trail wrote:
Your post is full of contradictions.Must not have taken the elective course "Sarcasm in Engineering". While I'm sure you're super cool in real life, this post has thus far confirmed my perceived stereotype of SparkE's ;-)
codygo wrote:
Aerodynamics is inherently non-linear.
Contact is inherently non-linear, but can be ignored in all but a couple critical areas
Loading given large displacement is inherently non-linear, but displacement on carbon road frames are generally small enough to ignore (except maybe the old Roubaix)
Plasticity is, by definition, non-linear (if a company is taking advantage of material plasticity in designs, please let me know so I don't buy their product)
Fatigue is, by definition, non-linear and should not be ignored
I would assume you are 100% correct, bikes are the product of completely linear analyses - whether that is applicable or the company has the capability to do anything else is a different story.
Collectively, academia/industry has agreed to simplifying assumptions. Doesn't everyone - actually looking to accomplish meaningful work in this flow regime - assume inviscid, incompressible, irrotational (although I suppose that's redundant after assuming inviscid), steady flow? So we test in the tunnel to gauge the disparity between computational and experimental results and attempt to quantify the effect our assumptions have on results, although 30mph isn't really what some people would consider tricky (ok, maybe the interactions are worthy of experimental verification and there are a couple areas where the inviscid assumption would be detrimentally non-conservative). It's not like the N-S equation was analytically solved before producing the newest superbike, although every marketing department would have you believe that...
Same story with structures. Do we assume homogenous material properties for composites? No. Orthotropic? Technically should be fine but if we're all going to be pedants about our field of specialty, no. Anisotropic? Yep, and having all those material properties with statistical significant which is representative of the individual's processing is expensive. What about element testing to qualify a certain design feature? Population statistics on the assembly? All those cost money and would sell fewer frames than a cool paint job by a long shot. On a good day we can get our models to correlate to test data in either strain or displacement, generally not both simultaneously (although the degree of anisotropy is not helping). At the end of the day, a [reputable] FEA code really is only good at predicting sensitivities given the right inputs (material properties). Remember, Solidworks and ProE/Creo/Wildfire are modeling software packages which include a FE module to sell licenses, which is nice if a small company doesn't want to spend money on a FE license and analysis tokens for an accepted good solver to produce results which don't garner the same wow factor as ultra sweet wind tunnel results. The redistribution of load given contact is very much a gray area in structures; we use three really good solvers (~$50k/seat/year each, not counting analysis tokens) for both implicit and explicit, linear and nonlinear analyses and only one solver really does an acceptable job (which does not begin to describe how computationally expensive that analysis job is) yet we and everyone else have contact everywhere - one of my recent projects modeled contact between components as not including it was too conservative, each job takes ~32 processor-hours on 50gb of RAM (even on 16 cores, iterating on a 2-hour lag is irritating). To get those results, a license had to be purchased for the pre/post processor (say $50k) and 16 tokens ($20k each), plus my burden rate. Or they could put out a gif like the one posted above from Creo Simulate, 99.99999% of the people would have starry eyes and I just click back on my browser without even reading - apparently a converged mesh is optional?