FEA Solver that doesn't need defeaturing | Hacker News Reader

FEA Solver that doesn't need defeaturing

coreform.com

69 pointseks39125 comments

I don't work at this company, but know some of the people involved and I think it is really cool. For those not in the know, most FEA programs require the model to be simplified and a custom meshing put on it. They also have difficulty solving traction between two objects. Coreform's IGA can create a mesh on any shaped volume and solve traction between two surfaces easily.

It is amazing the number of hours sunk, even in our little business, into defeaturing or really just remodeling, into a more experimentally conductive fashion, any of our prototypes which involve complex lofted multi-body assemblies. This just adds another layer into the design-data stack between the design phase and cam phase, which all needs to be maintained. I'll be quite interested to see their software evolves.

Meshing is truly a PITA, if this does away with it I'm sold.

If I get this, which I may not, it's fea using splines instead of straight edges. Very clever stuff, and some properly hard math. Impressed!

Going by the name 'isogeometric analysis' is is probably analysis on level-sets, not sure if splines are the easiest to translate to level sets but the effect is probably very similar.

I wasn’t sure what this meant so I found this paper that seems to give a good overview of what isogeometric analysis means. https://reader.elsevier.com/reader/sd/pii/S0045782505005451?...

Nurbs, or non-uniform rational b-splines, to be specific

It's a generalization of B-splines, called U-splines (splines over Unstructured meshes). If anyone is interested in the math behind it on the level of a journal article, the last two links here go into the technology in detail: https://docs.coreform.com/latest/html/coreform-public/index....

(Disclaimer: I work for Coreform, developing U-splines)

Existing FEA codes, including popular commercial ones are based on huge bodies of academic fortran code. Does this system replace all that (well maybe expect for stuff like matrix math and linear solvers) with a novel framework?

What kind of work goes in to validating correctness of these sorts a systems? A serious flaw in a FEA system could result in a substantial loss of life.

Are life safety applications allowed to rely on FEA alone?

If it's life safety chances are there's an acceptance/certification test that needs to be done to demonstrate safety. I think some applications that are life safety can use FEA for acceptance test but the criteria are strict and the FEA model needs to be well understood. FEA is mostly to aid in the design phase.

Kind of. My father is an engineer who uses FEA extensively in his small niche of pressure tanks. They use FEA to design the tanks and then stress test manufactured versions to ensure the manufacturing process matches what the FEA model predicts.

Absolutely. Models are usually validated against the set number of conditions when possible, but it is impossible to validate all of the outputs across the operating range

I Hope not, since bugs have been found in the reference model of some rocket engine 20 years after they had gone gold. In other new comparison operators eat NaN for lunch and carry on with the computation like they had a sensible answer.

Never. Complementary regulations (varies by location and industry) define methods and procedures, validations and control, based on statistics where necessary, for assuring an accceptable level of risk of failure (never zero but very very small) in balance of resource use. Errors might be introduced in any stage of realization of an engineering product from concept to delivery (including calculation errors, misuse of methods and tools, etc.) but we could prepare for know - and to some extent unknown - sources of trouble implementing assurances and validations, approvals and checkups, coupled with criminal accountibility. What comes out of a FEA software (which could have bugs, which could have been misused, using wrong data, etc.) will never be realized without scrutiny. Even the scrutiny and validation methods can have errors in it so redundancy is expected but straight delivery of FEA calculations (actually any stage of the process) into production without checks can be considered as criminal neglect (of which is impossible to eliminate like any other crime).

>A serious flaw in a FEA system could result in a substantial loss of life.

Seems like a good reason to rewrite as a js saas that needs to be rewritten in the latest framework every few months.

FEA is used for design assistance usually, not considered good enough for verification in aerospace at least (not sure about other fields). i.e you can design the wing for a certain loading using FEA to model but then a proper test needs to be completed that tests the wing to breaking point to verify the design meets constraints.

This must be true for any other analysis and calculation methodologies too.

Well I wonder sometimes. Like you don't build a building and then stress test it... so there must be some fields where simply adding the 30-50% margin for error is good enough.

Absolutely. Safety factors allow engineers to have good night sleep. ; )

Luckily in constructions the use of novel methodologies aren't that common like in other industries so previous small and sometimes big scale tests - also previous spectacular failures - predict future behaviour to a good extent with predictable risk factors, considered in safety factors. Also known things combined the known ways are preferred (or expects huge costs for your one off object). National building regulations on methodology encode some of these experiments (considering load, material and manufacturing uncertainties on top, also the time factor in the many many decades range) laying a legal foundation on what can and what cannot be done and expected, where the responsibilities start and end. Basically no need to test all new designs if proper - and legally approved - methodologies and procedures are used, rules and methodologies cautiously formed represent the validation there. (unusual shapes, materials, compositions, methodologies not covered by regulations might require much more scrutiny then, and potentially much higher safety factors too. ;-) )

Exactly, in many safety critical areas its required by law. It's way more than 30-50% though. In medical devices it's 4x-10x.

Full engineering tests. In medical devices it's called V&V, verification and validation. You verify the design is suitable and meets the invite requirements, you validate that your manufacturing process creates that design, exactly. You use software like this, or the custom academic Fortran code of the times before you mention as a design tool, as a way finding tool to help maximize strength or minimize weight of a particular part.

Parts of this company were originally spun off of Sandia national lab, so I imagine the staff has at least some familiarity with the importance of V&V. In general, it’s on the engineering users to do their due diligence and develop the right validation cases for their design products. The code developer can get as far as coming up with the mathematical and numerical verification of the code, but the possible end uses are too numerous for it to be feasible for them to develop validation cases for anything beyond some basic industry benchmarks.

No plastic material, only basic contact, no advanced connectors... This is my problem with all these fancy FEA codes, nice in theory, not fit for industrial models.

if you are interested in mechanical contact, we developed a symmetrical algorithm (no master/slave) and can make it available for you on https://tanatloc.com

Reply on news.ycombinator.com