Boat speed calibration mysteries

[ronstory]

Your data looks pretty solid and remember the DST800 is a "smart device". I'll bet you a beverage their is an averaging algorithm in the DST to keep the data stream looking smooth and not noisey. Those algorithms don't have to be linear for increasing and decreasing and their maybe some programmed interia in it.

Think about a modern 'smart' gas gauge in a car. When you add gas with ignition on, it increased rapidly but will not drop a quickly since the usage it a slow burn rate then fill rate. That is programmed in to the algorithm.

You expect a boat to accelerate briskly but slow down slowly (unless you use reverse) so it I was programming that algorithm, it would tweak it to the expected customer usage. It helps avoid support calls. ;^)

... plus... you are not the typical user. Most would be thrilled with the accuracy you already have.

[mglonnro]

Originally Posted by Paul L

I can't come up with a justification for the dip at 4 kts, so it is hard to extrapolate that far out. Based on the data you have, I'd stick with the gain adjustment you have at 6 kts and use it above that.

I looked at the data with more resolution and tried to select parts with no or very small changes in heading/speed, and now I got a linear relation without the dip

[mglonnro]

I made changes to the graphing system to show calibrated values, so here are those, before and after

[Dockhead]

Originally Posted by Cadence

Not to be crass. Does it really matter? If you are racing maybe for pleasure who cares.
I'd guess your running against the current. Are all taken under the identical situations?

Just to underline what mglonnro already said:


Accurate STW is essential for a valid true wind calculation. Without valid true wind, you can't know VMG to windward so you can't find your best angle beating. You can't know your heading on the other tack. And you can't know where you're going to gybe. Accurate STW is also essential for knowing whether you are hitting your target speeds -- whether you are trimmed right.



Needed for any kind of half-keen sailing, not just for racing.

[Dockhead]

Originally Posted by tkeithlu

. . . Your GPS is comparing sequential locations. It does not actually measure speed. It derives speed by how much time separated readings at two different locations and how far apart they were. . . .

Originally Posted by Rapanui

There is one answer and it's due to latency in the GPS. Nearly all GPS have significant latency in their position data, often 0.5 secs or more, and then to calculate speed, they will use data from several previous positions which is of course even staler. So on acceleration the GPS speed will be lower and during deceleration, higher. Found this out by doing land tests on a GPS setup to measure the contractual stopping distance of a warship.
Latency should not be confused with update rate, the GPS may update at 10 Hz, but have considerable latency.
If your data is in a spreadsheet, you can artificially delay the water speed data by a varying amount until you get a consistent calibration across the sample. Mathematically, when the standard deviation of the two parameters is minimised, at that point, the slope is the speed calibration factor.

That is however not how our GNSS devices calculate speed. Don't know about on your warship.

Differential position calculations (what you and Tkeithlu described) are only accurate to a meter per second (2 knots). So all marine GNSS devices use carrier phase doppler shift to calculate SOG. See:

https://insidegnss.com/how-does-a-gn...mate-velocity/

These calculations are near instantaneous. But they are damped, which accounts for the lag. You can usually set the damping (or turn it off) in your network. I can do it with mine.

[Dockhead]

Originally Posted by mglonnro

So,

Today, before starting the sail towards our home harbor, I did some calibration runs again by motoring in opposite directions at a few different GPS speeds: about 2 knots, about 4 knots, and about 6 knots.

Here's the whole picture of the different runs (same as before, SOG in the background, slightly slower most of the time):



Here are the 2 knots runs:





4 knots.





6 knots





Now if I put these figures into an excel and average the opposite runs, I get:



If I then draw the correction percentages into a curve, I can get an estimated polynomial function for that:



This function I can then use as a "correction table" when feeding the N2K system with corrected boat speed values. I GUESS

Here are some sample corrections:

This is great, but what about heel? Paddlewheel transducers are greatly affected by heel angle, and we are going to be caring most about accurate true wind when we are heeled and beating upwind.

[Paul L]

Originally Posted by mglonnro

I looked at the data with more resolution and tried to select parts with no or very small changes in heading/speed, and now I got a linear relation without the dip

Now that is starting to look a little closer to how you would expect the world to work

If you want to play some more, you might want to try fitting a correction using a least squares regression. It will give you the best fit line for your data points.

[mglonnro]

Originally Posted by Dockhead

This is great, but what about heel? Paddlewheel transducers are greatly affected by heel angle, and we are going to be caring most about accurate true wind when we are heeled and beating upwind.

Yes, this is absolutely the next challenge on the challenges list.

So far I haven't come across any other ways of estimating this than doing similar ("measured mile") calibration runs and recording/analyzing the results.

How to then separate the heeling effect from leeway, I have no idea.

Any reasonable ideas?

I will try to record some data when I get out on the boat again.

[Dockhead]

Originally Posted by mglonnro

Yes, this is absolutely the next challenge on the challenges list.

So far I haven't come across any other ways of estimating this than doing similar ("measured mile") calibration runs and recording/analyzing the results.

How to then separate the heeling effect from leeway, I have no idea.

Any reasonable ideas?

I will try to record some data when I get out on the boat again.

I don't think you need to disagregate heel from leeway.


Heel (without regard to associated leeway) directly affects paddlewheel calibration, so if you want accurate STW you need to compensate for that. The H5000 has a lookup table for this calibration.


Leeway separately affects true wind. You need that separately from well-calibrated STW. But I think leeway can be computed in a straightforward way (discussed in a separate thread I started recently). If you can collect a few empirical data points then you can compute leeway over a wide range of conditions with just two variables -- STW and heel angle.

[mglonnro]

Originally Posted by Dockhead

I don't think you need to disagregate heel from leeway.


Heel (without regard to associated leeway) directly affects paddlewheel calibration, so if you want accurate STW you need to compensate for that. The H5000 has a lookup table for this calibration.


Leeway separately affects true wind. You need that separately from well-calibrated STW. But I think leeway can be computed in a straightforward way (discussed in a separate thread I started recently). If you can collect a few empirical data points then you can compute leeway over a wide range of conditions with just two variables -- STW and heel angle.

Ok! So any ideas on how I should do the calibration?

In the real world, I think I could manage to get data like this for a few minutes of sailing:

- SOG, boat speed average +/- 0.5 knots along the way.
- Average heading and sog +/- 2 degrees.
- Average heeling +/- 5 degrees (or something like that)
- Start point and end point (GPS position)

My first experiment was with (almost no) wind and current, so the boat was going where it was heading, and it was easy to adjust for that.

When heeling, it's not going where it's heading, so how do I separate the heeling effect on boat speed from the lateral movement effect? Or don't I have to?

[Dockhead]

Originally Posted by mglonnro

Ok! So any ideas on how I should do the calibration?

In the real world, I think I could manage to get data like this for a few minutes of sailing:

- SOG, boat speed average +/- 0.5 knots along the way.
- Average heading and sog +/- 2 degrees.
- Average heeling +/- 5 degrees (or something like that)
- Start point and end point (GPS position)

My first experiment was with (almost no) wind and current, so the boat was going where it was heading, and it was easy to adjust for that.

When heeling, it's not going where it's heading, so how do I separate the heeling effect on boat speed from the lateral movement effect? Or don't I have to?

You're asking the wrong person -- I'm sure there are people on CF who know vastly more about this than I do.

But FWIW, here is my amateurish take on it --

1. We're looking for TWD and TWS, right?

2. To get TWA and TWS, we need to start with our course through the water (not HDG) and STW. Subtract this from directly measures AWA and AWS to get TWA and TWS. OK?

3. So the first problem is STW. The raw STW data needs to be compensated for heeling (and for any other non-linearity or any other error, to the extent possible). Once we've done that, then we have usable, calibrated STW.

4. The second, and entirely separate problem is course through the water. We need leeway, calculated somehow, to apply to HGD to get this.

5. Another problem is calibration of AWA and AWS. This needs to be compensated for heeling and for non-linearity or other errors of the transducer. There are standard tables which can be put into an H5000 to do this with standard mechanical wind transducers. For ultrasonic ones I don't know. I started a separate thread about that. One possible answer may be that because these transducers are much less sensitive to heel, we can ignore this.

6. So once we've done all of this: (a) calibrated STW, compensated for heel and for other inaccuracies; (b) calculated course through water once we know leeway; (c) calibrated AWA and AWS -- then we have all of the elements needed to calculated usable TWA and TWS -- good data in, good information out, instead of garbage in, garbage out.

Does that make sense? Some more knowledgeable person will no doubt correct anything which I misunderstand or didn't describe correctly.

[DEC2017]

Mglonnro: I think tkiethlu nailed it. The difference in pressure between acceleration and deceleration is the key perhaps due to a little back pressure. The key is to calibrate on the averaged data at constant speed - steady state

[Cynara]

why? are you going to be constantly looking at the chart to figure out how fast you are going for that second.

Feel fortunate you don't have any other work to do on your boat

[Paul L]

Originally Posted by DEC2017

Mglonnro: I think tkiethlu nailed it. The difference in pressure between acceleration and deceleration is the key perhaps due to a little back pressure. The key is to calibrate on the averaged data at constant speed - steady state

You have two sensors that are sampled at different rates and averaged over different periods, as well as having different reporting timing lags. Not sure why you would then attribute accel/decel differences to pressure.

[LakeSuperior]

I didn't read all the responses in detail so I may be repeating.

You need to recognize that the true velocity vector is the vector sum of the forward velocity vector and the slipage or leeway velocity vector plus or minus 90 degrees from forward velocity. This true velocity vector, speed and direction, is what is measured by the GPS using doppler.

Your paddle wheel boat speed sensor is only measuring the forward velocity vector. This paddle wheel sensor does not respond to the leeway speed. For example, if only leeway and no forward speed the paddle wheel would measure zero speed. I don't think the paddle wheel errors because of the heel however the leeway would be correlated to the heel angle.

Leeway speed will vary between zero and some amount depending on your boat, wind speed, and how you are trimmed.

Consequently, you will see a varying speed mismatch depending on the slippage speed along with all of the other calibration, surface current, and GPS data damping errors.