Doing More with ShapeAI in HyperWorks

Imagine if computers recognized geometric shapes as easy as humans.  What would you do with your newly discovered free time this would cause?

In one of my first jobs, I sat near a younger engineer who was working on a physical test correlation project.    One afternoon, I inquired what he was doing after hearing the same repeating patten of noise from his desk all day: 5 key stroke clicks followed by a few seconds of silence.  He was manually entering in recorded test data from a spreadsheet to an inhouse software tool.  The tedious nature of his activity made me feel bad enough to offer to write him a scripted automation.  I could imagine a solution to his task, but I wished I could find an automation for my own role’s tedious work.  As a project engineer setting up and running finite element analyses, I spent a lot of time searching and organizing models to identify similar parts and features to ensure that equivalent geometries were simplified and meshed correspondingly.  Unlike the straightforward steps of a mapped data entry automation, I had no tools to replace my brain’s inherent geometry recognition ability with an automation.  Lucky for today’s project engineers, the shapeAI technology in Altair HyperWorks can do this exact task.

To understand shapeAI, imagine Legos of all shapes and sizes spread across the floor. Incidentally, you are now visualizing the living room of my house!  Organizing the Legos into piles of similar shapes is a task that comes naturally to the human brain, even for a child.  This same task is more challenging for computers using conventional programming.  Specifically, it is hard to write all the rules that will properly sort the pieces with the accuracy we take for granted.  A first attempt to sort by the number of pegs quickly falls apart when such a rule cannot properly distinguish between clearly different bricks with 2x2 and 4x1 peg layout.  We could add additional rules such as the ratio of volume to area, but creating these rules is a challenge unto itself.  In machine learning these attributes with discriminatory value are known as features.   

Our shapeAI technology contains automatic feature extraction for the specified geometry itself without any additional input or intervention.  Combining these features with machine learning algorithms in HyperWorks’s matching tools puts the power of geometric machine learning at the fingertips of every user.  Only a few clicks are required to sort your own virtual Lego pile.   The video below demonstrates shapeAI being used to organize an automotive model by geometric similarity so that modifications to one part can be synchronized to all.

The time savings provided by the workflow in this video is modest, but just like my old coworker’s manual data entry tasks, the time savings becomes significant when the same tasks are repeated frequently.   Furthermore, the value scales well with increasing model size where our brains struggle to keep up with the complexity.  Injecting shapeAI technology to compliment human judgement has amazing potential for automation and efficiency in building simulation models.  I wish I had this years ago.  I’d love to hear about some of your tedious work tasks in the comments.  Maybe we can find automations for you next!