Hacker News Reader

Comment by ta1234567890 | Hacker News Reader

> The next day, our flight profile was duplicated on the SR-71 flight simulator at Beale AFB, Calif. The outcome was identical. Steps were immediately taken to prevent a recurrence of our accident.

Wow! Super impressive that they could simulate that. Didn't know that was possible, especially in 1966.

Is there any current simulation software (available to the public / off the shelf) in which you could accurately simulate something like the situation from the story?

View on HN

A bit of searching failed to turn up anything on how the simulation was done. I'm curious if it was all digital, like a modern simulator, or if it included analog simulation.

It was built by Link, which had been building flight simulators since 1929 [1], so certainly had extensive experience with analog simulators.

According to Wikipedia's article on analog computers [2],

> Analog computers were widely used in scientific and industrial applications even after the advent of digital computers, because at the time they were typically much faster, but they started to become obsolete as early as the 1950s and 1960s, although remained in use in some specific applications, such as aircraft flight simulators, the flight computer in aircraft, and for teaching control systems in universities. More complex applications, such as aircraft flight simulators and synthetic aperture radar, remained the domain of analog computing (and hybrid computing) well into the 1980s, since digital computers were insufficient for the task

It sounds like the '60s, then, were well before digital flight simulators were common.

[1] https://en.wikipedia.org/wiki/Link_Trainer

[2] https://en.wikipedia.org/wiki/Analog_computer

I'm going to have to guess that CFD simulation above mach 1 that is good is probably not something that is publicly available but from reading RUAG's website they use open source software[1]. From my experience with CFD in F1 the main issue you have with CFD is correlation but maybe driving a car around a track is harder to simulate than going really fast in "clean" air.

This reason is what leads me to think that it is a major issue and is why both China and the United States started going into mach 5+ with autonomous craft to gather data for correlation.

[1]: https://aerodynamics.ruag.com/en/aerodynamics/simulation-ana...

For low speed flows like in F1 the viscous effects are much larger and this can make flow prediction more difficult. It’s actually pretty easy to predict the shockwave pattern from 1D theory without the need for CFD [0]. The boundary layers are much thinner and you can assume inviscid flow at first. At hypersonic speeds the interactions get more complex however and you’re dealing with a lot more materials science challenges so CFD is important in predicting thermal loads and shock boundary layer interactions.

[0] https://en.m.wikipedia.org/wiki/Oblique_shock

The CFD simulation in F1 isn't really a track simulation. Normally a straight or curved domain (for car in yaw) is used. Computational limits in the rules mean simulating dirty air, like when following a car, is incredibly inefficient.

What is correlation?

https://en.m.wikipedia.org/wiki/Correlation_and_dependence

Kinda makes you wonder why they didn’t simulate it before they flew the test.

The test itself wasn't fatally flawed, it snowballed when they had an unrelated engine malfunction.

Several minutes into cruise, the right engine inlet’s automatic control system malfunctioned, requiring a switch to manual control... Without proper scheduling, disturbances inside the inlet could result in the shock wave being expelled forward–a phenomenon known as an “inlet unstart.” That causes an instantaneous loss of engine thrust -- TFA

It wasn't unknowable or even an unexpected situation, those engine unstarts were common until they worked out the control system dynamics. They were moving fast and killing test pilots right at the beginning of being able to use computers to simulate flight. Exhaustive simulation of all circumstances might not have been possible with the available computer time, or honestly it is quite possible nobody thought to do it.

Test pilots though are generally aware of and on board with the risks. Unfortunately simulation isn't all it's cracked up to be and very often aerodynamic models are trustworthy only after they have been validated experimentally for the specific scenarios they apply to even now. Nevermind half a century ago.

Right, the engine unstarts were a common and expected problem, but the key is that I don't think they were in any way predictable, other than being limited to certain condition regimes that included basically everything except subsonic flight.

That would be a lot of unknown conditions to test for.

They probably used an analog computer for the simulator.

Really interesting technology that is barely remembered today.

http://www.analogmuseum.org/english/examples/lfsim/

You probably don't need simulation. Just the inlet dimensions and the freestream air properties will tell you when the engine should unstart.

Reply on news.ycombinator.com