A2 Problems Catalogue

[Benjamin Kenobi]

Agreed. I'm all for vibration dampening of the sensitive components (IMU). I still use PU gel for everything as I have done for years. I want the IMU detecting the large movements of the aircraft rather than the vibes going through the frame.

 

[jdennings]

they would find that the response bandwidth of the system is only around 10-20Hz, with 50Hz being the upper limit for very small machines.

But to be sure of this, you'd need to log/sample observations at 100hz minimum, with 500hz/1khz a better indicator. This also really depends on the imu, the built-in filtering there and settings used.

 

[jdennings]

Agreed. I'm all for vibration dampening of the sensitive components (IMU). I still use PU gel for everything as I have done for years. I want the IMU detecting the large movements of the aircraft rather than the vibes going through the frame.

It was my impression that DJI does not recommend dampening?

 

[R_Lefebvre]

But to be sure of this, you'd need to log/sample observations at 100hz minimum, with 500hz/1khz a better indicator. This also really depends on the imu, the built-in filtering there and settings used.

It doesn't have to do with the IMU, filtering, etc. It's just the physical truth that aircraft cannot possibly respond at a rate faster than that.

 

[Tahoe Ed]

You are right. They recommend that you hard mount the IMU with the thin 3M double stick tape. I know that many do not agree with that statement. You can do what you want but do not come to DJI if you have issues because that will be a question that is asked. How is your IMU mounted. Just my opinion.

 

[gtranquilla]

True, the NAZA and the A2 are to be hard mounted.
But during the course of S900 development the anti-vibration module was implemented to deal with some WKM IMU related issues.
So technically the IMU is being dampened externally via the anti-vibration plate assembly.

You are right. They recommend that you hard mount the IMU with the thin 3M double stick tape. I know that many do not agree with that statement. You can do what you want but do not come to DJI if you have issues because that will be a question that is asked. How is your IMU mounted. Just my opinion.

 

[jdennings]

It doesn't have to do with the IMU, filtering, etc. It's just the physical truth that aircraft cannot possibly respond at a rate faster than that.

Oh, I see ... So what you are saying is that a copter with an imu sending a square wave with, say < 20ms pulses (>50Hz) would behave the same way as if no pulses were sent? No matter what the amplitudes? Don't ESCs and motors respond faster than this, and ultimately props, at least in a subdued way?

Having a hard time wrapping my head around this: A 400kv motor running on 6S can spin props up to 10,000 rpm ... That's more than 3 revolutions in 20ms. Even if in practice it's less than that, still plenty of room and time to adjust, no?

 

[Ronan]

True, the NAZA and the A2 are to be hard mounted.
But during the course of S900 development the anti-vibration module was implemented to deal with some WKM IMU related issues.
So technically the IMU is being dampened externally via the anti-vibration plate assembly.

Another reason why not to try to 'double' anti-vibrate the A2, which could cause issues.

 

[gtranquilla]

Please correct me if I am wrong but only the FC output controls are confined to normal MR rotational speeds in the 3 axes.
Whereas with respect to the sensor inputs, on the basis of a typical 9DOF IMU using conventional Kalman filtering, the accelerometers in particular have to deal with much higher frequencies because acceleration/deceleration can occur in meters per sec squared which is almost exponential.

Regardless of the frequencies, there is also the risk with high magnitude accelerometer signals that the accelerometer ranges can be exceeded such that input signal could be clipped at both top and bottom end. Net result would be that averaging under such conditions would always come out as a zero acceleration when in fact the MR is actually accelerating or decelerating in a specific direction.

 

[gtranquilla]

These IMUs contain at least 3 accelerometers, one per axis. In the event that acceleration/deceleration magnitude/force in any axis exceeds the sensor +- range limits, that data becomes invalid and has to be ignored by Kalman filtering. If not ignored, the averaging results in a false zero acceleration condition. In both cases the Multirotor would be flying without acceleration input data, relying only on the magnetometers and the gyros......

IMHO - If one is lucky it might remain level for a short period then get worse or simply fly-away.

While it is true that the airframe has limited mechanical rotational speeds well within the FC output control ranges, the input sensors have a wider frequency sensing range, especially the accelerometers since they must accurately sense acceleration which is in meters per sec squared.... almost an exponential curve.

In radio theory a band pass filter would be most appropriate to eliminate unwanted highest and lowest frequencies. But for Multirotor FCs, I believe the filtering is almost entirely done by means of Kalman filtering which is actually a computer algorithm which does the filtering on the basis of statistical analysis, discarding bogus input data then assigning weighted averages etc. to the remaining data.

 

[R_Lefebvre]

Oh, I see ... So what you are saying is that a copter with an imu sending a square wave with, say < 20ms pulses (>50Hz) would behave the same way as if no pulses were sent? No matter what the amplitudes? Don't ESCs and motors respond faster than this, and ultimately props, at least in a subdued way?

Having a hard time wrapping my head around this: A 400kv motor running on 6S can spin props up to 10,000 rpm ... That's more than 3 revolutions in 20ms. Even if in practice it's less than that, still plenty of room and time to adjust, no?

That is more or less it. Think about it this way. Do you know why AC current is 60hz? It's because it's so fast, that your eyes cannot see the light filament going dim and bright as the AC voltage fluctuates. So that's 60 hz. So fast, that your eyes can't even see it. Now, can you imagine an entire copter frame vibrating at 50Hz? It would be a blur. In fact 50hz is even too fast to see, they use that AC frequency in Europe. Do you think that the motors and props are capable of generating enough thrust fast enough to make your copter appear a blur? Not a shake, we've all seen that, but an actual blur so fast that you can't even see it's position at any point. That would be 50-60Hz. Ever heard a 60hz transformer hum? That is the sound the *airframe* would make (not the propellers!).

30-40 Hz is a very violent shake where you may be able to see discrete positions. If you can imagine the worst helicopter shake you've ever seen, from a bent main shaft or similar, where the helicopter is literally tearing itself apart, that is usually at 30hz. Do you think your propellers/motors have that much power?

Yes, the motors and props can respond more quickly than 60hz. But that does not mean they have enough power to make the airframe respond at that rate. The props just don't have that much force potential with the air.

The default filter rate for IMU data going into the attitude controllers in Arducopter is 20Hz. Most people don't even notice, which is an example of the fact that the controllers just don't need to be updated at these crazy high rates. And filtering down, helps avoid many problems.

 

[R_Lefebvre]

gtranquilla, you're talking about a few different things at once.

First, if the vibration is so intense that it is over the measurement limits of the sensor, you're done. There's no need to talk about a filter, etc.... it's going to crash, period. That's a whole other problem.

So, you mention the Kalman Filter. Yes, the Kalman Filter should be exposed to full-rate accelerometer data. It is an integrating filter, which means it naturally averages out any vibrations. (assuming it's been programmed correctly!). But the angular rate controllers need filters to get rid of noise. We actually had a fascinating situation just last week. A user was experiencing horrible vibration, that he just could not get rid of. It was at a specific frequency. He didn't want to go to the 10Hz LPF option, because he felt that made it less stable. I use 10Hz all the time on helis, though they are naturally stable. Anyway, one of the other devs thought that what he was seeing is "structural vibration", meaning that the control loop was actually exciting his airframe's natural frequency. The prop was getting control inputs at whatever Hz the frame resonates at (I think it was around 30hz). So the guy hacked in a notch-filter at the resonant frequency (beauty of open source...) and the problem was solved.

Cool stuff, and it reinforces what I'm saying about the response rate of these machines.

 

[gtranquilla]

R_Lefebvre,
Thanks for the good info. Notch filtering is a great idea, more specific versus low pass filtering, but I believe it would still have to be matched to the dynamics of each specific airframe which remains a major variable in the hobby world.
We use configurable Active filtering in the industrial electrical power world, but too complex, heavy and expensive so far for MR use. It works much like noise cancelling headphones and can be tuned to overlook a wide range of needed frequencies.
BTW - were you in any way involved with the development of INGRobotic's Responder?

gtranquilla, you're talking about a few different things at once.

First, if the vibration is so intense that it is over the measurement limits of the sensor, you're done. There's no need to talk about a filter, etc.... it's going to crash, period. That's a whole other problem.

So, you mention the Kalman Filter. Yes, the Kalman Filter should be exposed to full-rate accelerometer data. It is an integrating filter, which means it naturally averages out any vibrations. (assuming it's been programmed correctly!). But the angular rate controllers need filters to get rid of noise. We actually had a fascinating situation just last week. A user was experiencing horrible vibration, that he just could not get rid of. It was at a specific frequency. He didn't want to go to the 10Hz LPF option, because he felt that made it less stable. I use 10Hz all the time on helis, though they are naturally stable. Anyway, one of the other devs thought that what he was seeing is "structural vibration", meaning that the control loop was actually exciting his airframe's natural frequency. The prop was getting control inputs at whatever Hz the frame resonates at (I think it was around 30hz). So the guy hacked in a notch-filter at the resonant frequency (beauty of open source...) and the problem was solved.

Cool stuff, and it reinforces what I'm saying about the response rate of these machines.

 

[R_Lefebvre]

Nope, not at all. I actually don't have much about them, even though they are close to me. I actually plan on trying to contact them soon. It's looking like I'll be getting out of the Arducopter world and looking for other opportunities.

So, the thing is, you say that notch filtering would be better than low-pass. The implication being, you want to let the very high frequency stuff through. I have to ask why you would want to that? What is the point of having the motors respond at a frequency that is higher than the inertial response of the frame? All you're doing, is making the motors respond to random noise in the signal line. It's not doing any actual useful work. It just wastes energy really.

 

[gtranquilla]

Good points.... I will continue briefly by PM.

Nope, not at all. I actually don't have much about them, even though they are close to me. I actually plan on trying to contact them soon. It's looking like I'll be getting out of the Arducopter world and looking for other opportunities.

So, the thing is, you say that notch filtering would be better than low-pass. The implication being, you want to let the very high frequency stuff through. I have to ask why you would want to that? What is the point of having the motors respond at a frequency that is higher than the inertial response of the frame? All you're doing, is making the motors respond to random noise in the signal line. It's not doing any actual useful work. It just wastes energy really.

 

[gtranquilla]

See PM.

Nope, not at all. I actually don't have much about them, even though they are close to me. I actually plan on trying to contact them soon. It's looking like I'll be getting out of the Arducopter world and looking for other opportunities.

So, the thing is, you say that notch filtering would be better than low-pass. The implication being, you want to let the very high frequency stuff through. I have to ask why you would want to that? What is the point of having the motors respond at a frequency that is higher than the inertial response of the frame? All you're doing, is making the motors respond to random noise in the signal line. It's not doing any actual useful work. It just wastes energy really.

 

[jdennings]

Yes, the motors and props can respond more quickly than 60hz. But that does not mean they have enough power to make the airframe respond at that rate. The props just don't have that much force potential with the air.

The default filter rate for IMU data going into the attitude controllers in Arducopter is 20Hz. Most people don't even notice, which is an example of the fact that the controllers just don't need to be updated at these crazy high rates. And filtering down, helps avoid many problems.

I hear you, and have no problem believing that 20Hz is plenty at this point in the game. But while I can't prove it, something tells me intuitively that better can be achieved and that there is something to gain going higher. Sure, it may lead to something barely visible, or actually plain invisible, yet it would still be better stability that would make a difference in forward flight, hover transitions, wind handling, video stability, and the like. Or maybe it's just that other sensors are just not good enough right now for the whole copter to take advantage of higher rates. Look at the Vicon guys at U. Penn and their balancing a pole on top of a quad. I'd have to look back at the papers, but I remember 1000Hz refresh rates mentioned, although maybe it was a different control loop. In any case, it'd be interesting to know what the lowest rate would be to effectively balance these poles, and how that affects oscillation and movement of the top of the pole, which is greatly amplified. Something tells 20Hz or even 50Hz would not be good enough, or at least would limit the length of a pole successfully balanced, compared to a higher rate ... Could be wrong, to be continued ...

 

[R_Lefebvre]

What if I told you that a 600 heli running 2400rpm head speed, which is very aggressive, cannot physically respond to any control signal at faster than 80Hz? As in, it's impossible.

A helicopter can only exert a rolling torque for example, when the blades are oriented to the sides. At 2400 rpm, that is 40 Hz, and on a 2 blade head, you have the blades aligned side-to-side at 80Hz frequency.

Do you think it is possible to balance a ball on the rotor head of a helicopter using one of those Vicon rooms? I think so. Even though the maximum possible control response is only 80 hz.

Another way to think of it. A blazing fast servo, with a 0.04 second transit time, can only operate at 12.5hz at full sweep.

 

[jdennings]

Good points, and I believe you ...
For now!!!

I've always wanted to play with that MPU6000 frequency filter parameter in AC, along with a few related others ... Gotta reorganize that priority list ...

 

[R_Lefebvre]

Actually, there are already changes afoot there. The filter will basically be turned off. Unfiltered data will be sent to the processor, which will be integrated into the EKF. But the data will be heavily filtered before being used in the rate controllers... we should probably get a room...