danvel

ADI: Your integration graphs are wrong. You don't have to reverse sign. See correct graph below.

ADI

Danvel ......

I'm totally confused now .... damn it!

The way I understand it is: the more we move away from setpoint (centre), the more INT compounds (INT=PreviousINT+NewINT) to apply error correction drive to our 2 motors.
But in your graph, INT is still compounding and increasing and adding even more error correction, even when DF has reversed and is heading back toward centre again.

Cheers ADI


PS: I think I've figured out why I'm confused. (apart from being a dumbarse)
This is not a closed loop ! Your graph is the result we'd get open loop.
INT has gone crazy and run away on us because it's open loop.
With a properly tuned closed loop, our results will be dramatically different.
Am I correct in my thinking now ? Appreciate all your help Danvel.

Cheers ADI

tve

ADI,

The Integral term in a PID can suffer from what is sometimes called wind-up. It always has lag associated with it because it contains a series of samples from the past... Preventing wind-up can be handled multiple ways. One way is to make the Integral coefficient variable, such that it is less as it approaches the setpoint. (this is normally the function of derivative term). In this case, you are integrating the derivative term (your gyros are the rate-of-change). You can (and probably should) also put limits on the value of the integral term. Yet another modification is to only sample a "window" of N number of your most recent samples for your integration.


Regards,
tve

ADI

Thanks tve. I'll get to grips with this PI/PID stuff if it kills me.

Cheers ADI

zarx

Here's an update from yesterday,
Diode ON the motor not on the circuit board. Would you believe that the inductance in the wires from the board to the motor was enough to cause some crazy emf and IR drops.
Aussi, I am switching at 2 khz but now that you mention it I am thinking it might be good for us to switch faster. The ATmega32 microcontroller has built ing phase corrected pwm output, but the frequency options are limited to 16 khz, 2 khz, and lower. I will try 16 khz and let you know how it works out.

TVE, i have used "star-ground" techniques to prevent ground loops and my board is fully ground planed. Here's some advice for you however - dont simply use large caps for your filtering. Large value electrolytic caps have poor high freq response. Use a small value ceramic, or even better tantalum cap to help remove impulse noise. I am adding that to my board now, so we'll see how well it works out.

ADI, if possible implement a circular buffer large enough to hold enough samples to represent the last couple seconds or so of A/D conversions. Then use the sum of this buffer for your integral (don't think this will be obscenely large, the buffer is filled with small signal setpoint error values which are above and below zero, so it shouldn't get too large). This provides a natural lowpass filter on the integral and helps remove the drift that occurs by integrating sensor errors. Remember, integrals are the TOTAL AREA UNDER A GRAPH. So with the graphs provided, as long as the signal is above zero, the integral will continue to increase. Derivatives are the slope of the graph at that point, and in the case of the DF this relates to the rate of change of the gyro output (this is angular acceleration)

Cheers

Spaceclam

Jesus guys, i think i only understood about 10% of your guys's messages.
Greetings from about 9000 feet, about a thousand below the human ceiling for out of shape, american fat-asses like me (except for the fat part)
I should be able to keep in touch, but only morning and evening. the package arrived today according to the tracking number, but too bad i will have to wonder about it for another 3 days becuase my dad wants to keep me in suspense and NOT use the caliper meter and tell me.[:@]
i will probably be exhausted at the end of each day so dont expect any particularly intelligent responses out of me for a day or two, although i will try my best to keep up.
Cheers and good wishes,
Clam

ADI

Zarx Wrote:

>and I was wondering if any of you guys have had any success with your PID controllers..........
................. Also, I'm getting serious noise on the regulated 5v power to the gyros.>

Hi zarx,

Welcome aboard. Good to have some new DFers here.
Check out Danvel and Crac's DF videos for PI control success. Videos speak for themselves.
Re supply line noise .... I chose a different route and utilised a DC-DC step up convertor (MC34063A). 7.2v - 10v out. The 10v output is used raw to drive the motor FET gates and then the 10v is regulated back down to 5v (78L05) for micro, D/A and A/D etc. and 3v (LP2950CZ-3) for Tokin gyros.
The DC-DC convertor does a pretty good job of filtering all by itself and the other linear regs provide some more. The other advantage to this system is that the 10v holds up with input as low as 4.5v.
Here's the finished PCB. http://www.rcuniverse.com/forum/m_14...41/key_/tm.htm
Post# 1226
You can see the DC-DC inductor standing up next to MC34063A chip

Cheers ADI

Crac

Hello Guys !
To have some fresh news of my DF : http://perso.wanadoo.fr/grzflyer/
- new mechanical design (videos coming soon) : it flies much better !!
- some adjustment in the program : it is more stable
- already in french & many pages in construction [&o] but many photos...

Just an information for "PID" maker : my DF use only Ã* PD... (for the moment)

<-- Crac flying several times per day

tve

zarx,
You never did say what freq and level of noise you were talking about, so I was guessing at your problem. Star ground techniques are exactly the right thing to do for this circuit. Of course you are going to get IR drops across the conductors supplying the motors! These are some pretty good currents.... There won't be much (if any) detectable radiated EMF though.... I tested this in my case with an RF spectrum analyzer. Don't get worried about affecting your receiver anyway.... Much more important to get the power supply noise under control. Actually now that I think about it, you will have some radiation from the brushes on the motors... I am using a brushless setup (no sparks ). Your case might vary.

Yes using ceramic .1uF (or similar) bypass caps on every IC close to the power pins is important. The large caps ARE very necessary in my case (and probably would help you also). In fact putting a couple of those low-ESR 1500uF 35V caps in parallel at the regulator input will help reduce ESR even further if you can afford the space and weight. You also typically need a larger value cap right at the output of the regulator for stability (somewhere around 10uF)

I am drawing up to 30 amps per motor.... try filtering those drops with only small ceramic caps! It wont work.

In short, The larger caps provide charge-supply. You need this in addition to decoupling and filter caps. Try it and see! Try to get Low-ESR types.

Good luck!
Regards,
tve

tve

Hi Crac,

I also took the PD approach... works great!

ADI

EUREKA ! I finally figured it out !

Danvel ..... I closed the loop in the Excel spreadsheet with a percentage feedback.
Does this look familiar ? (Ignore the tuning)

zarx

You can think of a PID controller as the suspension on a car.
-The proportional component is the spring - the further you are from the set-point the more it attempts to bring you back
-The Derivative component is the shock - It damps the oscillations and keeps the thing stable, but too much and the response is S L O W and sluggish to a step change
-The Integral is hard to represent physically. Technically it would be the mass of the suspension system but I dont think that is intuitive. But put it this way, The integral of angular velocity is just the angle itself, so the integral component is what brings the DF back to horizontal when it has tilted a few degrees off. I would think that including the integral would be almost a requirement for extremely stable flight.

I should let you guys know, our team has the ambitious plan of trying to stabilize our DF without human input. We have no receiver! We use a ribbon cable to provide simple human control for testing, but after that, its all on it's own.

Question for you guys though: what are the gain constants you use for your controllers?
Before you respond, make sure we are on the same scale here. I have taken my analog input, and converted it into a floating point voltage in the code. Then, using the specs, I approximate the current deg/sec (I think the tokin/murata gyro has .8 mV/deg/sec). Therefore my gain constants are acting on the actual estimated deg/sec value that the gyro is getting.

tve - Yes obviously the large caps are required, but low esr and high value caps are hard to come by. A cheap and easy solution is to put a number of different types of caps in parallel. The large caps keep constant the voltage drops due to current draw, however (at least for me) the inductance in the motors causes huge spikes when the fets switch off. After lowpass filtering with an electrolytic, my voltage is remarkably constant...until you zoom in on the oscilloscope. I usually put a small tantalum cap in parallel with the large cap as they tend to have an esr about ten times smaller than even the best electrolytic.

tve

zarx,
It would probably very benificial for you to go back and read some of the previous posts on this thread. There are basically two approaches to control feedback here. One is integrating output from the gyros to achieve good stability for RC flight. The other is to use some sort of level-detecting sensor in conjunction with the gyros to achieve flight without human input. I, like your team, am working on autonomous flight. If you want to fly for an extended period of time without human input, you will need more than gyros. A stable hover can be acheived as Danvel has proved, even the best gyros (fiber optic) will drift over time if not corrected. What I found out on my first prototype was that I could use an accelerometer for level detection (the second prototype uses 2 accelerometers mounted vertically). This is used as the input to the proportional term. The gyros are the input to the derivative term as they measure the rate-of-change. I could give you the contants for these terms, but it would be different for each design. You will need to tune the PID algorithm if you make significant changes to the frame, position of batteries, etc. See my post to spaceclam a couple of days ago for a procedure on this. I also found that the integral term (summation of previous errors) is not a good thing for such a dynamic system. It will make it very stable in your hands or on a rotating surface, but it will not fly very well! This is not to be confused with the Integration that Danvel and ADI are doing.

Tell us more about the details of your craft... do you have any images of it?

Good luck to you and your team.

Regards,
tve

danvel

Always keep in mind that the reading of the gyro is angular speed; when it starts to decrease, it doesn't mean that the DF has reversed direction; it has only slowed down its motion, but angle continues to increase. Only when speed goes negative (or below 128 if we take mid-byte reference) the integrator starts decreasing.

Closing the loop in excel will only confuse you even more, because to do it properly you would need to obtain the dynamic equations of the system and take into account inertia, motor equations and so on...

Just implement the control and fly!.You will see that it's not as complicated as it might look.
Good luck.

danvel

Aussie Mechatronics Guy:

I've seen in your graph that brushed motors seem to give more thrust at low power levels than brushless. Is that correct? How can you explain that?

ADI

Danvel Wrote:

(1) >Always keep in mind that the reading of the gyro is angular speed; when it starts to decrease, it doesn't mean that the DF has reversed direction ..........

(2) >Closing the loop in excel will only confuse you even more, because to do it properly you would need to obtain the dynamic equations of the system and>

Thanks Danvel ........ re (1) above: yes I now understand. You are absolutely right.
Re (2) above: I realised I had no dynamic data in the Excel spreadsheet. What it did prove to me though, was the fact that you were right (again) and there was no need to change the sign of INT. This was the only thing I was testing for.

I promise I'll shutup now ..... 'and just start f#♀♣α$* flying" he he

Cheers ADI

ADI

Zarx Wrote:
>(I think the tokin/murata gyro has .8 mV/deg/sec). Therefore my gain constants are acting on the actual estimated deg/sec value that the gyro is getting..>

We've discussed these specs in the past and I for one had some trouble believing them.
Whatever you do, don't just rely on the spec sheet, especially for autonomous control, or your DF might fly off into the sunset. he he
Do some bench tests, if you need to know actual absolute data.
For example: 'Output voltage at zero angular rate' for Tokin CG-L43 is specified at 300-500mv, but the reality is well over 1 volt, with regulated 3v supply.

Cheers ADI

tve

Spaceclam,

hope you're having fun in the mountains.... I can't stop thinking about how cool pitch control will be. It will make the 4 rotor hovering thing into a 3D Nightmare! Do you know of any other similar parts that might work for my larger AXi motors? I'm not asking you to research it or anything, just asking if something comes to mind.

On an un-related note and topic...
I just noticed you changed your quote... pretty good. I have to know though... Is your icon (I forget the correct term) really an amanita mushroom? Is that what inspired your handle? To me it looks like a mushroom, but I can see how it also looks like a UFO with a clam type of protuberance.... I'm just curious for some reason. Did you airbrush it?

Spaceclam

the snow is insane up here. about 4 feet of fresh powder in some places, temps hitting the 20s, it's all good. however, i had never been down a double diamond run before, (i usually just stick to the single black diamonds ad double diamonds are like olympic level) but i managed to loose one of my two poles on from my lift chair on a double diamond run, and believe me, doinig a double diamond for the first time in 4 feet of powdere with only one pole was truly a nightmare[:@]

anyway, i dont mind doing any research for any of you guys, but i would look into a 60 size helicopter tail rotor assembly. here are a couple links to tail rotor assemnbly parts for a kyosho concept 60, but 60 sized rc helicopters are easy to come by, and i can also think of off the top of my head, the hirobo shuttle 60, concept 60, and the raptor 60. those may be a little big though, i would then look into tail rotor parts for a 30 size helicopter. things like the concept 30 would probably work just fine too. it does not sound like weight is a real issue for you at this point, so i am sure that either woudl work. just remember, it may not be a drop in fit for you. i bought the unit i did because it was supposed to be fit for a 3mm shaft so i should not have any problem there, but when in doubt, ask becuase it is really easy to drill out a hole to the right size, but if it is much more difficult to make an adapter. I would go with the 60 size one, and if the hole in the hub is too big for your shaft, i would reccomend that you get your hands on a turning lathe, and make a new shaft that is the right size at the end if you can. eithe way, just get one unit at first to scope out the situation. i am not going to order a set of four pitch units until i see for myself that it will fit, although it would be really easy to make a new shaft in my case anyway. however, i have no plans to 3d this, not yet anyway. if it gets boring i might try flips and stuff...
Also, i have absolutely no clue what kind of mushroom that is in my avatar, (icon). i just wanted to make a "funky mushroom" thing for my avatar so i google immaged "mushroom" and this was one ofthe results. i then photoshopped that onto the background you see. i am nowhere near good enough to airbrush that...
Who told you that was an amantia mushroom? i know there has been a lot of speculation on these forums.

here are the links to the helicopter parts, but you could also try something like www.hifirc.com www.fxaeromodels.com www.heliproz.com or www.helihobby.com and see if they have the tail rotors for 30 and 60 size aircraft. (i think they would)
http://www2.towerhobbies.com/cgi-bin...?&I=LXW457&P=0
http://www2.towerhobbies.com/cgi-bin...?&I=LXW458&P=0

tve

Spaceclam,

Thanks for the links! I might just order one of these to play around with! Do you think the concept 30 blades would generate quite a bit of thrust?

As far as speculation goes about your avatar, I haven't seen any. I am only reading this thread on rcuniverse though... I was just curious. Those type of mushrooms grow in the cascade mountains here in Washington. Maybe California too. They are considered poisonus, but some people cook them and eat them.... (from what I read, I wouldn't recommend it!!!) Just some trivia I ran across in the last couple of years. I think they are pretty cool looking. Did you google/image for amanita mushroom?

http://images.google.com/images?q=am...e=off&c2coff=1



Don't break any bones up there....

zarx

Well, the voltage noise is about 50mV peak to peak centred on 5 volts. There are also little impulse spikes that appear and I CANT GET RID OF THEM. Even the good tantalum caps didn't help! I ordered some schottky diodes - they will work better as the flyback diodes, so that might help the noise. I also tried turning the PWM freq up to 16 khz - the motors actually ran SLOWER on the same duty cycle. Anyone got an explanation for that?

tve, as far as pics go, I'll take a few and post em soon. I was reading your previous post on tuning the gain constants. Are you saying you never even used the gyro integral at all? You simply used the accelerometer output and the gyro outputs. Did you ever consider differentiating the gyro output to get angular acceleration? If you use washout on that term to prevent any reaction from voltage spikes, it will speed up your step change response time considerably. The problem we are having now is that for some reason the damn DF wants to remain tilted. I built a little cradle for it to sit in that allows freedom of motion on only a single axis, and the DF remains constantly tilted to one direction, ugh.

zarx

Oh yea, I should tell you that I basically made a 6 Degrees of freedom inertial measurement unit. It has XYZ accelerometers and gyros. The gyros are well filtered...the accelerometers aren't. We use software filtering which isn't as good unfortunately. I am using the analog devices 2 axis accelerometers (you can get samples of these from analog if you want to try them out) and they are very effective.

ADI

zarx Wrote:
>I also tried turning the PWM freq up to 16 khz - the motors actually ran SLOWER on the same duty cycle.>

I'm not an expert on FETs but consider this:
Everytime you switch a FET on/off there are power losses associated with the actual switching time period. (rise and fall times)
Switch it 16000 times in a single second time period and you have 16000 x 75ns switching period losses. Switch it at 1000 x 75ns and you only have 1000 switching period losses over the same 1 second of time.

tve

Hi zarx,

I do average samples from the gyros between motor updates to reduce noise, but thats it. No complex algorithms. This doesn't mean I'm not going to try different approaches. I am very interested in Kalman filtering, I just haven't taken a stab at it yet. I hope to have some new video of the thing flying this weekend. I hope the new setup is as stable as the last one.

I know this is going to sound wrong, but when I was using brushed motors I found that a pwm frequency of 120Hz or so worked much better. The FETS did not get warm and the motors ran fine. You read that right. 120Hz. Try it if you get a chance.. I think this due to eliminating switching losses as ADI was pointing out because the FETS would get considerably wamer even at 2khz.

Regards,
tve

Spaceclam

the concep't tail blades will not provide nearly as much thrust as your props. our beloved airplane propellers are design to be used in dynamic airflow situations, and of you were to take a let's say *x5 prop, and put it on a mechanical pitch setup, they would be rather inneficient at most pitches, and the efficiency will spike up, and then drop rapidly again once you pass it. helicopter tail blades, becuase of their blade design, have a wide efficiency range, but are not very efficient anyway at the speed they are run at. they have to have a wide efficiency range because they are constantly being varied and you dont want your helicopter to be really sensitive around neutral on the yaw and then as the blade pitch increases/decreases, it would not do much so you would run out of control real quick.
for our purposes, we are much better off either making or buyinig a pitch gauge, measuring the pitch at the tip of the blade, (in degrees) and setting that as the constant pitch for throttle, and so any little variation will make a world of difference, becuase it would still be close to it's efficiency range, even if it is 3-4 degrees off.

While all of that is true to my understanding, the concept's tail blades are simply way too small. you will probably have to get some helicopter blades (main rotor blades, not the tail rotor) and cut it to size. with all the horsepower you have on your craft, propeller innefiency should not be a concern of yours