Reducing the vibration effect on accelerometers

[yuri_base]

Yo!

I'm using accelerometers to measure glide ratio of wingsuits. (Here is the device I built (http://www.dropzone.com/cgi-bin/forum/gforum.cgi?post=3069522#3069522) for this purpose.)

Here's the problem. I'm using Eagle Tree's G-Force expander - the 1.7G version (compared to their stock 38G version). The vane vibrates due to burble behind it and the vibrations destroy the accelerometer data with excessive noise.

Some of you probably faced similar problems although in UAVs the main source of vibration is probably the motor, not the turbulence.

Is there a way to suspend an accelerometer in such a way that it dampens the vibrations?

Yuri

[kd7ost]

Just think of fluid dynamics. That is a nice build but looks like it belongs in a vacuum. You can’t address the quality alone without addressing the dynamic forces placed upon the unit. Can’t you put it on the plane somewhere that it is in laminar airflow? Or in the wing LE? Or other options. Pretend you are the sensor. Where would you work best?

Dan

[yuri_base]

It is in laminar flow, not in turbulent flow (it's placed on my belly, see the attachment below). The "airplane" is me. :) (If you have never heard about wingsuit flying before, here are some beautiful pictures and videos (http://www.wingsuitphotos.com) to get the idea.)

The turbulence created by the vane shakes it and that creates excessive noise in accelerometer data.

[dmgoedde]

The "airplane" is me. :)
I can't help but make an obvious wisecrack: This forum is for "unmanned" aerial vehicles! You ought to post in the "manned aerial vehicle" group :)

OK, now that I ripped on you, my hopefully helpful advice is a thought ramble like this:

1) you seem to want a solution that doesn't involve nuts and bolts component/code level stuff, i.e. you are using a commercial unit that wouldn't be easy to hack anyway
2) In this case, I see why kd7ost answered the way he did. My advice along similar lines would be something like "wrap the unit in foam", but as you mention you are not fighting engine vibration, but rather buffeting by the wind.
3) If I was doing this (with one of my homemade units) I would use a Kalman IMU, adjust the Q and R values, take more averages of the accelerometer channels per iteration, etc... These methods I am sure would work, but obviously you can't do these things.
4) I guess I'm saying your problem has constraints that in my opinion prevent a solution. You should scrap that unit and buy an AttoPilot. I could customize something for you. I am only half serious, but am eager to see what other people have to add. I bet someone out there has a workable idea. If RCGroups is good for anything, it will generate lots of discussion.

[dmgoedde]

Can you post-flight analyze te raw data? Perhaps a Runge-kutta or moving average method of smoothing the noisy accel data.

The Eng side of me is brimming with all sort of solutions to you goal of determining glide ration that don't use acceleromters. GPS, barometer, pitot, Kalman IMU, etc.... But your question is obviously only how to take what you got and make it work. I'm eager to hear what a guy like JackCrossfire has to say.

dmgoedde

[poynting]

It seems like making the probe into a more aerodynamic body might do the trick. Remove all of the drag you're creating with the open back by making a fairing out of foam or something else lightweight and workable. Or better yet make a faring to go around the entire thing that's more like a teardrop.

[yuri_base]

If I was doing this (with one of my homemade units) I would use a Kalman IMU, adjust the Q and R values, take more averages of the accelerometer channels per iteration, etc... These methods I am sure would work, but obviously you can't do these things.

I will use Kalman filter, however, I want to reduce the unnecessary noise in data as much as possible so that Q and R are as small as possible. My goal is to determine lift and drag coefficients from data, so accuracy is important.

Attached is a sample of altitude, speed, and accelerometer data.

[yuri_base]

Can you post-flight analyze te raw data? Perhaps a Runge-kutta or moving average method of smoothing the noisy accel data.

Yes, the data will be only analyzed post-flight. This is not a real time system, so there is no need for embedded Kalman here. Kalman will be used on a computer.

The Eng side of me is brimming with all sort of solutions to you goal of determining glide ration that don't use acceleromters. GPS, barometer, pitot, Kalman IMU, etc....

I do use Pitot-static for altitude and speed. GPS is not useful here, since the glide ratio is skewed by the upper winds that are typically 20-30mph but sometimes reach 60-80mph. IMU is also not necessary since I do not measure the orientation of "aircraft" in space.

The principle behind this device is very simple: vane points into relative wind, so it points at an angle equal to glide ratio (relative to air). Accelerometer installed on the vane measures this angle. There is a little more physics involved here in situations when "MAV" ;) moves with acceleration, but it turns out that the ratio of the components of apparent acceleration is equal to lift to drag ratio L/D at all times.

[yuri_base]

It seems like making the probe into a more aerodynamic body might do the trick. Remove all of the drag you're creating with the open back by making a fairing out of foam or something else lightweight and workable. Or better yet make a faring to go around the entire thing that's more like a teardrop.

My first prototype before this one was of traditional vane design - a tube and a fin at the end. Unfortunately, this design is too messy for skydiving. I don't want to have sharp objects during parachute deployment and landing. Triangular vane is more compact and clean.

But your idea is sound - reduce turbulence in the first place. Perhaps, make the vane of metal mesh, so instead of one huge burble behind the vane the turbulence will be broken into thousands of pieces that will smooth out the jitter?

[zitron]

Why are you using a accelerometer to measure L/D? If I did it I would just use a GPS logger since it gives me location and altitude. Then after the flight I would just plot distance traveled against altitude, the slope of which is L/D.

-Z-

[kd7ost]

I’m a fan of preventing the issue at the first place it shows up. The turbulence is being caused by the collapse of air behind the triangle. As was mentioned above, a teardrop shape has been around for a long time to reduce drag caused by the collapsing high pressure air behind a mass moving through a fluid. (Air in this case is to be considered a low viscosity fluid) Consider wheel pants to reduce drag over wheels in vintage aircraft. It seems simple but sometimes the best solutions are.

If that’s not enough by itself, you could further reduce the effect by having a rough or pitted surface such as you might find on a golf ball on the outer surface of your housing. This would serve to expand the boundary layer around the tear drop shaped unit. The boundary layer is increased in amplitude by the shape first of all, then the speed and the mass of the fluid. At your speeds you’re not going to get much compression being that atmosphere has such a low viscosity. But still, the pits probably wouldn’t be needed if the housing is tear dropped shaped in a nice 3D airfoil.

My .02

Dan

[Jack Crossfire]

Accelerometer results R skewed by vibration as we found out. They measure acceleration by measuring vibration. The biggest problem is preventing vibration from being transferred through wires. It may require packing the accelerometer, power source, & wireless transmitter completely isolated from the airframe.

[zitron]

Gah, I really need to read the thread before posting! Ok, so GPS won't work.

You already have time, altitude and speed, that's all you need to compute glide ratio!

From the graph you posted, I can already calculate your L/D.
Say you averaged 90mph for 120 seconds, that's 4828m traveled
In the same 120s you fell 10000ft, that's 3028m fell
So assuming your speed is recorded along direction of travel, the amount of ground distance traveled = sqrt(4828^2 - 3028^2) = 3760m

==> Your L/D in that flight = 3760/3028 = 1.42

Cheers,
-Z-

[yuri_base]

zitron the purpose is to study aerodynamics of wingsuits, find optimal body position, make modifications to wingsuit and see glide improvement, etc. The goal is to achieve accurate measurements. Averaging L/D over the whole flight contains no information about dynamics at different AoA. The idea is to use the equations of motion to derive coefficients of lift and drag. All without using a windtunnel.

The speed is currently not measured accurately, because the Pitot tubes are in the boundary layer, so 90mph is less than actual speed. 90mph is my typical horizontal speed, 45 is typical vertical speed for good maxed out flight, so total speed should be sqrt(90^2+45^2)=101mph.

[yuri_base]

Jack Crossfire Interesting. Some portion of the wires is exposed to 100mph wind and that's causing (or at least, contributing) to high noise. I'll need to route wires inside the tube supporting the vane.

Thanks for the thought!

[yuri_base]

kd7ost You're right, the turbulence behind solid triangle is intense.

I'm thinking about using metal mesh as drag elements so that there won't be large "burble" created.

I wish I could use teardrop shape but the limitations of mounting it on my belly are prohibitive. When I deploy my parachute, my body swings forward and legs with the leg wing come up in front of my chest. Traditional vane design - with fins and teardrop body - may hit my legs or tear the suit. This triangular vane design works well - it's compact. Creating drag by using mesh instead of solid surfaces and paying attention to wires should solve the vibration problems, I think.

[kd7ost]

Does the unit need to pivot in a manner different than your body position? Does it need to free float? I see in your picture it is mounted away from your body. But I imagine that your body can also induce turbulence in close and that is why you have it out on such a long arm. You’re trying to keep it out of turbulence induced by the drag of your body. As you stated, the arm itself is in the air flow as are the cables, and being side mounted that is going to have flex. The flex would likely cause more residual motion than the air alone as the unit begins to flop back and fourth. Not to mention the single attachment point is a soft mount when you consider it is on a strap against your body with only one foot.

I also see it is about midway down your longitudinal axis when in flight. At those speeds I imagine that turbulence along your body is increasing away from your body as it moves farther along your body. The blunt head and shoulders would almost dictate that.

What if you built a mount that was tripod in nature? I don’t know what you have to do to contort your body in flight so am just throwing out an idea here. Let’s say something like a chest flat bar that goes just beyond your breasts. Maybe closer to your collar bones would work better. A leg on each end that is short goes straight to an apex below your chin. As close to you as reasonable but close enough to your head that the disturbed air flow is still close in to you. A third leg goes from the apex down to you original mount. This would prevent any side to side or front to front oscillations that would induce increased error. The cable could wind down one of the tripod legs.

Also, being closer to your head would get it out of the disturbed air your body creates as it moves down your longitudinal axis in flight. It will put it in cleaner air.

I could do better if I understood what the mechanics of motion of the sensor needs to be. It must need to pivot or you would just strap it on your chest I’m thinking?

Dan

[yuri_base]

The principle of using accelerometer to measure L/D is explained here:
http://www.dropzone.com/cgi-bin/forum/gforum.cgi?post=2563139#2563139

Yes, the accelerometer needs to pivot and one of its axes needs to be pointing into relative wind, hence the use of a vane. Besides, vane helps point Pitot tubes straight into the wind which helps with accurate speed and altitude measurement.

The mount itself is not wobbling, it sits there very stable. Most of the vibrations come from turbulence created by vane and wires and Pitot vinyl tubing exposed to 100mph wind.

For the next version of the device, I'll increase the length of the mount so that the vane is in more clean air (i.e. farther from the boundary layer).

[kd7ost]

Thanks Yuri,

This is what I see in my head. Please excuse my crude drawings. I know you can see the dynamics in this though. These are both cutaway side views.

Dan

[yuri_base]

Dan, yes, the first design was what I built initially (although it wasn't a teardrop shape, it was just a simple 1/4" tube with a thin fin at the end), but I had to abandon it because it's simply not good to have sharp objects pointing straight at your crotch as you go through the opening shock of slowing down from 100mph to 10mph. :)

[kd7ost]

I kinda figured that jumping out of a plane at high altitude and flying in with that suit marked you as having cast iron gonads anyway. :D

I got a mere 77 jumps in in my Army days. I loved it but today I would be more than happy to take on a ground support role. ;) We went out static line at 800 to 1200' AGL anyway. The fastest we ever moved through the air was the first couple of seconds after going out that side door. I remember thinking "Yay" every time I felt the opening shock. Not much time to fly around. I bet what you're doing is quite a rush.

If you could move that device closer to your chest, I think you could have the clearance. I know you don't want to go back to the drawing board though.

Dan

[yuri_base]

Dan,

Yes, it's very different these days. ;)

http://www.biertijd.com/mediaplayer/?itemid=4262

Yuri

[kd7ost]

Eeeee ha, That's some hairy looking flying. I can't believe that guy skimmed that close to the ground by that highway switchback. One false move and.........Yowsa.

Dan

[mlbco]

Yuri,

If all you want is to estimate your steady state L/D you only need airspeed and altitude data to do this. You'll need to make a pitot system that sticks out 2 ft in front of you (attaching it to your helmet is fine) and you'll need to correct for air density effects from 14Kft down to deployment height. Plan on separate tests for each trim condition and average the data over the flight to get your L/D. L/D is the true airspeed divided by the sink rate (averaged). For example, if we assume the data you presented has an accurate true airspeed value and I average the speed/sinkrate, I get an L/D for that flight of 1.9:1. Ideally you should do each glide at a fixed airspeed, or try to fly constant speed for as long a period as possible. I've done this kind of testing with full size sailplanes many times and the results are good if you correct for density altitude and if you have a good airspeed probe. I don't think you'll ever get the data you desire with accelerometers, they measure too many other effects to be useful for this.

Steve

[CenTexFlyer]

Yuri,
You'll need to make a pitot system that sticks out 2 ft in front of you (attaching it to your helmet is fine)

Attached to the helmet?!?!?! The moment of that 2ft tube at 100 mph could be significant. Shaped incorrectly, even a tube, could start an oscillation that could be deleterious to the wearer's neck regardless of whether the chute opened or not!

[mlbco]

My pitot is made of 1/8" brass tubing and has performed quite well at those speeds (its about 1 ft long). The drag load on a 1 ft 1/8" tube would not exceed 0.4 lb and the torque would be less than .25 ft-lb on the helmet. Don't skydivers strap cameras on their heads with more torque than that?

Steve

[yuri_base]

Steve,

Yes, the torque is not a problem. We jump with cameras and even at 5g's opening shock it's not a big deal.

I'm not just interested in average L/D, I'm interested in the full and accurate dynamics of wingsuit flying. It has a lot of information embedded in it, and the wingsuit equations and my device allow to extract it.

It's like doing a windtunnel test without windtunnel. Instead of air moving and you staying still in the tunnel, it's you're moving and the air is still.

Besides being much less expensive ($20 for a 2-3 minute flight), it's also much more fun!

Yuri

[yuri_base]

Dan,

The guys in that video have hundreds of BASE jumps and thousands of skydives, on average. They totally know what they're doing. :)

Yuri

[mlbco]

Yuri,
I think I understand what you want now.

For dynamic estimation of the aerodynamic forces all you'll need is 2 accelerometers and a pitch rate gyro. You can filter these to resolve the net aero force (in the 'body' axes of the system). Forget about making a pendulum to keep the accelerometer level, this is just a kludge for not having a pitch angle estimator. Assuming you can read all the sensors into a small programmable computer, you can estimate pitch (theta) as follows:


1) Measure the value of the longitudinal accelerometer (Ax) and form a pitch angle 'measurement' from the equation theta_measure=arcsin(Ax/go) where (go) is the gravitational acceleration constant.

2) Estimate theta by integrating the pitch gyro and combining with a correction term from step 1: theta=theta+(pitch_rate-K*(theta-theta_measure))*dt
K is a filter gain that you can solve for using numerical simulation or bench tests of the hardware. When K=0 you are only integrating the gyro. As K increases you are blending in the accelerometer value more quickly. You want a value for K that minimizes any gyro drift effect yet doesn't make the estimate for theta overly sensitive to pure longitudinal acceleration.

Now that you know the pitch attitude the body axis aerodynamic forces are:

Fx_aero=mass*(Ax+go*sin(theta))
Fz_aero=mass*(Az-go*cos(theta))

Steve

[yuri_base]

Steve,

How accurate is gyro's integration over about 20 minutes? (ride to altitude is usually 15-20 minutes, plus 2-3 minute jump itself)

Once I measure accelerations, I still need components of speed to plug into equations of motion to extract lift and drag coefficients. See http://www.dropzone.com/cgi-bin/forum/gforum.cgi?post=2563137#2563137

Ax = g*V*(Kl*Vy - Kd*Vx)
Ay = g*(1 - V*(Kl*Vx + Kd*Vy))

I want to know Kl and Kd. So I still need Pitot probe to measure Vx and Vy, and it needs to be in clean air, so it looks like the kludge is unavoidable.

On a BASE jump, the apparent accelerations measured by accelerometer can vary from 0 (in the first moments of jump, when it's true weightlessness) to ~1.5-2g's (during plane out) over 10-15 seconds - how well gyro can take such changes and still output correct pitch angle?

Yuri

[mlbco]

Yuri,

The method I mentioned uses the accelerometers to zero out the gyro drift automatically. It assumes that accelerations experienced during free fall are sustained for short in durations (less than 15 seconds, for example) and that in the long term the accelerometers accurately measure the tilt angle. The filter combines the accel's and gyro to give you a pitch angle estimate that won't drift and is not sensitive to pure acceleration effects once you properly choose the filter gain (K). I use this type of system on my UAVs and it doesn't drift. It only gets confused during sustained accelerations, such as continuous high-banked 360 deg turns. There are other ways to handle the high-g constant turning case, but they aren't necessary here.

The equations for force that I mentioned are the aerodynamic forces in body axes. They equal the total aerodynamic force on you during the fall and their vector sum is equal to the vector sum of the lift and drag. If you really want to know the lift and drag components in coefficient form (instead of the x and z body axis values of the force) you will need to sense the angle of attack and airspeed. The equations you mentioned require Vx and Vy information, which is equivalent to angle of attack and airspeed information. The airspeed is obtained from the pitot system and angle of attack requires a small weathervane sensor mounted along the pitot tube in free air but not interfering with the pitot function.

The coefficient values of the aerodynamic forces are important for simulation and trajectory optimization studies of cases other than the flight you flew. The total force values in body axes will tell you what you achieved during a particular flight, but will not allow you to study how to optimize your pitch control for future flights. If you really want a complete performance model, it seems like you'll need 2 accelerometers, 1 pitch rate gyro, pitot-static airspeed, and an angle of attack vane sensor.

Steve

[kbosak]

Perhaps a Runge-Kutta or moving average method of smoothing the noisy accel data.
Runge-Kutta is a numerical method of approximating the first derivative in time, when solving Ordinary Differential Equations (i.e. containing d/dt term).
However, when reading accelerometer data, there is only acceleration in 3 axis (not even calibrated: drift, temp etc). This is simply a set of 3 scalar values, nonlinearly related to each other (via bank angle, turn rate, speed etc unpleasant physics behind guaranteed). What can you achieve with RK here? What equation do you want to solve? :confused:

[Tuner]

First off I think your main problem is the 1.7g sensor is not good for what you are doing I would use an 5-8g sensor. The problem is when you max out your sensor you loose data. If the sensor does not max out you can eliminate the noise using post processing of the data.
As fare as a 38g sensors???? I would assume its resolution would be to poor for your measurements but maybe not. From the looks of it a 38 sensor has a sensitivity of 50mv/g and an 8g sensor has a sensitivity of 250mv/g thats a factor of 5. Im sure a 1.7g sensor is even more sensitive but this is what is killing you.


Does the device have to be close to you body.
I personally would use the shuttlecock concept for the device shape and suspend it with a bungee cord.
Make a large shuttlecock and have all the electronics and battery contained in the shuttlecock.
The reason is the more mass in the shuttlecock the more momentum and more it dampens vibration.
I would then use a bungee cord between 2-12feet long. I would attach it to my helmet.

This way the drag of the shuttlecock will stretch the bungee and act as a dampener with the bungee cord, but if the cord stretches to easily it will stretch to its limit and have less of an effect inversly if it is not elastic enough in relation to the drag of the shuttlecock it may not absorb any oscilations also in some situations in between it may cause it to bounce/oscillate.
I would error on using something that stretches easily as you have a lower chance of it oscillating.

Connecting it to the very top of you helmet will have the shuttle cock more in laminar airflow but the longer the cord the more turbulence it will experience. In a perfect world I would use a 1ft mast sticking straight up from the top of my helmet and I would place the device on a 2 ft. cord.

The only issue is its obvious awkwardness. Maybe mounting it on the feet might be better?

If you are going to mount it to you body here are some suggestions.

Well that a totally different way to go about it.
Really I think the problem is your Sensor 1.7G is like nothing.

Hands and head are already stabilized by the body and absorb lots of vibration, but placing the device in your hand seems like a bad choice due to what I would assume are some strong forces generated by the suite.
If you could and I know this sounds stupid but if you could place it in your mouth or attach it to your head very securely and eleminate any contact with your helmet this could solve any issues as you head is very well stabilized.

If you want to do as little as possible?
Mount the unit using shock absorbers. High end portable CD players have some great shock systems in them especially Kenwood players. Or good old FOAM.
The problem is you need to ensure vibrations in your wires are not translated into the unit.

Thats My perspective opinion.
Scott