gatnerd said:

The Air Force was recently saying much the same - that computer simulation is not an adequate replacement for real world testing.

The situation did not really change...

https://en.wikipedia.org/wiki/NASA_X-43

"NASA's first X-43A test on June 2, 2001 failed because the Pegasus booster lost control about 13 seconds after it was released from the B-52 carrier. The rocket experienced a control oscillation as it went transonic, eventually leading to the failure of the rocket's starboard elevon. This caused the rocket to deviate significantly from the planned course, and it was destroyed as a safety precaution. An investigation into the incident stated that imprecise information about the capabilities of the rocket as well as its flight environment contributed to the accident. Several inaccuracies in data modeling for this test led to an inadequate control system for the particular Pegasus rocket used, though no single factor could ultimately be blamed for the failure.^[6]"

We should keep in mind that even a 6 DoF model can't predict every time something as simple as the the flight of a bullet, so being able to predict something more complex is pure hubris...

EmericD said...

We should keep in mind that even a 6 DoF model can't predict every time something as simple as the the flight of a bullet, so being able to predict something more complex is pure hubris...

To expand on this, if the physical test uses a bullet exactly as designed, at least to significant tolerances, and the air it was flying through was still and at constant temperature and pressure, and gravity was constant throughout the flight and the starting point (say muzzle exit) was consistent shot to shot  and all of those factors were covered in the model, you'd expect the model to be accurate. Bit of a tall ask for test range and conversely it's a bit of a tall ask for a model to account for all of those being variable. I wouldn't say bullet flight is simple.

In order to make it possible to run the model in a reasonable periods of time, a lot of these factors are assumed or abstracted and measurements aren't always correct. Then you've got very small effects that can build up over time and change the end result - this was a big thing in weather modelling in the '90s, where they changed from one set of starting conditions to get one set of results to a set of variances from measured conditions to get a range of results and they can comment on probability. This kind of modelling is dependent on computing power, so it becomes more accessible as computing power increases. Getting a good idea of how much to change the starting conditions requires test evidence though.

Simulation allows you to look at areas you can't test, or can't measure, either easily or at all. Ultimately you've got to test to prove that the end result matches (or not)

If anyone tells you that you don't need testing and can do it all on computers is trying to sell you something. Probably simulation software. 

RovingPedant said:

Simulation allows you to look at areas you can't test, or can't measure, either easily or at all. Ultimately you've got to test to prove that the end result matches (or not)

If anyone tells you that you don't need testing and can do it all on computers is trying to sell you something. Probably simulation software.

Every engineer o sciencist knows that since they allready learn it at the university.
As mentioned this is nothing new.

The problem is nowadays decisions tend to be made by people who are lacking in basic sciencific and technological understading. But they do understand "its soooooo much cheaper". Which is also exactly how far their understanding reaches. So they use the only benchmark they can comprehend for their decision. At the same time they are very poor business economists (if they weren't they wouldn't have to work for the government in the first place) because they never understood full cost calculation and holistic approach.
Since we also live in times where very seldom someone is held resposible and evasion of responsibility has been up lifted to an art form by the decision makers there effectively are no consequences for stupid decision. Negating any learning effect and any chance of personel changes in the decision making proces.
That is the root of the problem. The fact that simulations are limited is merely a environmental conditon.

RovingPedant said:

To expand on this, if the physical test uses a bullet exactly as designed, at least to significant tolerances, and the air it was flying through was still and at constant temperature and pressure, and gravity was constant throughout the flight and the starting point (say muzzle exit) was consistent shot to shot  and all of those factors were covered in the model, you'd expect the model to be accurate. Bit of a tall ask for test range and conversely it's a bit of a tall ask for a model to account for all of those being variable. I wouldn't say bullet flight is simple.

To be more precise, here are 2 cases of bullet behavior I observed, and that you simply can't predict:

- when you shot the Lapua HPS 170 gr 7.62 mm bullet at 800 m from the SCAR-H PR (20" barrel; 1-in-12" twist), the bullet holes are perfectly round, when you shot the same load from a HKG28 (16" barrel; 1-in-11" twist) most of the bullets are flying sideways at the same distance.

- the 168 gr SMK fired from a cold hammer forged  20" barrel with a 1-in-11" twist are flying sideways at 1000 m, while the same bullet fired from a button-rifled 20" barrel with a 1-in-12" twist is making round holes on target.

EmericD said...

To be more precise, here are 2 cases of bullet behavior I observed, and that you simply can't predict:

If anything interior ballistics, to get the starting conditions, are even more complicated because you need propellant behaviour and a lot of high deformation and surface interaction modelling for projectile/bore interaction. None of which are done very well at the moment.

Given enough time, money, and supporting tests, you could probably predict what's going on there, but it would probably be cheaper and quicker to test which barrel/ammunition combination works and select which you want unless you really needed a particular solution.

Fascinating examples.

It makes me shudder a bit contemplating what else we now "Know" based on computer modeling.

Much of Climate Change predictions and subsequent energy policy seems to be based on computer models of what the planets weather might be in 30-100yrs... 

https://archive.ph/LJKaa

Tacking closer to our area of interest, nuclear weapons testing is now based largely on supercomputer simulations. And increasingly they are trying to simulate things they have little to no test data on for calibration.

gatnerd said:

It makes me shudder a bit contemplating what else we now "Know" based on computer modeling. Much of Climate Change predictions and subsequent energy policy seems to be based on computer models of what the planets weather might be in 30-100yrs... 

Well, we all know that those models are "false", because they predict an increase of the temperature in the lower AND upper atmosphere due to greenhouses gases, when the measurements show that only the lower atmosphere is heating, and at the same time the upper atmosphere is cooling.

But that does not mean that those models are not useful (remember "all the model are false, but some are useful"), and probably someone in the years to come will increase the reliability of those models, as some new mechanisms are proposed to explain the cooling of the upper atmosphere due to CO2.

gatnerd said:

Tacking closer to our area of interest, nuclear weapons testing is now based largely on supercomputer simulations. And increasingly they are trying to simulate things they have little to no test data on for calibration.

Well, that's probably because a nuclear weapon is the "perfect" form of a large energy release in a very small volume, so all the equations collapse in very simple forms, as you don't have to take into account energy losses, viscosity, aerodynamics...

For example, G.I. Taylor (a British physicist) was able to evaluate the energy released (a highly classified data) during the Trinity atomic bomb test with just 4 pictures of the test and a ruler. This really upset the US Gov that thought that Taylor was a spy, when he was just a genius.

Although aviation related, there are a couple of articles linked below, which may indicate that there is finally some healthy scepticism being applied to the rush to utilise the latest snake oil in procurement:

https://www.thedrive.com/the-war-zone/air-force-boss-gives-reality-check-on-over-hyped-digital-engineering-revolution

https://www.thedrive.com/the-war-zone/t-7a-delays-compound-pilot-shortage-expose-digital-engineering-pitfalls

We have probably all seen the effects where promising technology has been over-hyped and actually unfairly reputationally damaged and blamed for programme delay. This is often to justify unrealistic timescales on underbid/underfunded programmes, driven by under pressure and technically illiterate MBA managers.

Your example of inconsistent bullet behaviour is a great practical example of why there needs to be a proper balance between theory and testing, even when designers are convinced that testers love to break shiny things (as a tester, I must confess that there is element of truth in this!). 

TV still hasn't given up, partnering with FN for 6.8 TV conversion for M240.

https://www.thefirearmblog.com/blog/2023/06/02/true-velocity-partners-fn-america-develop-m240-conversion-kit/

Possibly for Australia or some other contract?

gatnerd said:

Possibly for Australia or some other contract?

Possibly, but I think it's more likely that TV is just hoping to cash in on the M240 conversion they developed a couple years ago and get a contract somewhere, sometime.

Given Australia's increasing alliance with the US ( see 3:33-4:10 in https://youtu.be/gGLvUyzEMj0?t=213 ), I expect they would want to use the same type of 6.8x51 ammo.