very small details of the flow that cannot be captured by the mesh.

If the computational resources allow, a DNS (Direct Numerical Simulation) approach will resolve all the details of the flow, and there is no need for a turbulence model. These simulations are normally restricted to simple academic problems and are not feasible for industrial simulations.

For slightly more complex simulations, normally restricted to one region (e.g. aircraft components, flow in a pipe, …), the turbulence can be modeled using LES (Large Eddy Simulation) models, which as the name suggests resolve the largest eddies while modeling the smallest features (on both length and time scales). These models can be seen as a filter, and achieving good results requires very specific meshes that can become prohibitively expensive to simulate. Therefore, for industrial applications, they are reserved for very sensitive flow problems, where the turbulent behavior of the flow plays a major role (e.g. aeroacoustics).

Besides the LES models, there are the RANS (Reynolds-Averaged Navier-Stokes) models which are the industry standard. These models, as the name indicates, take an average of the full spectrum of fluctuations. As the turbulent behavior is fully modeled, the mesh can be coarser than for the LES simulations. This allows to focus the computational power on problem scale and complexity (e.g. Multiphysics, full industrial machinery, combustion, ...)

Both LES and RANS models are extensively covered by the literature and validated against experimental tests with very good results.

Diabatix Coldstream has several different RANS models available:

• KOmegaSST
• KOmega
• kEpsilon
• RNGKepsilon
• Laminar
  ‍

What turbulence model should I choose?

The turbulence model selection depends on the problem at hand. As a rule of thumb:

• kEpsilon - One of the most commonly used models, it tends to better predict the flow away from the walls and tends to be quite robust. It is mainly used for external aerodynamic flows and fully turbulent flows.
• kOmega - Tends to perform better close to walls and for low Reynolds flows than the kEpsilon model, and can predict separation.
• RNGkEpsilon - a variation of the kEpsilon model with a mathematical approach that attempts to take into account different length scales to model the eddy viscosity, instead of a single one.
• kOmegaSST - This model is a variation of the kOmega model and can be seen as a hybrid between the kOmega and the kEpsilon model, taking into account the best of both, uses kOmega near the walls and switches to kEpsilon away from the walls. It is the default turbulence model used in Diabatix Coldstream.
• Laminar - for very low Reynolds flows or purely laminar flows. A low Reynolds number means that viscous forces are much more important than inertial forces. The high viscosity can dampen out the chaotic movement, so there is no turbulent behavior in the fluid.‍

Is it possible to start from an existing design and modify it?

Diabatix Coldstream uses a topology optimization approach, meaning it can transform one material into another, thus creating a new design. For each design region, you have the option to upload an initial design. You can ask ColdStream to perform any of the following three options with this initial design:

1. Only remove: you strictly want to remove material from the initial design. This is a highly robust optimization mode as well as a very fast mode.
2. Only add: you strictly want to add material to the existing design. This is the complementary mode to the 'only remove' option. This is also a highly robust optimization mode.
3. Add and remove: this is a combination of the two previous options. ColdStream has complete flexibility in removing or adding material as it sees fit. This mode is slightly slower compared to the previous two.

How many objectives can be used in a single design optimization?

The number of objectives is not limited to 1, so multiple objectives are possible. To control the relative importance of the objectives, the weighting factor can be set to a value in the range [0,1]. The values of the different weighting factors do not have to add up to 1, this relative weighting is done automatically by Diabatix Coldstream.

Can I restart an optimization from the same point after stopping it?

No, once the simulation is stopped it cannot be restarted from the point where it was killed. A new optimization should be submitted. If you press the 'Stop calculating'-button, all data is irretrievably deleted.

How does the fanInlet/fanOutlet boundary patch work? Can I set my fan?

The fanInlet and fanOutlet are boundary conditions that model the behavior of a fan, meaning that they alter the momentum of the flow based on a performance curve.

Diabatix Coldstream provides the user with a large library of performance curves for fans currently on the market, however, the user is free to create their fan. To do this, a set of at least 2 pairs of points representing volumetric flow (m^3/s) and total pressure loss (Pa) should be provided. Please note that the units should be in S.I. and that the total pressure should be used and not the static pressure.

The mean outlet temperature is too low, can I trust the simulation results?

For every finished case, ColdStream will output all the key performance values to the tables tab. One of these automatically generated tables displays the maximum, mean, and minimum temperatures of each entity. These values are generated as follows:

• maximum temperature ($T_{max}$): mesh element with the highest temperature

• minimum temperature ($T_{min}$): mesh element with the lowest temperature

• mean temperature ($T_{mean}$): surface/volume average over the entire entity

$T_{mean} = \frac{1}{S}\int_S{T\space dS}$

or

$T_{mean} = \frac{1}{V}\int_S{T\space dV}$

Where:

• S is the surface of an entity if that entity is a boundary

• V is the volume of an entity if that entity is a region

• T is the temperature distribution of an entity

From thermodynamics, we know that the mean outlet temperature of the coolant can be estimated using the following equation:

$T_{mean} = \frac{\dot{Q}}{\dot{m}\space c_p}+T_{inlet}$

Where:

• $\dot{Q}$ is the amount of heat transferred to the fluid

• $\dot{m}$ is the mass flow rate of the coolant

• $c_p$ is the specific heat capacity of the coolant

• $T_{inlet}$ is the inlet temperature of the coolant

In some cases, there is a small difference between the mean temperature reported in the tables and the mean temperature calculated using the last equation. This might raise questions about whether or not these results can be trusted.
The cause of this difference lies in how well the heat has mixed with the bulk fluid region. Some heatsink shapes don’t allow for good mixing, skewing what ColdStream shows in the table view. A more accurate representation would be:

$T_{mean} = \frac{\int_S{\rho\space c_p\space T\space\vec{U}\cdot\vec{n}\space dS}}{\int_S{\rho\space c_p\space\vec{U}\cdot\vec{n}\space dS}}$

Where:

• $\rho$ represents the density of the coolant

• $c_p$ represents the specific heat capacity of the coolant

• $\vec{U}$ represents the local velocity of the coolant on the outlet

• $\vec{n}$ represents the normal on the outlet

• $T$ is the temperature distribution on the outlet

The above equation is thus a measure of the amount of energy carried over to the coolant independent of the mixing. This measure is available in ColdStream in the ‘imbalance’ table. The h_relImbalance value for the fluid is a measure of all the energy carried to and from the coolant. The closer this value to 0, the better the case has converged.

👍

In short

Yes, you can trust these results, as long as the h_relImbalance on the fluid is within acceptable margins.