Problems in Model testing with Physics AI.

In hyperMesh 2024 using the modeulo PhysicsAI , I follow the tutorial to train the model and in the test part of the model I get an unexpected error message, all at the time of show results.
Answers
-
Hi @Mejia_21,
Kindly provide the below requested details so that we can suggest accordingly on the query.
- Kindly provide the log file (.txt) from the 'scoring' folder created in the PhysicsAI project directory.
2. Kindly let us know which PhysicsAI tutorial has been referred? Is it the HVAC one?
3. Also, from the snapshot, we can infer that you are using HyperMesh on the Altair PBS/Access interface than on a local machine? Is the understanding correct? Kindly clarify.
Many thanks.
Kind Regards
Garima Singh
0 -
I have located the requested log file in the 'scoring' folder within the PhysicsAI directory. I am attaching the .txt file for your review.
Yes, the tutorial I'm using is the one related to HVAC in PhysicsAI. I attach the link
For the third question: I'm not using Altair PBS/Access specifically, but it looks like I'm working in a centralized or managed environment. HyperMesh runs from a dedicated directory (/appl/hyperworks/), suggesting that the software is installed on a server or shared machine, likely in a managed corporate environment.
So, while this is not an Altair PBS/Access managed environment, it does appear that I am using a centralized infrastructure and not an on-premises machine.
0 -
Hi @Mejia_21,
Thank you for providing the files.
As per the files provided, the Model Testing step has been executed without any errors.
I infer that the error is related to visualizing the .h3d file created in the 'scoring' folder as it may have been corrupted.
Hence,
- Kindly confirm if there is/are any .h3d files created in the 'scoring' folder.
- If the .h3d files exist, are these empty result files (can be checked by checking its size)?
- Also, provide any one of the .h3d files from the folder for us to check.
While we work with the development team on the issue,
- Kindly perform the 'Model Testing' step again. To do so, delete all the contents of the scoring folder (do not delete the scoring folder) & click on 'Model Testing' again. If you get an error such as 'testing results exist' then, create a dummy testing dataset & perform 'Model Testing' with this dataset.
- Kindly perform the same tutorial in any of the latest HyperWorks version that you may have such as 2024.1 or 2025.0 & let us know if you get the same error.
Kind Regards
Garima Singh
0 -
Hi @Garima_Singh ,
Indeed, the .h3d files are present in the scoring folder, and after checking their sizes, I can confirm that they are not empty. I am attaching one of these .h3d files for your analysis so you can check whether the file might be corrupted or if there is any other underlying issue.
Following the instructions you provided, I deleted all the files within the scoring folder (without deleting the folder itself) and then repeated the "Model Testing" step as suggested. However, unfortunately, the error still persists, and I am encountering the same issue. Despite following the steps correctly, I have been unable to resolve the problem.
Regarding testing with newer versions of Altair HyperMesh, I am unable to do that, as the version I am currently using is the only one I have access to at this moment. It’s important to note that I performed the same procedure in a Windows environment, using the exact same files, training parameters, and steps. In the Windows environment, everything worked as expected, and no errors were encountered.
At the moment, I am working in a Linux environment. It is possible that the operating system I am using might be influencing the behavior of the software in some way. Therefore, I believe this could be a relevant factor that might be contributing to the issue I am experiencing.
I would greatly appreciate any additional recommendations or suggestions you may have to help resolve this issue in the Linux environment.
0 -
Hi @Mejia_21,
Thank you for the providing the .h3d file & additional details.
The development team tested the HVAC tutorial on the Linux (server) machine on the 2024.0 version & 'Show Results' under 'Model Testing' is working an intended (as shown below).
Hence, we infer that this issue is machine specific as we could not replicate in our Linux machine.
Few additional details are required by us to suggest further on the issue:
- Does the license attached to this machine has HyperView feature in it? Kindly check & let us know. If HyperView feature is not there in the license, then, it is indeed the root-cause for this issue.
If HyperView feature is there in the license, then, the following steps to be followed:
- Can you test the same HVAC tutorial on another Linux machine or perhaps a machine of your colleague? If it works on the other Linux machine, then, this machine has some issues with respect to HyperView loading the model.
If the HVAC tutorial does not work on the other Linux machine as well:
- Provide Linux Operating System version details.
- Can you repeat the whole tutorial again on your machine from start in a new folder? Note: create a folder name with no spaces in the naming convention.
- Kindly test the 'Predict' option by first loading an input .fem file from the tutorial & then click on 'Predict'. Note: First set this model as 'Active' in the 'Model Training' tab & then load the new design (.fem) from the 'newDesigns' folder (provided in the tutorial) & let us know if the you get any further error or the 'True & Predicted windows'.
Kind Regards
Garima Singh
0 -
Hi @Garima_Singh,
Thank you for the guidance and the steps you provided.
I’ve verified that the license I have does include the HyperView feature, and I’m able to use it without any issues. However, the error persists. I’ve tested the tutorial on another machine, but the problem continues to replicate.
Regarding the Linux version, I’m using Red Hat Enterprise Linux 8.10. I’ve followed the tutorial again from the start, using a new folder (without spaces in the folder name), and tested the "Predict" option by loading a new
.fem
file, but the error still occurs. I’ve tried different files, but the issue persists across all cases.I was told something about tokens, but I’m not entirely clear on how this aspect works. Could it be that the issue is related to a token limit or some restriction in the license? If that’s the case, could you provide guidance on how to manage this or if I need to adjust any settings?
I also suspect there may be a compatibility issue with the version of Linux I’m using, but I’m not entirely sure. Could you confirm whether HyperView is fully compatible with RHEL 8.10?
I look forward to your feedback and suggestions so I can proceed with resolving the issue.
Best regards,
0 -
Hi @Mejia_21,
Thank you for providing the additional details.
While we check with the development team on the RHEL 8.10 support for HyperWorks products, kindly check if there is 'EngDataSciPaiShape' feature in the license & there are 75 units/tokens available.
Important note:
- After model creation in the 'Model' , do not click on 'Set Active Model' & if the model is set to 'Active', kindly click on 'Release model'. The tutorial is being updated with this detail.
- If an active model is used in the 'Model Testing' step, then, 150 (75+75) units will be consumed instead of 75 which may be causing an issue in your use-case.
Kindly test the recommendation (performing 'Model Testing' step with model not active) & let us know.
Kind Regards
Garima Singh
0 -
Hi @Garima_Singh ,
Thank you for the update and for the effort your team is putting into testing the compatibility with the RHEL 8.10 version. I would greatly appreciate it if you could notify me as soon as you have more information regarding this.
Regarding the license verification, I'm not entirely sure how to do this, but what I can provide is the license log. I’ve attached a screenshot of the log, which might be helpful for investigating the issue in more detail.
As for the recommendation you gave me about not selecting the model as "active," I followed your instructions and performed the test without marking the model as active. However, the error still persists, even with the model inactive. The logs continue to show the same error message.
Thanks in advance for your help, and I look forward to any updates.
Best regards,
0 -
Hi @Mejia_21,
We have coordinated internally on this query & our colleague Paola Alvarez from Mexico will coordinate directly with you on this issue.
We have also requested one of the PhysicsAI development team members to join the meeting as well.
Note: Red Hat Enterprise Linux 8.10 is not the supported Linux version. But, as per the PhysicsAI development team, it is less likely to be the root-cause.
Many thanks.
Kind Regards
Garima Singh
0