Battling the Jello Monster (Quantitative Vibration Analysis)

[Snarls]

In my ever-present battle against the infamous jello monster I've decided to drop the guess-and-check approach for a more quantitative analysis of my enemy.

My quest against vibrations stems way back to my first multirotor, and has haunted me up to this point. Miniquads with an action cam have been generally acceptable when recording video, but my TBS Disco with a gimbal has always had issues. I've been very close to perfectly smooth video, but it has never been consistent. Now it is time to attack this problem head on.

To begin, I am looking at the primary source of vibrations, the propellers. I've pretty much always flown my Disco with ten inch Quanum carbon fiber props which I speced out to be the most efficient. I've also believe that carbon fiber props would produce less vibrations because they are stiffer. Was I right? Well lets find out.

From previously trying to smooth out my video I have ended up with four different propeller options:



(From Left to Right)
10 x 3.8 Quanum Carbon Fiber
10 x 4.5 Hobbyking Thin Electric
10 x 4.5 Hobbyking Slow Fly
9 x 4.5 Master Airscrew Multirotor

I am especially interested in the new Master Airscrew multirotor propellers which claim to be engineered to be more efficient and have better performance.

Initially I wanted to test these on the bench, but I could not get accurate vibration data using my phone, and I could not treat each prop equally since I did not know what throttle value would best simulate the rpms required for general flight. Luckily my Disco flies using a Pixhawk which records the sensor values for every flight which can be viewed later on the computer. Using this method I can look at the current draw and vibration data for each propeller after a test flight.

To perform as much of a controlled experiment as possible I planned out a waypoint mission for the Pixhawk to fly. The mission includes basic movements that should represent a normal flight; fast forward flight, quick stops, slow forward flight, turns with yaw, takeoff, and landing. The pixhawk would fly this autonomously, leaving my choices as an operator out of the equation and ideally putting each propeller through the exact same movements.



I waited for a calm day and performed these initial tests. Taking the flight log from each flight, I exported the data to Matlab and made comparison plots for each propeller with respect to current and X,Y,Z vibration. Here are the initial results:




(CF for carbon fiber, TH for thin electric, SF for slow fly, MA for Master Airscrew)

The plot of current indicates pretty clearly that the carbon fiber prop is the most efficient. And to my surprise the new Master Airscrew prop is the least efficient. I take this with a grain of salt though, as the Master Airscrew prop is the only 9 inch prop and logically would require more rpm and current to lift the multirotor.

The vibration plots clearly indicate one prop that is worse than the others, and that is the slow fly prop. The other props look generally quite similar, with the carbon fiber prop maybe again being the best. To check this I looked at the averages and standard deviation of the vibrations.

	Mean (m/s/s)							StdDev (m/s/s)
	X	Y	Z				X	Y	Z
Carbon Fiber	3.8596	4.3842	10.6373				1.0836	1.0803	3.9469
Thin Electric	4.2934	6.1323	14.7218				1.1189	2.0334	6.1874
Slow Fly	7.4226	9.9663	18.1682				2.1127	2.3583	5.4049
Master Airscrew	4.5176	5.9404	13.8779				1.4113	2.4533	6.4499

This data confirms that the carbon fiber prop is the best in terms of least vibration with the least X, Y, and Z vibration means and standard deviation. The Thin Electric and Master Airscrew are not terribly far off, and the slow fly is significantly worse.

So this proves underwhelmingly that the carbon fiber props I've been using all along are the best right? Well after doing these tests I realized there are some variables I had not accounted for. The carbon fiber props are the only ones balanced, the others are straight out of the package. Another issue is wind. If one test experienced wind and another did not then the flight log might reflect that as more vibration. I was also flying using PIDs calibrated for the carbon fiber props, and it is possible that PIDs not suitable for a propeller may change its vibration behavior. The next step is to eliminate these variables and perform the test again, to see with further proof, what prop is best. [to be continued]

 

[LitterBug]

Good stuff Snarls! I've been guilty of flying cheap and am just now starting to use carbon props on my larger quads. Found some really good deals since nobody flies big stuff any more. If you are at FFE, I have some more props you can test including some T-motor CF 9x3 antigravity. Check out this deal https://www.readymaderc.com/store/i...oduct_info&cPath=534_537_572&products_id=2220

Was hard to pass on that......

Cheers!
LitterBug

 

[Snarls]

Thanks. Unfortunately I won't be at FFE, wish I could though. Thanks for the tip on those props, yeah that's quite a deal. I just picked up a set for myself to test along with some other super cheap props. I guess I'll do another round of tests with a whole new set of contenders

 

[Snarls]

A couple days passed while I further eliminated as many variables as possible. First I spent an hour or two balancing every prop as best as I could. The carbon props needed a minor adjustment and the Master Airscrews were pretty much balanced out of the package except for one prop. The thin electric props needed sanding and balancing on each prop and the slow flys needed sanding and some hot glue on the hubs to balance. I did a completely new autotune while flying the Master Airscrew props as these are the props I was hoping would prove to be the best. If the new tests showed the Master Airscrews as best then I would know that PIDs have a significant effect on the vibrations and consider finding optimal PIDs for each type of prop. I then waited for as close to a windless day as possible for test two.

I changed the autonomous mission of the first test to further represent real flight. I increased the flight distance, the takeoff and operating altitude, and added three loiter (yaw) turns at different radii. These missions would be longer and give a larger quantity of data for each flight condition (fast, slow, turns, etc) to analyze.


Each flight took about three and a half minutes. The first flight with a full battery was with the Master Airscrew props and I noticed that there may have been some oscillations for slightly high gains. After doing flights for the other three props I lowered the gains slightly and flew the Master Airscrews again. This would also give me a look at how battery voltage effects the vibrations. If it does, then the second Master Airscrew test would be better to compare to the carbon fiber props, which were tested just prior, and these were the two props I wanted to focus on. From the telemetry on my computer during the flights I already noticed that battery voltage may have an effect on current draw, and the data would tell a better story. After finishing the testing I once again saved the flight logs and processed the data in Matlab.




(CF for carbon fiber, TH for thin electric, SF for slow fly, MA/MA2 for Master Airscrew at beginning and end of battery.)

These results were much more interesting. With all props balanced it becomes difficult to discern which prop draws the least current. The carbon fiber props have some of the lowest current spikes, but also high ones, and is not very consistent in comparison to the other props. Also worth noting is that the Master Airscrew test at the beginning of the battery is pretty low value and consistent, while the second Master Airscrew test near the end of the battery has larger current draw and is less consistent like the carbon fiber props. Is this change due to changing PID behavior at lower voltages or the low voltages themselves? I'm not sure. The best conclusion I can draw from the current data is that each prop is about equal in terms of current draw when properly balanced.

Now looking at the vibration data, surprisingly the slow fly props are not clearly the worst anymore; they are actually pretty comparable to the second Master Airscrew test and the thin electric props in Y and Z. The first and second Master Airscrew tests are significantly different, once again suggesting that battery voltage plays a bigger role on vibrations than I thought. The first Master Airscrew and carbon fiber look to be the better contestants once again. Looking at the means and standard deviations:

		Mean (m/s/s)					StDev (m/s/s)
	X	Y	Z			X	Y	Z
Carbon Fiber	5.8451	6.4938	16.0964			1.9631	2.1617	7.1951
Thin Electric	4.6703	7.0269	15.6583			1.9568	2.9099	8.3173
Slow Fly	7.2551	8.854	22.4083			3.2559	3.7981	10.5264
Master Airscrew 1	5.0610	6.7362	15.7337			2.0520	2.4206	8.0858
Master Airscrew 2	5.9666	8.6922	20.5374			2.2950	3.7812	10.2448

Looking at the numbers I think the carbon fiber props are once again the winner. The standard deviations are the lowest out of all the props. The means are not as low as the first Master Airscrew or parts of the thin electric, but it is important to remember the effects of the battery voltage on vibrations. Assuming that the vibrations get worse as battery voltage decreases, the carbon fiber props which were tested closer to the end of the battery are only slightly higher in vibration than the Master Airscrew and thin electric props tested near the beginning of the battery. The carbon fiber props also have lower means than the second Master Airscrew tests. Also remember that these flights were performed with PIDs that if anything would favor the Master Airscrews. So if battery voltage and tune are taken in to account the carbon fiber props are arguably the best overall.

To further defend the carbon fiber props I was recording on-board video for the flights involving the Master Airscrew props and carbon props. The video with carbon fiber props has some shakes and jello, but the Master Airscrew videos are clearly worse.

In the end I thought I would find one clear winner, but alas that is not the case. The slow fly props are the worse, but not terrible. It is interesting to note how some cheap plastic props can actually be pretty comparable to more expensive carbon (albeit cheap carbon) props or engineered Master Airscrew props. Personally I will be sticking to the carbon fiber props I have been using. This is not the end of the vibration testing though. Now it's time to step away from the propeller and down the line; to the frame itself. [To Be Continued]

 

[Snarls]

Now looking past the propellers it is time to check the path of vibration from the props to the flight controller, and further to the gimbal. I'm looking at two areas, where the motors mount to the arms, and where the arms mount to the frame.

I designed some motor mount pads and printed them out in TPU on my 3D printer. I also designed some pads to go between the top of the arms and the top frame plate. They are wedged ever so slightly to hopefully angle the arms up a little, giving slightly inward facing motors and more stability. These were also printed in TPU.







There were four tests I performed; a control test with no dampening, one with just dampening at the motors, one with just dampening where the arms meet the frame, and one with both. The flight path would be the same as the last prop test. To eliminate any influence from voltage as the battery drains, I ran each test with a full battery, recharging after each run. I tried to run all four tests in one afternoon, but the wind picked up before I could do the test with just the dampening at the arms. So that test was done two days later. Here are the results I got:





Looking at the current plot, there is no significant difference between any of the data sets. Some traces have more noise than others, but I am going to attribute that to outside variables like wind. Equal current draw is expected here since the props are not changing (Quanum carbon fiber props). This is supported by the current averages:

Mount	Control	Motor	Arm	Both
Mean Current (A)	11.5084	11.5929	11.4954	11.5996

Moving on to the vibration data there is also no significant difference between any of the traces. You could say the 'arm' (only) data set has less vibration, but because that test had to be run on a different day I am going to say the difference is not from the arm mounts.

		Mean (m/s/s)				Stdev (m/s/s)
	X	Y	Z		X	Y	Z
Control	4.7046	6.5859	16.9027		1.5054	1.5906	6.6395
Motor	5.1351	6.0037	16.0968		1.5540	1.6819	6.3539
Arm	4.1046	5.0858	12.5501		1.5290	1.5670	6.2195
Both	4.7070	5.7937	14.5389		1.4287	1.3435	5.4107

So it looks like adding TPU motor mounts and arm mounts did not significantly alter the vibrations reaching the flight controller. If anything using just an arm mount is best, but once again it is not significant enough of a difference to rule out environmental factors as the cause. Even if the difference was purely from the arm mount, it is not a significant enough improvement to warrant dealing with them. Maybe other materials might fare better, but so far vibrations are better controlled at the props than below the motor and at the arms.

 

[PsyBorg]

Interesting data collection and analysis. I have always thought soft mounting motors and arms would not produce results. No mater what the conditions the FC's will ALWAYS try and do their best at what they have been programmed to do. you can not change somethings best. You can make it easier to do its job but it will only do the best it can under all circumstances.

That said I will throw something into your mix to maybe make your experiments more controlled. I know the KISS fc's have a TPA adjustment that will work according to battery voltage and current. Does the FC you are using have any similar method to govern pids according to voltage and current thus lessening the changes as time passes during the flights?

What I think you will end up doing to solve the gimbal jello dance is to stiffen the mounts to the gimbal to change the resonant frequency inherent in the build. The FC seems stable across the board and is doing its thing from what I see on the graphs. You will never get rid of vibrations in these things. I think changing the resonant frequency to get best performance is all we can do at this point if we want perfection.. We will probably have to wait for the Aliens to come and fix our Aliens....

There probably is a way to add mechanical shakers that can vibrate in harmonics to the vibrations the FC reads and cancel them out but more power use and weight is not what we need for things that fly.

 

[Snarls]

Thanks PsyBorg. Yea the Pixhawk does allow you to enable PID voltage scaling. I enabled it earlier today, but have not had the opportunity to test how good it is. I agree that stiffening the gimbal mounts will probably be the better design. In my experience the softer the gimbal mount (more loose) the less vibrations, but more physical bobbing around of the camera which leads to jello. The stiffer the mount the more vibrations, but the camera does not bob around. I need to find the sweet spot. That's some testing I have on the to do list.

 

[PsyBorg]

Get one of them sponges that has sanding material on it. I regularly cut up used ones to make my camera mounts and I only get jello when I don't strap them down tight enough in the foam or I have seriously damaged props. Use what ever mounting is provided for the gimbal and supplement that with custom cut and fit foam from the sanding blocks. Just make sure if it is those rubber bobbin type mounts you run a few zip ties thru the centers in case that sponge is to tight. Don't wanna pop the complete assembly off mid flight and lose the camera and a gimbal.

 

[LitterBug]

Keep up the great work Snarls! I have been amazed at how little jello (if any) there is on my AWK210 with the camera hard mounted to the top. It really does sound like you have "lucked" in to the resonant frequency that your camera scans with your mounts. Have you tried using different frame rates or exposure settings? Sometimes that can get you out of the Jello zone too.
Here's my setup with a Runcam HD2 (on the left):


And here's a sample:

I don't have any good footage from either of my gimbled quads... Hope to get some at FliteFest.

Cheers!
LitterBug

 

[Snarls]

Thanks guys. I've had very few problems with jello on my miniquads. Likely because the props spin at a much higher frequency than the camera, and the props are lighter and smaller so any eccentricity will produce less of a vibration. With larger quads the RPMs produce much more visible vibrations, which may hit a resonant frequency as you suggest. The props are also larger so any imbalance produces a larger amplitude vibration.

 

[PsyBorg]

That and you tend to fly at a stable rpm in hover for filming or slow movements so a resonant shake could easily build of the varying harmonics of the air frame.

 

[Snarls]

After Litterbug tipped me off to the prop sale at RMRC I ended up buying a handful more props to try. Knowing how well the carbon props performed in the last tests I wanted to make this into a carbon prop showdown.

Here is what I have for this test:



-Same Quanum CF props as before
-HQ 10x3.8 MR reinforced plastic
-Gemfan 10x5 CF T style
-T-Motor Antigravity 9x3 CF

I also got some wood props but made a goof and bought only reverse rotation and half were 10x5 and the other 10x4. Wood props are still on the radar though just to experience.

The HQ props are the only non CF props, but they are almost exactly the same form as the Quanum props. I'm not sure who copied who, but they are definitely the same design. Testing these against the Quanums will be a plastic vs carbon comparison to test my hypothesis that carbon is better.

Out of the box the Gemfan props were not impressive. They are the only props that required balancing and the mounting holes were not perfectly in the center. I had to ream the mount hole from 4mm to 5mm and tried to center the hole while I did it, but they are still off.

The T-Motor Antigravity props are the real star of the show. They are normally way out of my price range but the sale at RMRC gave me the opportunity to test them. Out of the box they are balanced and almost paper thin. Because of this even though they are CF they flex as much as a reinforced plastic. They are also super light and look great.

On to the testing:

For testing I once again ran the same autonomous flight path on an almost windless day. To avoid voltage issues I also again ran all the tests with a fully charged battery. I used the PIDs for the Master Airscrew props from the last round so no prop should have a tuning advantage, with the exception of the T-Motors. The T-Motors oscillated and felt so loose on the Master Airscrew props that the mission was in danger of crashing. So I did an autotune on the T-Motors, which I'll keep in mind when analyzing the results. Also note that I took off the camera gimbal for this round of testing so that is why the hover current is less than the previous prop testing.

The T-Motor props are so light that it is very easy for the motors to accelerate them. That got me thinking about the effects of damped light on vibrations. Damped light actively slows the motor and prop. In theory quickly slowing the props which have rotational momentum could result in additional vibrations. So for one additional test I disabled damped light and tested the T-Motor props again.

Here are the results (click to see closer):





To my expectations, the T-Motor props required more amps to hover, being a 9x3 prop vs 10x3 and above. They, along with the Quanum props are very consistent in the hover current draw. The plastic HQ props and Gemfan CF props on the other hand are pretty varied in their current draw. There appears to be no difference between the T-Motor props with and without damped light (DL). Overall the HQ prop has the lowest instantaneous current draw, but the Quanum props are consistently the lowest. To put numbers to these results:

Prop	Quanum	HQ	Gemfan	T-Motor	T-Motor no DL
Current Avg.	9.2459	9.6030	10.1685	10.8546	10.7706

Just as a fun note, with the Quanums and my 5200mAh battery I could possibly hover for 33 minutes (in a perfect world).

Now looking at vibrations. The Gemfans stand out right away as the worst of the bunch. The vibrations are consistently higher than any other prop, including some of the plastic ones from last round. There is also a strange spike in vibration at around the 600 samples point. These vibrations could be from the high pitch of the prop (10x5), or more likely because the mounting holes are not centered. I don't think these props should perform this poorly so I may come back with a way to mount them centered and test them again.

The HQ props are comparable to the T-Motor and Quanum props at points, but they frequently spike to higher values. Because they share the same profile as the Quanums this supports my hypothesis that carbon props are better than plastic props. The HQ props are still a very respectable plastic prop I would recommend.

The two T-Motor tests and the Quanum test are quite comparable. Notable however, the T-Motor traces have occasional spikes throughout the test, whereas the Quanum trace remains generally flat overall. Interestingly the T-Motor props with no damped light have significantly less vibrations than the Quanums in the Y axis. This is odd because there is no significant decrease in vibration in the X and Z axis. Worth noting is that I moved some wires touch the flight controller before that test which may explain this change, although I would think the wires effect all axes.

Looking at the numbers:

		Mean (m/s/s)				Stdev (m/s/s)
	X	Y	Z		X	Y	Z
HQ 10x3.8	5.8716	5.9075	12.5438		1.6816	1.4776	4.2951
Quanum CF 10x3.8	4.4948	4.9809	8.4606		0.9245	1.0035	2.5645
Gemfan CF 10x5	9.3948	9.6050	19.1838		2.3122	2.4718	5.5360
T-Motor Antigravity CF 9x3 no DL	4.7059	4.0228	10.4306		1.1979	1.5024	5.8038
T-Motor Antigravity CF 9x3	5.5463	5.3732	9.2713		1.3331	1.4330	4.8213

Once again by the numbers the Quanum carbon fiber props prove to be the best in both mean and standard deviation. The T-Motor props are not far behind, followed by the HQ plastic props. And unfortunately the Gemfan carbon props are significantly worse.

Conclusions:

So are the expensive T-Motor props worse than some cheap Quanum props? Well that can't be concluded here. Comparing 9x3 props to 10" props, the 9x3 props will have to spin faster and lift the same weight with less area. That could result in the props flexing more and thus contributing more vibrations. In comparison to other 9" props I'm sure the T-Motor props would be one of the best. They are certainly well performing props overall.

After this second round of prop testing the Quanum props are still the best choice. Who knew these props I originally chose for my Disco would be the best. It's kind of frustrating I can't solve my vibration issues with changing my props, but at least I know I made the best choice.

 

[LitterBug]

Where do you get these Quanum props? Think I need a set to test too.

 

[PsyBorg]

Something you said struck a cord in my foggy mind. You said you changed damped light and just the Y axis changed. I have noticed thru all of your testing that Z axis (yaw) is consistently higher. I am wondering if there is nothing wrong with your set up other then all of your motors are not pointing 90 degrees straight up. If one or two were off that axis it would induce a touch of yaw that would constantly be fought but the fc. Since this is an AP ship p gains may not be as high but I and d are more likely higher then for quad freestyle.

I would take a look and check with a square that your motors are in fact all still pointing straight up in relation to the FC.

 

[Snarls]

LitterBug here are the Quanum props. I guess they are only available from the global warehouse.

Psyborg the data is accelerometer data so the Z axis is not yaw, it is up and down acceleration. Actually it is the standard deviation of the noise on the acceleration. From what I've seen from other pixhawk logs online the Z axis is consistently higher than X and Y. Probably because the lift of the multirotor and any up and down turbulence from the props will be seen as Z acceleration. Interestingly arducopter claims values below 60 m/s/s are acceptable for stable flight.

Interesting you bring up the yaw. I checked my motors and none of them are noticeable angled to the side. I know however that my front left and back right motors are consistently require higher pwm values in flight. It is a strange issue that other people have seen even with perfectly aligned motors, and the cause is still pretty unknown.

 

[PsyBorg]

DARN.. I though I was on to something smart... Carry on.. Ill be over here quietly hiding my head in shame.. I'm a codfish.

 

[Snarls]

Ha no need to hide in shame. Any insight here is good insight. At least you bring up my strange yaw issue which may or may not be related to the vibrations.