Curve-Prediction using Machine Learning
Altair EmployeeHi Altair Community!
This is the first time (hopefully out of many) that I am posting a blog post, so bear with me in my attempt to start a discussion regarding my use of Altair Tools. I was recently in a discussion with a client who was looking at predict stress-strain curves for their materials, specifically to reduce the amount of time they spent testing new material to achieve the optimal stress-strain curve for the application. I thought to myself that this problem should be pretty straight forward to solve with the data analytics tools from Altair, but as so often happens, the client did not have access to the data from the test-bench.
Often when working as a Pre-Sales Engineer for Altair Data Analytics I run into the problem of not getting access to the data. This can be due to a wide variety of reasons, like data security or having a non-centralized system for managing data from the client. What I usually try to do in those cases is to produce my own data to represent the client use-case and by doing so, show the functionality of the tools on representative data, without having to breach any security concerns from the client.
I started by drawing curves representing a generic stress-strain case:
You might say that this curve looks like a five-year old tried to draw a stress-strain curve, but this blog post is about non-linear modelling using neural nets, not my artistic abilities.
I drew 9 different curves and saved these as .png images. I then used Python code to extract the X/Y coordinates of the lines, which is what I would use for the Machine Learning Model. The data from the curves were extracted to a .csv file which I could analyze using Altair RapidMiner.
I created a Deep Learning model which I could use to make predictions of the curves. Finally I visualized the new equations together with the original data points in Altair Panopticon.
The colored dots represent the points I drew in my image and the black lines represent the curves predicted by the deep learning model. Pretty good results if you ask me and the customer was happy that they could get an introduction to the Use-Case for Data Analytics and was also happy that we did not have to spend 6 months signing a NDA or work go through the tedious process of getting real data.
This show-cases one of the visions of Altair, where artificial data can help speed up the process of Data Analytics and enrich the data set. Altair wants to combine Simulation, Data Analytics and HPC for making better decisions, hopefully this blog-post showcased how this vision can be implemented.