Characterization Converts 3D CFD Simulation into a 1D System Model Component

Integrating slower FloEFD simulations into a fast FloMASTER system model without co-simulation.

3D CFD simulations are accurate, but too slow. Co-simulation is too time consuming to set up early in development. 3D characterizations into 1D simulations is just right. (Image courtesy of Mentor Graphics.)

3D CFD simulations are accurate, but too slow. Co-simulation is too time consuming to set up early in development. 3D characterizations into 1D simulations is just right. (Image courtesy of Mentor Graphics.)

Optimizing a massive fluid system, like an engine cooling system, can be a considerable task for even the most seasoned engineers.

Perhaps two of the most powerful tools engineers have at their disposal to design these fluidic systems are 1D and 3D computational fluid dynamics (CFD) software. By combining these tools into one simulation, engineers can gain the benefits of both speed and accuracy.

3D models provide the accuracy needed to optimize a system part-by-part. However, they are often powerless to optimize the big picture without massive compute times and computational resources.

This is where system level modeling comes in. These 1D simulations simplify each part into equations, charts and/or response surface models (RSM). These 1D models speed up the simulation of the whole fluidic system allowing for a more complete voyage through the design space.

Unfortunately, 1D simulations may not have the accuracy an engineer needs to get a clear idea of what is happening within each part. And if that part is the bottleneck or point of interest needing optimization then there is a problem.

“When you bring 1D and 3D together, the 1D simulation provides an increase of speed and reduction of computation resources,” Doug Kolak, technical marketing engineer at Mentor Graphics (a Siemens company), said. “At the same time, 3D simulations provide improved accuracy that can mirror a system without empirical data. If you mix the two correctly, you get just the right level of accuracy and speed to make the important design decisions in the appropriate place in the design process.”

By characterizing 3D simulations into 1D simulations, engineers can democratize their analysis and improve the saleability of their simulations. This frees them from the worries of expensive HPC and quick deadlines early in the development cycle of a complex fluidic system.

1D/3D Co-simulations vs. 3D Characterizations into 1D

This 3D model may be accurate, but it’s also calculated slowly. Characterizing it in a 1D model using an RSM will speed up system simulations for product optimizations. (Image courtesy of Mentor Graphics.)

This 3D model may be accurate, but it’s also calculated slowly. Characterizing it in a 1D model using an RSM will speed up system simulations for product optimizations. (Image courtesy of Mentor Graphics.)

Some might think that the best way to get the benefits of both 1D and 3D is through co-simulations. They just hook up the inputs and outputs of their 1D and the inputs and outputs of the 3D models and venture forth into that unknown design space.

The problem is, these coupled simulations are only as fast as the 3D model. This limits the engineer’s voyage into the design space to a close orbit at best.

“1D/3D co-simulation also takes a lot of work to set up,” Kolak said. “You historically do this later in the design cycle. It takes so much overhead that you don’t want to do it if the models are not matured. With characterization, you can push a lot of this early in the development.”

The reason co-simulations are so useful later in the development cycle is that they can help bring a part through the verification and validation step of an otherwise well-known system. This calls for high levels of accuracy for the targeted component. However, the accuracy of the remainder of the system can remain at the system level as it is mostly just cycling boundary conditions to the target part.

Co-simulation is also useful when a part is inherently 3D, while the remainder of the system is accurately simulated with 1D. For example, let’s say an engineer were to design new gaspers in an airplane environmental control system (ECS).

“Here you are more interested in how the air comes from the gaspers whereas the fans and ducting can all be 1D,” Kolak suggested. “Modeling the air as it comes into the cabin is a 3D problem so you want to see the 3D reactions here. You can get detailed 3D results from the simulation, but what is affecting those results at the boundary conditions is the flow in the vents. You replace those parts of the simulation with 1D.”

The truth of the matter is that much of the applications one would use co-simulation for are far too slow for early development design space exploration. This design space can be explored much faster, without sacrificing accuracy, by converting the 3D simulation into a 1D model using characterization tools like RSM.

“Once you run the 3D characterization, you are free to run system level trade-offs at the speed of 1D early in the design process. You don’t have to run one hundred slow 3D simulations at the same time as one hundred 1D simulations,” Kolak noted. “This gives engineers the ability to better define large design spaces.”

Some programs that offer this feature are Mentor Graphics’ 3D CFD tool FloEFD and the company’s 1D system level CFD tool FloMASTER.

Characterizing 3D Simulations into 1D Models

The RSM created in FloEFD that will define the behavior of the N-arm component in FloMASTER. (Image courtesy of Mentor Graphics.)

The RSM created in FloEFD that will define the behavior of the N-arm component in FloMASTER. (Image courtesy of Mentor Graphics.)

So, how does one take a slow 3D CFD simulation and turn it into a quick yet accurate 1D model? Well there are few ways this can be done.

First, let’s say we have a well-defined component with enough information to build the part from scratch in FloMASTER.

It will take a considerable amount of time creating the analytical model and code that defines the inputs and outputs of the part, but what the user gains is control. After some intense coding, the user can input some matrix coefficients and publish the model. Unfortunately, for many, this method is a large barrier of entry into the 1D modeling world.

Instead, Kolak suggests that users define a new N-arm component. This will help them to avoid much of the of coding. An N-arm is a user-defined system part. It gives engineers the ability to define a part’s characteristics and number of arms (assuming there are more than three arms).

To define the N-arm component, all you do is input whatever empirical data and equations you have that define the pressure drop and temperature distributions, based on one reference arm to all of the other arms, into the part’s code. From here, the N-arm element does all the math for you.

“When you write the N-arm part script, you can add up to five curves and surface-maps to the characterization,” Kolak said. “Inside the script, you then have the option to reference this additional data and curves.”

Unfortunately, engineers don’t always have the luxury of a well-defined component with empirical data. “For 1D CFD simulations, the challenge is finding empirical data to describe each component,” Kolak said. “There are several sources for loss and thermal data, but they don’t have every possible design to be considered.”

This is where 3D CFD simulations, like those performed in FloEFD, becomes a necessity. Once the part is modeled in FloEFD to the engineer’s satisfaction, the engineer can run multiple simulations defined by a design of experiment (DoE). The results from the DoE are used to create the RSM that characterize the part.

“The idea of 1D/3D characterization is based on RSM,” Kolak said. “Creating an RSM for your simulation from a DoE is a fairly automated process. The user specifies inputs, boundary conditions and the design space using minimum and maximum parameter settings. They then select the number of experiments to run for a design of experiments (DoE). FloEFD then calculates the best-fit RSMs for the simulation data based on the lowest fit error. When it’s all done, you will have all the temperature data and pressure drops from the reference arm compared to all the other arms.”

At this point, the user can inspect the RSM for its coefficient of determination (R2), error, aliasing, overfitting or any other unexpected behaviour. If the model isn’t up to standards, then the user can define how many more runs should be added to the existing DoE. Once those runs are done, the fitting is reassessed.

“When looking at the RSM, you want to see If you increase flow rate will pressure drop as you expect. Or if you are getting values or spikes you shouldn’t expect,” Kolak noted. “Make sure to see the trends you expect from engineering knowledge. If you see a point in the RSM that looks odd, then run a simulation at that point to further investigate the behaviour of the model in that area.”

“Generally, an error of less than five percent is acceptable; however, a well-fit RSM can have less than one percent of error,” added Kolak. “Early in design, you can be comfortable with large variances as you are making large decisions. Later in development you want to see lower error values.”

Opening the RSM of the N-arm component in FloMASTER. The boundary conditions and labels of each arm is imported from the 3D model. Be sure to verify that the boundary conditions and labels fit your system model. (Image courtesy of Mentor Graphics.)

Opening the RSM of the N-arm component in FloMASTER. The boundary conditions and labels of each arm is imported from the 3D model. Be sure to verify that the boundary conditions and labels fit your system model. (Image courtesy of Mentor Graphics.)

Once the RSMs are defined, FloEFD can save them in a format that FloMASTER can open. This data is then used within an N-arm model to define how the part will react within the system model. From that point, the model can be used within FloMASTER like any other part in the library.

“FloMASTER will create the new flow component,” Kolak said. “Select the file and it will process it automatically. All the necessary data from FloEFD is included. You don’t even have to define the boundary conditions, just review them to see that FloMASTER read them correctly.”

At this point, the user can take a snap shot of the 3D image to help label and define the part. The user can also arrange where arms can clip into the part within the system model so they align with the image.

Now, this new element can be saved to the network and used in any FloMASTER systems model. This will help the engineer to seek out new local optima within the design space at warp speeds while not having to worry about computing power.

To learn more about FloEFD and FloMASTER and their 1D/3D characterizations, follow this link or watch this video.

Mentor Graphics has sponsored this post. They have no editorial input to this post. Unless otherwise stated, all opinions are mine. —Shawn Wasserman

Written by

Shawn Wasserman

For over 10 years, Shawn Wasserman has informed, inspired and engaged the engineering community through online content. As a senior writer at WTWH media, he produces branded content to help engineers streamline their operations via new tools, technologies and software. While a senior editor at Engineering.com, Shawn wrote stories about CAE, simulation, PLM, CAD, IoT, AI and more. During his time as the blog manager at Ansys, Shawn produced content featuring stories, tips, tricks and interesting use cases for CAE technologies. Shawn holds a master’s degree in Bioengineering from the University of Guelph and an undergraduate degree in Chemical Engineering from the University of Waterloo.