What is the guidence on SpaceX rockets based on ? PID's?

[Patrick]

Are the vertical recoveries were seeing based on PID loops?

My Professor insists so.

[Mads Barnkob]

I think you can get the "best" idea from BPS.SPace on youtube or this paper I just found on google: https://project-archive.inf.ed.ac.uk/msc/20172139/msc_proj.pdf

The conclusion on the paper is that the basic PID presents the best landing performance, but highest fuel consumption. It was also my own idea that they were using some kind of model prediction or gain scheduling. I doubt a PID with 3 static settings could do the job

There is also a lot of speculation and interesting links to papers in this thread: https://www.reddit.com/r/spacex/comments/4ewrpq/why_classical_control_for_the_rockets_control/ f.ex. to JPL G-FOLD system descriptions: https://www.lpi.usra.edu/meetings/marsconcepts2012/pdf/4193.pdf

[Patrick]

Ok done more research and desperately want to understand this stuff:

Also, is it possible to have two PID's in series ?

[Mads Barnkob]

Yes that is called cascaded. But that is even harder to control, due dead time from the first is likely a 2.order system and adding that to another PID will result in a 3 or 4.order system. Tuning gets harder as change on any of the PID affect the process, but its hard to evaluate on which PID parameters is optimal.

[Patrick]

Ok thats what I thought, ganging things up often makes them goofy.

But with a common quad copter do you think its possible to have a separate PID for roll pitch and yaw ? or would they continuously interfere with each other ?

[Twospoons]

Since roll, pitch and yaw are, in theory, orthogonal, then yes you can have independent PIDs for each. In practice you would have to ensure your gyros are aligned correctly, or there will be some cross-coupling that will have the PID's fighting each other.

[Patrick]

Is it possible to, based on real time flight data, change the P, I, and D values of a loop to predetermined values know to be more ideal at a particular region?

For instance,

At Angle 42 degrees, speed equal or less than 11m/s P 4.2, I 0.0, D 1.1
At angle 11  degrees, speed greater that 11m/s P 3.7, I 0.1, D 0.7
And so on....

I would plot a g(x) nonlinear function for a micro-controller to execute. like a lookup table.
 or parallel P(x), I(x), D(x) curves all plotted for the micro-controller to pull from using real time air data from Pitot probe/ gps/ interial data streams.

[Mads Barnkob]

In regular automation PID we call that "gain scheduling" and is often used in process startup scenarios or to run with different settings in winter/summer.

You could also look into feed-forward PID if you want a short of "prediction" to not solely rely the process value changing, but you take into account the change of the input variable that will affect the process value.