Dear All,
As you are aware the UDFRC homogenization tool is live on compositesAI and it works by invoking GMSH and Swiftcomp to compute the effective material properties.
The fiber can be isotropic / Transversely-isotropic / Orthotropic.
The matrix can be isotropic / orthotropic.
General prompt to invoke the tool would be something like :
I’d like to know the effective stiffnesses for a composite with 60% fiber. The fiber is orthotropic (E1=150 GPa, E2=150 GPa, E3=150 GPa, ν12=0.3, ν13=0.3, ν23=0.3, G12=57.7 GPa, G13=57.7 GPa, G23=57.7 GPa) and the matrix is orthotropic (E1=10 GPa, E2=5 GPa, E3=5 GPa, ν12=0.25, ν13=0.25, ν23=0.3, G12=2 GPa, G13=2 GPa, G23=1.92 GPa
For such structured prompts the tool is invoked without any issues.
Could you please try it out in a more general context and let me know if there are any prompts where the code is not triggered as expected? This could be especially relevant for users who are not well-versed in composite modeling and give very generic/vague prompts.
Thanks for your time!
@Haodong @Admin @Wenbin @lzhang @su.tian.22 @banghuazhao @SichenLiu
3 Likes
@Avinash are you comfortable to change the category to be public in CompositesAI?
Yes Professor, I changed the category to CompositesAI now
Hello Everyone,
Just wanted to send out a reminder to check if you have used the tool and have any feedback on how it can be improved , it would be great to incorporate that now.
I am specifically looking for feedback in the following areas - (Also feel free to add any other guidelines that the tool must follow)
- When the inputs are vague but the tool is invoked correctly and the results are given with appropriate remarks/assumptions.
- For a new user trying to understand the tool, the description given by compositesAI is correct and self-explanatory.
- Tool appropriately terminates the simulation if the parameters are unrealistic - Asking the user to revisit the inputs.
Thanks for your time! @Yu_Group
1 Like
For No. 1, my finding is that when inputs are vague, CompositesAI tend to guess what you want instead of asking you for inputs again. Maybe there is a way to minimize guesses and ask for results if unclear.
I know for No. 3 to be absolutely sure that parameters are acceptable, you can implement a verification method in your API code similar to what I have done. For example, if the user give unrealistic materials, the API will output the error message.
Ideally, we want a separate API whose sole job is to check inputs before passing it to the API where computation happens, but this idea was not that straightforward, so for now we can just implement it inside the API code in case CompositesAI did not catch the errors in the inputs.
1 Like
Thanks for the feedback Sichen,
For point 3, currently my material property checks are also within the API.
But if feel it would be better to have as many checks possible, at the schema level instead of having it inside the API. This might enable the AI system to understand the code better (and may contribute to its intelligence) rather than it being a mere tool to just call the API. This is my understanding, but you can correct me if it’s wrong!
1 Like
It might be hard to implement such checks inside the schema level because in many cases, the errors in the user inputs aren’t obvious. For example, I demonstrate that sometimes you may have an unrealistic material when your shear modulus is a bit too large, or any other cases when the stiffness matrix is not positive definite (can refer to the equations in the AAE 555 textbook), so computations must be carried out, which in the schema level it is hard to implement because the job of the schema is not to carry out computations (or any other logic). That is why we are still doing the checks inside the API.