You are on page 1of 1

How can we use Machine Learning to improve Finite Element Analysis algorithms?

Ad by Learnbay Data Science


6 months | IBM certified data scientist course | Learnbay.
Learn data science and machine learning with projects to uplift your career and get
certified from IBM.
Learn More
1 Answer
Rahul Raviprasad
Rahul Raviprasad, Top Writer at Quora (2018-present)
Answered Oct 24, 2017
In Finite Element Analysis and Computational Fluid Dynamics,
You are trying to predict something by solving a higher order differential equation
(mostly the case in CFD and FEA) mostly by using numerical methods.
Now you don’t have good means to solve these problem, you end up by making some
assumption about your data, a rough starting point.
Now you apply this solution to the problem and see it is wrong to some extent, so
you correct it by changing your input, until you converge to a solution.

In Machine learning,
so far mostly I have come across problem of the type where based on past
experiences(data), we try to do prediction of two types, either classify something
or give out a regression number. People who have stats background would feel that
this is basically fancy terms for statistical analysis.

Now combining both, the only place I could think of is when you are collecting data
for boundary conditions or as a starting point, based on past data we could
possibly predict some inputs for our FEA or CFD models, such that convergence is
achieved faster.

Note this will only improve time and that too in theory. Also a computer is trying
to converge here on bad input vs Data Scientist trying to come up with ML model
that really helps a FEA person with better input which again has to converge to a
solution we wanted.

This is what I could think of, please do let me know your thoughts. If someone has
really done something of this sort, please share your experience, I am interested
to know more .

Thanks for reading!

https://www.altair.com/resource/the-impact-of-deep-learning-in-cae

You might also like