More weaverbird plots. Should have started with a picture of the raw data before reduction to medians then furthur reduction by a smoothing function but the coding is more difficult so saved it for last.

I created polar plots showing the point scatter about the median STWs (red squares) separately for each TWS bin. If combined into one plot it'd be a hard-to-interpret mess of points. Only the "even" TWS bins (4,6,8,... knts) plots are shown below. The "odds" plots which use the other half of the data are similar and show the same trends and can be found here:

The medians tend to be in the center of mass of the point scatter. If sample size is small then that median is suspect.

If sample size per bin is large relative to the natural variability then outliers have very little effect on the median.

If the intent is to use these data to detect "before" vs "after" rig performance on the same hull then:

Might be tough....the variability is large and so the data overlap will also be large given you're trying to detect small percentage changes.

One, perhaps convincing, result would be if all the medians for rig A were higher than rig B even though the points overlapped.

Another possibility if comparing bermuda vs junk is a take-off from the perception that cambered junk is slower beating but faster off the wind. If true then there should be a STW crossover around 90 deg TWA? in the data.

rself

Anonymous wrote:

Reading the manual for the LJC Capteurs CV7 wind sensor, I note that it updates the output at the rate of 2 per second. Wouldn't that be a reasonable rate for water speed as well? Is there any advantage in collecting water speed data at 5 per second?

The new multiplexer should allow us to comtrol these things better than the old one. The old Vyacht Mx had no priority controls, so in practice the boat speed sentence was being crowded out by other sentences which were less important or irrelevant to us. This meant that regardless of how frequently the ST900 sent boatspeed data, it only changed every few seconds, as you can see from the .csv file.

Personally I think recording the data every second is plenty. bear in mind that if you double the rate you double the amount of data to be processed.

Reading the manual for the LJC Capteurs CV7 wind sensor, I note that it updates the output at the rate of 2 per second. Wouldn't that be a reasonable rate for water speed as well? Is there any advantage in collecting water speed data at 5 per second?

Anonymous wrote:

Anonymous wrote:

That is very interesting, Robert, and thank you for doing this work, but I think the fact you are taking the median values from a group of data points will give a lower performance than it is realistically possible to obtain.

It would be better, IMHO, to take the maximum figure from each "box" of data. This is the approach taken by Polauto, and by Anthony Cook.

It appear so me that median values will give a better picture of the performance than just picking the maximum figures:
High peak values can easily be obtained by picking up speed on a lower course and then shoot up a little, which would give a very good VMG reading if caught in the right moment.
Note: I don't accuse anyone for cheating, I just know from my own experience that it is difficult to keep a correct course all the time. Sometimes I aim a bit low (and pick up speed) and sometimes I aim a bit high. I guess I am not alone on this.

Arne

That is a risk, and when doing the trials we developed a technique of setting the sails for each point of sail and then hunting up and down a bit (quite slowly) through the pointing angles to get the best data we could with the wind that was available at that moment. It is clear that much of the data we gather is at less than top performance, which is why I think it is right to use the best performance we can measure.

With Anthony Cook we are just now trying out a filtering process to take out data where the wind is dropping rapidly, which can also lead to falsely high performance, as the boat speed drops more slowly than the wind speed. We have looked at taking out data where the wind speed drops in one second by 0.7 knots or more, and also 0.6, and 0.5. The number of data points rises rapidly as the filter is reduced.

The problem with using median values is that we record a lot of data at low boat speed, when we are starting and finishing a run, for example, or retrimming the sails, and these data points will pull the median values down in a fairly arbitrary manner.

I tried to avoid any fast luffing manoeuvres when doing the trials, as they would create misleading data. I guess we can try editing out data where the wind angle is changing more than, say, 2 degrees per second, which would help to protect against any such cunning manoeuvres.

Fortunately, once we have the data, we can choose how to analyse it, and discuss the merits of each approach. With no data, we have nothing.

Anonymous wrote:

That is very interesting, Robert, and thank you for doing this work, but I think the fact you are taking the median values from a group of data points will give a lower performance than it is realistically possible to obtain.

It would be better, IMHO, to take the maximum figure from each "box" of data. This is the approach taken by Polauto, and by Anthony Cook.

It appear so me that median values will give a better picture of the performance than just picking the maximum figures:
High peak values can easily be obtained by picking up speed on a lower course and then shoot up a little, which would give a very good VMG reading if caught in the right moment.
Note: I don't accuse anyone for cheating, I just know from my own experience that it is difficult to keep a correct course all the time. Sometimes I aim a bit low (and pick up speed) and sometimes I aim a bit high. I guess I am not alone on this.

Arne