Vario Comparison

On Tuesday, August 28, 2018 at 7:06:24 PM UTC-7, Mike Borgelt wrote:
On Tuesday, 28 August 2018 23:13:24 UTC+10, wrote:
Mike,

Congrats on the new gadget. It sounds great. Since you are here and talking about TE compensation, there is a dumb question that I've often wondered about.

In theory TE = MGH + 1/2MV**2

but ideally, should V be the plane speed vector with respect to the air or ground?

-Stu

The air. But that isn't as simple as you might think.

More on Dynamis on our website. See the article "Horizontal Gusts" and under products - Dynamis Variometer System.

I'll put up some more articles and maybe comment under "blog" on the website over the next few days. Also pricing.

I'd like to get the next few local systems installed and test flown before release to the wider world.

Mike

My guess is that you incorporated a vertical accelerometer (G meter) to differentiate between a lifting air mass and horizontal gusts (which don't produce vertical acceleration). I believe this has already been done in the Butterfly vario.

Tom

On Monday, September 3, 2018 at 5:11:28 PM UTC-7, 2G wrote:
On Tuesday, August 28, 2018 at 7:06:24 PM UTC-7, Mike Borgelt wrote:
On Tuesday, 28 August 2018 23:13:24 UTC+10, wrote:
Mike,

Congrats on the new gadget. It sounds great. Since you are here and talking about TE compensation, there is a dumb question that I've often wondered about.

In theory TE = MGH + 1/2MV**2

but ideally, should V be the plane speed vector with respect to the air or ground?

-Stu

The air. But that isn't as simple as you might think.

More on Dynamis on our website. See the article "Horizontal Gusts" and under products - Dynamis Variometer System.

I'll put up some more articles and maybe comment under "blog" on the website over the next few days. Also pricing.

I'd like to get the next few local systems installed and test flown before release to the wider world.

Mike

My guess is that you incorporated a vertical accelerometer (G meter) to differentiate between a lifting air mass and horizontal gusts (which don't produce vertical acceleration). I believe this has already been done in the Butterfly vario.

Tom

Unfortunately horizontal gusts definitely do produce a vertical acceleration, since lift is 0.5 * d * Cl * V^2. Because of the V^2 term, they produce quite a lot of vertical acceleration. A 10 knot gust at 60 knots airspeed will give you near 0.4G acceleration.

Butterfly does seem to have worked this out first though.

Lift from horizontal gusts can be filtered out with rates of dynamic pressure and attitude, perhaps with application of polar.

Lots of fun with math.

On Tuesday, September 4, 2018 at 9:07:09 AM UTC-7, George Haeh wrote:
Lift from horizontal gusts can be filtered out with rates of dynamic pressure and attitude, perhaps with application of polar.

Lots of fun with math.

More holistically, you run a Kalman filter using every sensor you have and then some. You treat the system as having three disturbances: two horizontal wind components plus vertical flow. You continuously calculate a solution yielding least square error for the six DOF system with those disturbances being of primary interest.

On Tuesday, September 4, 2018 at 10:59:26 AM UTC-7, Steve Koerner wrote:
On Tuesday, September 4, 2018 at 9:07:09 AM UTC-7, George Haeh wrote:
Lift from horizontal gusts can be filtered out with rates of dynamic pressure and attitude, perhaps with application of polar.

Lots of fun with math.

More holistically, you run a Kalman filter using every sensor you have and then some. You treat the system as having three disturbances: two horizontal wind components plus vertical flow. You continuously calculate a solution yielding least square error for the six DOF system with those disturbances being of primary interest.

Making a trip in the wayback machine to my control theory and aerodynamics classes - apologies if I mess up the details...

If you have a dynamic model for the glider (a Kalman filter would typically require this) you may also be able to use the difference between activating the short period and phugoid dynamic pitch modes with respect to pitch rate. This difference is why a strong thermal has that seat of the pants surge that pitches the nose down instead of up - as you'd get with a gust under normal circumstances.

For all of this you'd get a better result if you also knew the control positions and Cm vs control position - primarily for the elevator. Also, we don't really have dynamic models for gliders though my guess is it would not be that difficult to measure with a reasonably instrumented glider and a couple of test flights. You might be able to get decent results with a generic model for a reasonably current generation racing glider (for instance), though model-specific parameters would of course be better. You're really just trying to filter out the gusts so you may not need anything all that precise to get an improvement, just Cm vs alpha and Cm vs V. Knowing the c.g. and weight will matter as well, but you might be able to calculate these effects.

I wonder if it's easier or harder to use machine learning to do this than a more deterministic least squares model...or if all of the above is overkill.

Andy Blackburn
9B

On Tuesday, September 4, 2018 at 12:49:36 PM UTC-7, Andy Blackburn wrote:
On Tuesday, September 4, 2018 at 10:59:26 AM UTC-7, Steve Koerner wrote:
On Tuesday, September 4, 2018 at 9:07:09 AM UTC-7, George Haeh wrote:
Lift from horizontal gusts can be filtered out with rates of dynamic pressure and attitude, perhaps with application of polar.

Lots of fun with math.

More holistically, you run a Kalman filter using every sensor you have and then some. You treat the system as having three disturbances: two horizontal wind components plus vertical flow. You continuously calculate a solution yielding least square error for the six DOF system with those disturbances being of primary interest.

Making a trip in the wayback machine to my control theory and aerodynamics classes - apologies if I mess up the details...

If you have a dynamic model for the glider (a Kalman filter would typically require this) you may also be able to use the difference between activating the short period and phugoid dynamic pitch modes with respect to pitch rate. This difference is why a strong thermal has that seat of the pants surge that pitches the nose down instead of up - as you'd get with a gust under normal circumstances.

For all of this you'd get a better result if you also knew the control positions and Cm vs control position - primarily for the elevator. Also, we don't really have dynamic models for gliders though my guess is it would not be that difficult to measure with a reasonably instrumented glider and a couple of test flights. You might be able to get decent results with a generic model for a reasonably current generation racing glider (for instance), though model-specific parameters would of course be better. You're really just trying to filter out the gusts so you may not need anything all that precise to get an improvement, just Cm vs alpha and Cm vs V. Knowing the c.g. and weight will matter as well, but you might be able to calculate these effects.

I wonder if it's easier or harder to use machine learning to do this than a more deterministic least squares model...or if all of the above is overkill.

Andy Blackburn
9B

I agree Andy; to do a Kalman filter you'd want gyros, accelerometers and control position sensors. Position sensors are not hard though and their mapping might be learned for each installation with a smooth air test flight. The problem with Neural network AI is that you have to begin the process with a comprehensive training set. Probably Mike Borgelt has a simpler and better way to get to just the goods that we care about without a plane load of sensors. I'll be looking forward to hearing more about this.

On Tuesday, September 4, 2018 at 2:40:05 PM UTC-7, Steve Koerner wrote:
On Tuesday, September 4, 2018 at 12:49:36 PM UTC-7, Andy Blackburn wrote:
On Tuesday, September 4, 2018 at 10:59:26 AM UTC-7, Steve Koerner wrote:
On Tuesday, September 4, 2018 at 9:07:09 AM UTC-7, George Haeh wrote:
Lift from horizontal gusts can be filtered out with rates of dynamic pressure and attitude, perhaps with application of polar.

Lots of fun with math.

More holistically, you run a Kalman filter using every sensor you have and then some. You treat the system as having three disturbances: two horizontal wind components plus vertical flow. You continuously calculate a solution yielding least square error for the six DOF system with those disturbances being of primary interest.

Making a trip in the wayback machine to my control theory and aerodynamics classes - apologies if I mess up the details...

If you have a dynamic model for the glider (a Kalman filter would typically require this) you may also be able to use the difference between activating the short period and phugoid dynamic pitch modes with respect to pitch rate. This difference is why a strong thermal has that seat of the pants surge that pitches the nose down instead of up - as you'd get with a gust under normal circumstances.

For all of this you'd get a better result if you also knew the control positions and Cm vs control position - primarily for the elevator. Also, we don't really have dynamic models for gliders though my guess is it would not be that difficult to measure with a reasonably instrumented glider and a couple of test flights. You might be able to get decent results with a generic model for a reasonably current generation racing glider (for instance), though model-specific parameters would of course be better. You're really just trying to filter out the gusts so you may not need anything all that precise to get an improvement, just Cm vs alpha and Cm vs V. Knowing the c.g. and weight will matter as well, but you might be able to calculate these effects.

I wonder if it's easier or harder to use machine learning to do this than a more deterministic least squares model...or if all of the above is overkill.

Andy Blackburn
9B

I agree Andy; to do a Kalman filter you'd want gyros, accelerometers and control position sensors. Position sensors are not hard though and their mapping might be learned for each installation with a smooth air test flight.. The problem with Neural network AI is that you have to begin the process with a comprehensive training set. Probably Mike Borgelt has a simpler and better way to get to just the goods that we care about without a plane load of sensors. I'll be looking forward to hearing more about this.

What I know is I don't have a Kalman filter going in my head: but I do have a butt which feels vertical acceleration. If it doesn't tell me I am going up, I discount the screaming vario.

Tom