TVC Model Verification
ฝัง
- เผยแพร่เมื่อ 22 พ.ค. 2024
- Shout out to blue light glasses for A. Helping me sleep better and B. Ending the “haha he look lik elon musk” comments???
Scout D1 Flight: • Scout D1 | Flight 2
Help support BPS.space: / bps_space
Second channel, mostly for KSP: / musicmakr
For more info:
/ joebarnard
/ bps_space
/ bps.space
/ bps.space
www.bps.space - วิทยาศาสตร์และเทคโนโลยี
![MOBYE - มีใจทำไมไม่จีบ (ASAP) [Official MV]](http://i.ytimg.com/vi/dCOdz92FU18/mqdefault.jpg)
L o O k D e E p i N t O m Y e Y e S
Ok cool
I’m planning to make a liquid fuelled model rocket engine sometime in the future. I’ll update in 2 years, and let u know on progress, if not sooner
I am currently building a micro hybrid motor with ten grams of delta m. (N2O /PE)
If you are interested in a throttleable version, I could maybe develop one and send it to you.
"Hello everybody, if you don't like the video please make sure to unsubscribe and hit the dislike button!" lol
This, is one of the reasons why I really like this channel. I really like your sense of humor, and I kinda need it right now...
"If you cannot do great things, do small things in a great way." -Napoleon Hill
Way too close to Valentines Day for that sort of joke :-(
"this is interesting" - gets exciting
"Actually it's not" - doubting but still interested
"It's about control theory" - explodes in excitement
Controls Engineer here. Really interesting video
Control Engineer here! really nice video, you should really consider do more videos on those simulink modeling/tests!
Just realized Joe used an extra Signal as a mic stand.
This was genuinely exciting and interesting! I've just been going back to and learning control systems in theory and to actually see it being used and modelled is cool! I refuse to believe otherwise. Great video!
I am not a control engineer.
But now I want to be.
Props for tackling performance validation. I’m almost disappointed we didn’t get to see an example of what your full model looks like at the end to see how complex it can get.
Justin Treptow yeah that’s what i was hoping for
This was actually super interesting and helpful because I've been looking at buying that same type of Sony cam to do some high-speed analysis of a pneumatic-mechanical launcher device where I'll need to fire it, see how long and for what durations various pistons and so-on are cycling for as I tune pressure and spring rates and volumes. I've been assuming setting up video in Premiere and using it to analyze the footage to get things like response time, velocity, etc is possible; but I've never done it and had no clue what the best ways to actually start collecting any sort of measured data from the footage would be.
Your video really helped with that a lot and I think it was very interesting, thanks for going out of the way to put it out for us!
You might consider drawing a vertical line on the movable unit and mount a clear protractor in front of it so you can see exactly what angle it's at at each frame of the video.
See now I feel like it's a challenge. "I'm not dumb. I'm gonna watch this whole video and I'm not going to be bored! Not for one second! Ha!"
It would be interesting to do a frequency response test using a sine wave on your tvc mount. That would give you a good model.
Yea that's what I thought.. put a scope on it
Yea, if he put a distance measurement sensor (capacitive or laser) and used a sufficiently wide noise signal he could get the gimbal dynamics in one direction. But considering he knows the model (a second order inertial mass rotating about an axis) so all he needs is one profile with acceleration and deceleration to estimate inertia and friction.
@@itzdaman A simple potentiometer is much simpler and more informative.
You know you should be taking notes during the video when Joey B puts the glasses on!
Also this video has a lot of connections to the new video we just started filming :D
All the videos you said were boring have never been boring yet they're all super interesting!
Very interesting actually! Ive been so curious about how you actually test and come up with all this. Id love to see more about matlab!
Thumbs up for the honesty up-front :) I'll watch it. Great Vids.
This video was interesting, informative and inspiring, like a first-rate science class. Thanks for everything you do at BPS.
Very well done Joe. I was hoping that the video would of been longer.
This video is excellent.
You can add an optical time element by blinking the LED at a known rate instead of just turning it on. Also, if you shoot long (move the camera away, stop it down, and zoom in) you can increase the depth of focus. Lastly, you could move the thruster to known deflections, and with the camera as your viewport, mark a sheet of paper mounted behind everything (relative to the camera). This would let fairly easily measure deflection in the video.
Not at all boring. Just watching your problem solving skills. And you multi task so well.
Thanks for the interesting video Joe. There is a piece of open source software called Kinovea that is used for measuring sports movements, you can track points, measure velocity, get angles etc.
Might be helpful to save you some time... Keep up the good work!
How are you holdin up bud. Hope the new year will bring up your spirits.
I once worked as a Process Controls technician (not engineer). The equipment I worked on controlled the process of making paper, or in my case, the process of making fiberglass mat which uses the same equipment as in paper making. At the end of the machine the computer monitors various parameters of the final material like thickness, color (pink, if that may be a clue), dryness, etc, and makes adjustments to the process to make the process run on target, or back on target if it strays. The system uses a computer algorithm to prevent control oscillation around the target called a *Proportional Integral Derivative algorithm* or PID in the world of process controls. It uses all you are talking about, input delay, machinery response time, sensor feed back, and (in my case, but maybe analogous to vehicle inertia in a rocket) material response delay. Maybe this is what you are avoiding talking about because the formula itself IS incredibly, mathematically tedious and hard to grasp just what it does. If not, you might want to investigate the implementation of a PID in your controls. When (mathematically) tuned properly your sensor graphs will show one control spike and a curve to target. If not, the graph will show "ringing" or a decreasing oscillation until steady on target giving your rocket that visual "wobble". The most difficult part of it's use would be to account for control effectiveness as speed and aerodynamics increases.
thank you for taking the time to write this comment. extremely informative! i always wondered how control systems handle the situation of an output parameter oscillating about the expected target value. time to dive into PID theory :)
Where can I buy that NASA hoody?
Hey, a video about control theory!! I'm a controls engineer at General Electric working on the GE9X! Gotta say I've loved watching your videos for a long time and definitely appreciate the love towards the controls community (even as boring as we are 🤓)!
I wish the RS25 would do this in kerbal, instead any rocket that uses it just fishtails all over the place. Love your vids, and your glasses. Keep up the great work!
2:33 Signal Microphone Holder
Control Engineer Here...I have never thought about using a high-speed camera to observe/measure the angle...Really great job and looking forward for such interesting videos in the future.
why do I actually enjoy watching this so much
Don't be so quick to call this video boring. As an mechanical engineering student who just took (and really enjoyed) a controls class, the controls videos are some of the most interesting on this channel!
Honestly I love seeing this stuff, the polished content is fun but if I really want to do this, then videos like this are invaluable.
You should tap off of the wiper of the pot in the rc servo and analog sample that. That would allow you to see real time what the servo is doing. You could also use that in the control to know when a command is complete and avoid sending multiple inputs that the servo isn't fast enough to handle. It would also make it easier to model small step vs large step responses.. .02c
6:19 "Not even minus, just plus!" This is the real value of this channel. Complete disclosure and truth.. :)
Pleasse more videos about the control model of the rocket, they are very useful!
You can open up the servo and run a pair of wires off the wiper and directly record what the servo is doing. You can find some servos that people have preconverted, but I'm pretty sure you have a particular servo you want to stick with. Sacrifice one servo and get raw data. Various servos also have different dead zones and you may find moving from 0 to 1 degree is different than moving from 1 to 2 degrees... lots and lots of things to double check. Cool work!
Working on my own rocket now, thank you for this information!
This is much more interesting than finishing my essay that’s due tomorrow
You had me at "unless you're a controls engineer." Great vid!
You could use blender's frame motion tracking functionality to characterize the transience of the response as well. It literally makes a graph of the motion vs frames.
Im just listening to you as i build my 20' wingspan RC Bomber plane for FFE20 combat in July. Designing & 3d printing servo mounts.
Nice! Are you basing off of any aircraft? I want to build a 5 foot mosquito bomber just haven't got gotten around to it yet lol
Actually this is not entirely correct.
The main issue is the diagram. The reason we put the controller before the dynamical plant (not only with respect to adding noise) is that the error is calculated by subtracting the reference, thus even if the reference is 0 there is still no arbitrarity to place the controller before the plant even if it is a loop. In this case it also makes it easier to model delay, since the plant contains the delay dynamics, so you can put it in front if the gimbal plant and encompass the entire thing as a new plant, and build on that.
Secondly, I'd like to suggest to do it continuously and model the gimbal as a simple system. Instead of using a discrete delay only, add a transfer function of the speed of the rotating gimball and measure that with a camera. If you know the speed (or even the acceleration) profile, you can identify the inertia (or friction) and use that as a second order transfer.
Hi, Joe! Thanks for an interesting topic! I'm wondering, are you going to continue your series with Blip & Blop?
I would definitely be interested in some videos on the matlab simulink stuff. I’ve been using matlab for data analysis stuff, and really don’t have any clue about what simulink can do or what it’s intended to be used for. I’m incredibly intrigued to learn more.
In all seriousness I found this very interesting. I'm looking forward to more of this stuff
You fact-checked yourself on that throwaway LM comment?? You were off by 1 out of 2 stages on the vehicle! This is the kind of effort I come to this channel for! xD
keep in mind that the quality of your servos will play a great role in there response time and accuracy .
Most servos will happily take an update rate well in excess of 50Hz. Even the cheap ones. They also don't mind a slower update rate, or an irregular/inconsistent update rate. You should be able to leverage this info to tie the servo updates to the output rate of the control loop.
Weekend engineering? I’ve also been tuning my Meteor M-2 satellite receiver setup. Not much to see over Scandinavia since there’s a big storm moving over Europe as of writing this.
I truly do not have anything better to do. This is truly fascinating stuff .
The servo itself has a feedback loop. For large angles you will reach max speed. That is called slew rate limitation. You can see that in your video! After some degree you have constant speed some degrees. Actually you probably first have a slew rate limitation also in the acceleration fase. If you do this excersise with different target angles you might see it!
The slew rate limitation is an extra -90degrees in the frequency plot (bode diagram).
I use to make the laplace modell as: one clean time delay e^-L and one time konstant (1/1+sT). Normally a rc servo has typicully 60degrees/s max speed. That is 83ms for 5 degrees. With infinite acceleration that gives 17ms for the linear response! Cross over frequency is approx wc=3/Tm=3/0.017=176rad/s.
Some guys suggests frequency response measuring. I think step response analyse are sufficient with these identifications. Knowing L and T separately gives you a mutch better bode plot!
Put your camera close! And put a millimeter ruler just behind. Then you can plot the movement i figures versus time. And measure also small angles!
With love! /BG
“I don’t know...unsubscribe....”. I like your confidence, sir.
Import a picture of a ruler and move it next to the rocket. Then rotate that image for a decree or two and now you have more precise idea of the angle. Also you could set your sequence settings to your footage, so the time info would match directly = no calculations needed.
This was actually interesting Joe great job
I was under the impression that the TVC of the LEM descent stage was used only to compensate for changes in center of mass, and the actual steering was done with RCS.
I subscribed because of this video. Interesting and honest.
I'm also interested about those simulink models! I have basic knowledge about matlab, but I haven't used simulink
Joe, in your control feedback modeling, do you account for the rate of change of mass & center of mass of the rocket system in adjusting your TVC?
Very interesting :)
But i would not believe, that "forceless" position control (ideal conditions) from the servo takes same time as position control "on-load" with thrust.
The deviation from position control should be more higher with thrust, too.
That's only my experience in working with these servos.
Please continue working on this stuff - it's very interesting watching your videos.
Yes I’ve been waiting for Joey b to teach controls
The term we used in controls class was actuator dynamics.
There is easier way to do this:
servo has a in built potentiometer to measure angle
open the servo and solder a wire from middle pin of that potentiometer
and measure the voltage between this pin and ground with analog to digital converter pin of your microcontroller
you need to only map the voltage to angle
you can get immediate angle feedback on the microcontroller itself which can be used to build a small pid subsystem for tvc
This is genuenly more interesting than the usual updates
I'd caution the idea of going to 1 degrees in a 5th of the time to get to 5 degrees. The control law in servos outputs less power to reach a closer angle (the proportional control aspect), so the angular acceleration to approach smaller angles will be lower. If anything at small angles this will probably lead to the reaction times being exactly the same, not shorter, for smaller angles. A way to practice this concept with servos is try modifying them for continous rotation, a small angle input results in slow rotation.
Ay man...even if I find these boring I shouldn’t just unsubscribe. I support your channel and that means I’ll give you a like and stay subbed. I want to see your success and sometimes that means you actuate a servo at 5° in 1000fps.
Great work!
Do servos perform the same when under a load (rocket motor is active)? Was there any voltage dips or spikes during actuation, servos are sensitive to voltage in relation to angular velocity.
Roughly the same performance during a burn - the servos don't resist any force from the thrust of the motor, they just change its direction. I tested with a loaded motor to make sure the mass that the servos were moving was roughly the same.
You can also simply go to the manuf. web site for the servo and get the data sheet that will tell you its speed. But ya, doing it real world is always best.
More of this pleaase:)
Did you think about measuring the frequency response of the actuators to get a transfer function? Wouldn't this be more useful for the model?
Very interesting indeed. This is the hobby I would like to have but that I'm too lazy for to actually take up. You would be an inspiration if I could be inspired 😐 Don't loose your fun in doing these things, they're amazing 👍
Came here for the technicall stuff. Stayed for those closeup eye shots.
As a building automation/controls technician, I find this interesting!!
Very interesting subject, keep it up
What about backlash in the servo gears? Might explain why returning takes a bit longer. Compensation should be easy as long as you remember the previous direction of movement.
Aspiring to be a better controls engineer. Great video
Another great video... and not too boring actually :-)
You should do this same test under load..... Mount it in a test platform, video with high speed, and light a motor. Does the Actual load cause the actuator to take longer to move?
Not boring at all, great for PID fans. These vídeos for control are really nice.
This is actually super interesting to me.. Love it
I am not a control engineer, but a medical engineer ;) Very interesting presentation. It is more interesting, than my bachelor control systems lecture was 😅
I guess boring is contextual, but a nice slow and detailed process is still interesting. I'm following along while cooking up my lunches for the week on a slow Sunday arvo
What if the sensors is not in the object itself but the multiple camera observing it, the control would be relative, and might miss a feedback along the way but the scope would be huge
I wanted to add thrust vector control to my rockets but I dont have a 3d printer. Is it possible you could make and sell a kit
TVC will always stand for "Television Centre" in my mind
What about taking signals from servo's potentiometers?
Were the glasses to emphasise the geek index? 🤓 (I wore glasses for 30+ years before LASIK).
Hrmm..... Would be a bit more accurate if you could actually monitor the encoders at the servo - that's how we usually do it. Then you can excite both closed and open loops to really get a good handle on how the vanes will respond... (On things like Giant Magellan Telescope, we back this up with optical sensors to also model the error between the servo encoders and the actual actuation target)
Isn't the reaction speed also dependent on the amount of moment the actuator must apply?
Oh hey it's that milk guy
Quite interesting! I hope you one day get to use this data on a real rocket! I'll follow your progress ;)
Ohhhhh i love your hoodie in the video. Very interesting video. Do you think you will ever work a liquid fuel rocket?
Useful ! More on rocket simulation with simulink please :)
This seems a bit overcomplicated to me. I would have just attached a scope (analog probe) to the servo's potentiometer and another to the LED control signal used as trigger.
Secondly using step function is not the best way to identify systems, this assumes it is linear. However using a chirp (increasing frequency sinus) you can tell even the transfer function.
Servos (that i worked with) never settle in the same position which is another parameter (precision).
Note that LEDs are not turning on immediately, they just like anything else have a delay, might be negligable in this case but it exists.
How do you calculate de transfer function of a rocket? Can you teach it to the world? Your work is awesome!
How far do you think your thrust vectoring design could be extrapolated into high power rockets (assuming it is brought to scale and materials are upgraded)?
How to get the transfer function of my rocket? What's yours? Can you do a video on it?
Wouldn't it be more accurate to measure the delay till the servo starts moving, the acceleration and the maximum speed of the servo?
Wish we had Matlab when we were doing control theory way back when. Wow
We need more Simulink Modelling videos!
I think a video on how to tune a PID in Matlab would be awesome
One recommendation is to utilize computer vision (python/opencv) for a test like this to validate 5° angular displacement and time duration(check LED illumination). You could prob get away with just a webcam. You may need to calibrate the camera and obtaining the cal matrix.