Geometry Prediction using Data from Optimization: A STOCHS App Application

Hello, this is Ertan Taskin from Ozen Engineering, and in this video, I'll be making a quick demonstration of the utilization of a STOCHS app to generate a geometry. Before diving into that, I would like to give a little reminder.

Previously, I have performed an optimization analysis for a simplified case, and the corresponding Workbench screen is over here. So that required 20 data points that are simulated, and we had diameters as an input, and the outputs are pressure drop and the uniformity data.

Just to give you a reminder on the geometry, a very simplified shower head, if you will. We have internal holes, and the diameters of those holes are the parameters.

In order to get the uniform pressure distribution on this particular region, as well as the pressure drop, we perform the optimization analysis. The results of that are already provided here before, out of 20 simulations with the Pareto Front Analysis, apparently.

The design 19 gave us the most optimal results, and the corresponding diameters are shown here. So, in this work, we will make use of the data generated by this optimization study and create an app to further predict the geometries utilizing an application.

So, the schematic structure of this work is pretty much shown here on the screen.

We collected the data from the optimization table, and which I showed you already before, and posted the data for all of each 20 simulations, and saving or exporting the data with the coordinates, translating Ansys scale.

All these three sets of information are performed with Python code, utilizing optiSLang functions, and then later with an additional Python code to generate the geometry, which ultimately leads up to the application. A glimpse of that code, which leads to the application, is pretty much over here.

Reading the VTK files and then further performing all the works to create the geometries with all the essential functions. So, in order to run the application, I will start it first. When we do this, a new page pops up and continues to read the code.

Once the app is ready, it is asking us to enter the path of the data, all the VTK files, where are they located. So, I will bring that link here, and then enter.

Load and predict prepares all the designs that are worked from 1 to 20 and creating the corresponding geometries colored with the required variables. It is already done. So, it is showing us the first design; we can definitely switch and take a look at the others, but let's start with the first one.

For statistical purposes, we can take a look at the mean or the 95 confidence level analysis too. We have options, as I mentioned before. We have pressure or velocity to color the entire field. Let's continue with the pressure.

On the lower side of your screen, as you see, we have the pressure distribution, and this is the color scale. The pressure is in Pascal, so from 100 to negative 400. While rotating it, this design was the uniform diameter design. As you see, this is how the pressure distribution is.

Now, you may wonder why I have this scattered view. Please note that we are not reading the already simulated data. This app utilizes some of the design data out of these 20 cases for training purposes and uses the others for the predictions. Remember, these two diameters were small.

And in fact, I think we can again take a look at it on the screen. The first two diameters were small, the third and sixth one was large, and the others are in between. Alright, yeah, the first two are small, the third and the last one are large, and the others are in between.

I guess that's in a pretty good agreement. And again, this is the pressure field. And when we make a comparison to the CFD predictions, I think we will see a good agreement.

So, on your screen here, the Ixtahoz prediction is on the left, and the CFD post image from the Fluent simulation is on your right, and the main features are definitely captured with the high spots here and there, and the low pressures definitely with the gradients of that, and I think this is, even though this is scattered and a bit coarse, I think this is a very good agreement.

And let's take a look at the velocity field. Yeah, similarly, the higher velocities due to the jet in the entrance region are depicted well.

Here are the ranges of the color, and similarly, when I zoom out, I guess the view will be slightly better, and so I can definitely take a look at different designs, such as design one with the uniform diameters, if you remember, the main flow features are very similar.

The scale is very similar, but apparently, the scale is impacted based on the design. Let's take a look at 20. Let's reload pressure first. Yes, it is better to load the pressures first. Yeah, again, the scale will be different, but the main features are similar.

So, in this video, we have demonstrated how to, very briefly, of course, create an app utilizing the optimization data. And briefly demonstrated the corresponding codes to create the necessary auxiliary files, as well as the main Python code to generate the application. Thank you for watching.

Please contact us at https://ozeninc.com/contact for more information.