Virtual Crash Testing

"Snuze: Needs another Swede" (markg) 02/19/2015 at 08:47 • Filed to: None

1

10

This is "crash testing" for ships:

Ok, well not exactly, its a shock trial, but I wanted a cool image to lead with, and it's related. This morning someone resurected a comment I made a few weeks ago on DeMuro's article about the regulations involved in importing cars. Link provided:

!!! UNKNOWN CONTENT TYPE !!!

Doug mentioned that you could, in theory, bring in a car newer than 25 years, but one of the biggest hurdles is the crash testing. So it got me thinking, why can't we do virtual crash testing?

I come into this with some experience working for the US Navy doing shock testing for shipboard systems. I'll give a little background because I think it's cool, and it's relevant to where I'm going with this. All of the Navy's frontline surface combatants have to be Grade A shock rated. The grade has to do with the amount and types of failures that are permitted on various systems throughout the ship. In the olden days, every single individual system was tested on a light, medium, or heavy shock machine, or on a "barge shot". It can get outrageously expensive in very short order. A shock machine consists of a frame, with a giant hammer on a pendulum, which strikes an anvil, a large metal plate with various mounting tabs and holes. For each piece of equipment, you have to design and fabricate mounting jigs to simulate shipboard installation, that also bolt up to the anvil. You then instrument the heck out of the test article, and drop the hammer from different heights and measure the response. You then flip the item around and repeat in different orientations (simulating X, Y, and Z axes). After each hit, you have to function check the article to make sure it's still working. It can cost anywhere from several thousand to a few hundred thousand dollars, and take weeks to months to test a single item. Barge shots are much cooler, and more straight forward - they take the test article, strap it to a barge in the middle of a pond, then detonate a few hundred pounds of explosives anchored to the bottom of the pond. The catch is that each shot costs almost $1M.

So as computers became more available and capable, the Navy started looking for alternatives to the tedious shock testing program outlined above. Eventually, they came up with !!!error: Indecipherable SUB-paragraph formatting!!! It uses finite element modeling to simulate the interaction of a test article and it's rigid mounting structure during a simulated underwarter explosion. The software results were compared with thousands of actual tests to verify results and ensure the analysis model was accurate, but eventually it was accepted. It's now very common practice to use DDAM in place of major shock testing, though some smaller components are still physically tested. Also, the first ship in a new series generally undergoes shock trials, which is very simliar to a barge shot - they detonate underwater explosives near the vessel, not enough to (in theory) damage or destroy it, but enough to give it a good shaking, to expose any possible weak points.

Doing some Google searching reveals that the same analytical tools are availabe for automotive use. In fact it appears that most manufacturers already do virtual crash testing to ensure compliance before submitting test vehicles to NHTSA, etc. So my question is, if this is so common place, why can't it be done? I have many years of experience in different facets of the Navy, and while I wouldn't call them technologically backwards (they have been pushing lasers and rail guns afterall), it feels that way sometimes. What it really boils down to, though, is that they are distrustful of new, unproven technology. But, they have been using DDAM for quite a while now, which to me is verification enough that the technology is there.

Manufacturers already make 3D models of all their cars. Many are even doing virtual crash testing already. I don't think this is a switch to be made hastily or lightly, but why not start gathering the data now? Compare virtual models of current cars with their physical crash test results to start refining the analytic model. We have the computing horsepower, so why not put it to good use? It can save money, and time, and most importantly to use enthusiasts, strike a nail in the coffin of the 25 year import rule.

1.21 JIGGA WATTS!!! > Snuze: Needs another Swede 02/19/2015 at 09:34

0

I think it's because of the sheer amount of computing power needed for finite element analysis (FEA). There are hundreds, possibly thousands, of variables that need to be addressed when you crash a system as complex as a car. Some of it is as simple as "does bolting these two parts mean they act as one part, or is there infinitesimal movement to be accounted". Most of it is more complex, like "The steering column is 0.25 degrees out of spec due to assembly constraints, but isn't captured or noted in the model. Will this affect the crash?".

A big reason could be system imperfections. An engineer will put tolerances/parameters on his design that accounts for manufacturing imperfections/flaws. However, the 3D model will almost always be the "best case" scenario of part dimensions, part location, and assemblies. The 3D model will then be tested in FEA and that would result in a "best case scenario" crash.

However, testing a real-life vehicle allows manufacturers to test "best-practice" scenarios that come from the factories, which means they test the car that was made and assembled the most efficiently. Best case /= best practice. There is just too many variables to account for in a FEA program that it will either take days to run one simulation, or have to use ungodly amounts of computing power.

Rock Bottom > Snuze: Needs another Swede 02/19/2015 at 10:06

1

I'm an experimental aerodynamicist at a major American auto manufacturer and we use computer codes to do quite a bit of the aero work, but then we validate the results in a wind tunnel or on the road. This is because no matter how good you think your model and grid is, your boundary conditions are defined, and your code works, there will always be a need to prove that you got the engineering right. It's the equivalent of an acceptance test. We don't road test every single spoiler or air dam that people dream up as a car is being developed, but the computational tools let us eliminate less effective ideas early so we can focus our limited (and expensive) test time on the features that are more likely to work well.

I assume that the crash test people are doing similar things. I would bet that they can do much of their development work virtually and then be fairly confident that the final product will crash well. This way they don't have to keep crashing prototypes for years as the design matures. But no matter what, in the end they still have to crash it to prove, beyond a doubt, that the design's right.

Sidenote: Your job looks AWESOME! I mean, bombs! Warships! More bombs! 'Murica!!!

Snuze: Needs another Swede > 1.21 JIGGA WATTS!!! 02/19/2015 at 10:23

0

Well, we have LOTS of computing power available to us these days. All the major manufacturers run, or lease time, on huge distributed networks to do FEA and CFD already. As I mentioned, they are already doing some of this virtual crash testing now, before a single model ever gets smashed, to ensure compliance before a single car is sent to IIHS or NHTSA for testing. So it's not like it's not possible.

I do agree with you that best case /= best practice. But I think you have to look at the bigger picture. Crash worthiness is generally inherent in overall design, not the minutia of how a bolt is tightened or if the steering column is a quarter of a degree off. I think if that were the case we'd have radically different testing practices than we do now just based on assembly line variance and the tolerances in the properties of the alloys that are used to make the car.

doodon2whls > Snuze: Needs another Swede 02/19/2015 at 10:34

1

Modeling, Simulation, and Analysis are used extensively in the Structures and Impact Analysis domains within the auto industry. The government requires compliance testing with actual production intent hardware which I kind of agree with. It puts the design up to the ultimate test, and ensures that there are no unsurfaced issues - especially with supplemental restraint systems.

In fact, there is so much variability in results from crash testing [there are a lot of hardware and software systems involved] that manufacturers impose conservative limits upon themselves for the design to ensure that the _one_ shot they have at a compliance test goes off without a hitch. Of course if they fail the compliance test, they can re-test, but who wants to generate the PR and data from a bad test under the direct purview of the gov't ?

1.21 JIGGA WATTS!!! > Snuze: Needs another Swede 02/19/2015 at 10:38

0

Very true. I can't argue with any of your points. I would say that virtual testing can't always account for the randomness that Murphy likes to throw in. I've got a pretty good feeling that there is a good percentage of physical tests that left engineers scratching their heads and saying, "Well that's funny. That wasn't supposed to happen." Maybe not the majority of tests, but enough to conclude that physical tests are necessary? I'm just guessing here.

If there is one god that engineers would pray and sacrifice to, it would be Murphy, in hopes that he does not appear...

Snuze: Needs another Swede > Rock Bottom 02/19/2015 at 10:48

1

Thanks. I don't do that kind of testing anymore, sadly. I'm still in the industry though and moving over more to the analysis side.

And you're absolutely right about verification, that's why I said, it's not an overnight switch. When I did shock testing, I got to test a few physical items on light and medium shock machines, and we did modal analysis and FEA before we ran the tests, and compared results in the report. It was always within an acceptable level of variance.

My thought is, and apparently the Navy's is, that most warships are pretty damn similar in terms of hull form, structure, etc. So if we can validate our model across multiple platforms (think the older CG41s, FFG7s, DDG51s, etc). then it should work on newer stuff. I suppose the one caveat to this might be radical designs, such as the LCS-2 Independence Variant, or the DDG-1000 Zumwalt Class. They are so radically different in terms of things like hull form that maybe they need to be tested and evaluated differently. But I haven't done any work on either of these platforms so I can't really speak to that.

Likewise, with automobiles, there's going to be some major differences between types of vehicles - unibody vs. body-on-frame, passenger car vs. SUV, FWD vs. RWD, front engine vs. rear engine, etc. But take any, say, compact sedan out there. They are pretty much all front engine, front wheel drive, unibody design, use McPherson struts up front, etc. I bet if you developed a good analytical model based on one particular car, it would hold well to any other car you throw at it, so long as its of the same basic type (front engine, fwd, unibody, etc.).

The only other driver I can think of here, really, is cost. Obviously doing destructive testing on a warship would cost literally billions of dollars, so that's not feasible. Providing 8-10 cars to crash test is a drop in the bucket, probably <0.5% of the budget, for most major automotive programs. So from that perspective I could see that there isn't really a reason not to do it.

But it could make sense for niche or boutique manufacturers who are only turning out a few vehicles a year. And as I mentioned at the end, it could be a strike against the 25 year law. If I wanted to import a car that was never crash tested in the US, why not allow me to compare virtual crash data against US metrics to see if it cuts the mustard? Well, I know why, I'm just making the point that I think this could be a valid practice.

Snuze: Needs another Swede > 1.21 JIGGA WATTS!!! 02/19/2015 at 10:56

1

I got to thinking in replying to another comment, and cost for cars, at least for major auto manufacturers is negligible, so I suppose there's not reason not to crash test. When you're talking hundreds of millions in development, supplying 8-10 cars is a drop in the bucket.

And you're absolutely right about Murphy. A few years before I got the job where I was doing physical testing, the department I was working in lost a test lab. They were doing a loose test on some ammunition to certify it for transport - they just throw a handful of rounds in a wooden crate and let them bounce around. Well one was defective, I think the primer wasn't seated properly, and a few hours into the test, the tip of one round contacted the primer hard enough to set the round off, which started a chain reaction and ended up burning the building down. Fortunately all that stuff gets done remotely, the controlling station is 1/2 mile away, so no one was hurt, but the lab was completely gone.

But I still think virtual testing could be beneficial for boutique manufacturers who only turn out a few cars a year. Or if they could loosen the regulations (I know they won't but a man can dream, cant he?) for importing non-US made cars by substituting virtual crash test data.

Snuze: Needs another Swede > doodon2whls 02/19/2015 at 10:58

0

All good points, and as I'm talking to more people here I'm beginning to see that. Also, automotive development budgets are usually in the hundreds of millions of dollars, so 8-10 vehicles for destructive testing is a drop in the bucket.

But I still think this could be a useful alternative for boutique cars, or as a way to circumvent the 25 year rule by allowing for the acceptance of virtual data.

doodon2whls > Snuze: Needs another Swede 02/19/2015 at 11:04

0

Hundreds of Millions for dev't costs ? Not sure they are that high - maybe for a clean sheet program.

The cost of developing, refining, and analyzing the models and simulation data is not cheap either, but it is amortized over a large volume program and eliminates many prototype vehicle tests. The designers can get within 80% of the design virtually and then tweak with bench and vehicle tests.

The cost to build a simulation model that has been Verified and Validated to pass muster with the homologation entity would be prohibitive for a single vehicle import.

Rock Bottom > Snuze: Needs another Swede 02/19/2015 at 11:41

0

All great points. Maybe one day we'll get to a fully virtual certification process, but you're spot-on about the cost having to balance out. For now, cars really are cheap to build and test to failure (especially compared to even a very small ship). Maybe it'll be all-virtual in 10 years, of course that's what the CFD guys have been saying to us wind tunnel guys since 1970!