Optimisation gets close to converging but then diverges

Altair Forum User
Altair Forum User
Altair Employee
edited October 2020 in Community Q&A

Hi Everyone,

I'm running an optimisation with multiple load cases. The solution gets really close to converging (in the .OUT file) the first and second convergence criterion are met and just as its about to converge the solution diverges. It runs for the maximum number of iterations (80) before it stops. Is there any way to change how long the optimisation will for so that a solution is reached before the maximum number of iterations are reached?

Or does it more than likely suggest my load cases are incorrect, or theres something wrong with my model? etc.

 

Tagged:

Answers

  • Altair Forum User
    Altair Forum User
    Altair Employee
    edited April 2018

    Hi Jerome,

     

    Welcome to the forum! I don't think your result will be better if we could increase the amount of iterations within Inspire. If your Optimization takes more than 50-60 Iterations, there is usually something wrong with your setup - or - at least we can tweak it to force a convergence.

    If you like to share your model wide, please upload it here - or - PM, so I can send you an upload link.

     

    Regards,

    Felix

  • Altair Forum User
    Altair Forum User
    Altair Employee
    edited April 2018

    So in Inspire, there's no way to change how many times a solution will iterate, or to change the convergence criterion?
    I checked over the files and it might have something to do with one for the load cases. 

    Also I recently downloaded a student version of HyperWorks after read some in this forum that a convergence graph can be viewed using the .hgdata file, however I was unable to open this file. 

     

     

  • Altair Forum User
    Altair Forum User
    Altair Employee
    edited April 2018

    Inspire is not allowing to set the max. amount of Iterations or any other values regarding the optimization process itself.

    Optistruct for sure is allowing this through Hypermesh - if there is a need for you to do it - please refere to HyperWorks user forum.

     

    Again - if there is no convergency within 80 Iterations, there is propably something wrong with the setup (maybe missing supports, large deformations...).

     

    -Felix

  • Altair Forum User
    Altair Forum User
    Altair Employee
    edited April 2018

    Okay, so I just ran two topology optimisation simulations for the exact same part (one after the other), the only difference was the Factor of Safety. My first one I selected a factor of safety of 8 and  the other of 5. The FS 8 ran for about 60 iterations before converging while the FS 5 ran for 80 but failed to converge so I'm 100% sure my load cases are correct. I looked at the .out file and the FS 5 got to the 2nd convergence criteria a few times, each time it would come up with warnings that the loads epsilon values are larger than 10% after that 2nd criteria is met and then it would iterate another 20 times before it says 1st and 2nd criterion met until 80 iterations are completed. 

    The 1st convergence criteria also doesn't seem to start when it should. 
    For example, in the FS 5 case:  (FS 5) iterations 67 through 75.

    The convergence criteria seems to change between iterations. Sometimes it will say that a percentage change of 0.37% is enough to satisfy the 1st convergence criteria like (iteration 54) or 0.28% is enough to satisfy the 1st convergence criteria like in (iteration 64) but then 0.19% change like (iteration 67) or 0.31% (iteration 68) or 0.22% (iteration 70) doesn't satisfy the 1st criteria. 

    Any help would be appreciated. 

    Unable to find an attachment - read this blog

  • Altair Forum User
    Altair Forum User
    Altair Employee
    edited April 2019

    Okay, so I just ran two topology optimisation simulations for the exact same part (one after the other), the only difference was the Factor of Safety. My first one I selected a factor of safety of 8 and  the other of 5. The FS 8 ran for about 60 iterations before converging while the FS 5 ran for 80 but failed to converge so I'm 100% sure my load cases are correct. I looked at the .out file and the FS 5 got to the 2nd convergence criteria a few times, each time it would come up with warnings that the loads epsilon values are larger than 10% after that 2nd criteria is met and then it would iterate another 20 times before it says 1st and 2nd criterion met until 80 iterations are completed. 

    The 1st convergence criteria also doesn't seem to start when it should. 
    For example, in the FS 5 case:  (FS 5) iterations 67 through 75.

    The convergence criteria seems to change between iterations. Sometimes it will say that a percentage change of 0.37% is enough to satisfy the 1st convergence criteria like (iteration 54) or 0.28% is enough to satisfy the 1st convergence criteria like in (iteration 64) but then 0.19% change like (iteration 67) or 0.31% (iteration 68) or 0.22% (iteration 70) doesn't satisfy the 1st criteria. 

    Any help would be appreciated. 

    Hi this is karteek 

    I am getting problem while doing optimization. Will you please help me. My problem is, I am not getting any iterations only 1 iteration I am getting and it is not optimizing. And I am attaching my 3d model file and fem file. And this is my mail Id kartheekalla66@gmail.com and will you please solve and send the file and to that address and plz explain where I did the mistake. I tried so many times and getting different different errors....... 

    Unable to find an attachment - read this blog

  • Altair Forum User
    Altair Forum User
    Altair Employee
    edited April 2019

    Hi Kartheek, 

    Seems as though there are some modelling issues with your current model. From the model, I gather that you are trying to optimize the lower part of a sandal. I am attaching the solved and optimized model and also, here is a list of rectifications you can make while running your own optimization.

    • Use a pressure load instead of a point load as that defines the pressure of the foot more accurately
    • Use a support on the lower end of the model to constrain it as opposed to using the supports on the upper end of the model

    Here is an image of the optimized model:

     

    <?xml version="1.0" encoding="UTF-8"?>image.png

    Unable to find an attachment - read this blog