Power calibration by power table #257
Conversation
This function is enabled by commandline option -L LED IO-port 25 shows USB-Tacx connected LED IO-port 25 toggle shows Data transfer LED IO-port 24 shows cadence present, to check the pacement of Reed-Contact and Magnet LED IO-port 23 show ANT+ Dongle connected
…usANT into Raspi-Status-LED
This added function compensates the non linearty at low power and high spieed by the magnetic Brake. It uses measured power values from a power table.
|
@decodeais Which head unit are you using with the Flow? T1902 (solid green) or T1904/T1932 (white/green or white/blue)? For the T1904/T1932 a lot of work has gone into the power curve (see #143 and #153) and the non-linearity is baked into the curve already. However accuracy depends a lot on the rolling resistance of the tyre (tyre material, pressure, trainer knob setting). The -c parameter can be used to account for that but there is no automatic calibration yet. The T1902 is still using an older, less-acurate power curve. I have done some work on applying the new one but haven't gotten around to implementing it unfortunately. |
|
I am using the t1932. Unfortunately I have no power meter to measure, but the tracks on Rouvy look much better. Now it is possible for me and my wife to ride with the same parameters. Our FortiusAnt program runs headless and so it was difficult to change the parameter every time. Our Problem only appeared on high speed with low power (<150W downhill). If we changed the c value we got to much power in other speed/power regions. |
|
@decodeais You are very welcome to try to improve the power model if you want. The two issues I linked have at least 3 independent sets of powermeter data that you could analyze. The data-quality may not be high enough for a (fine-grained) table-based model though (it was something we considered but ultimately didn't use). In theory, a table-based model could give higher accuracy, unfortunately it will not remove the need for individual calibration. The rolling resistance of the tyre is a large contribution a lower speeds and it varies a lot depending on the tyre, tyre pressure, temperature and cylinder pressure (the knob on the trainer). The power curves we currently use look something like this (this is not the final model but similar): When it comes to ride-feel, rather than just power estimation, it is also important to remember that the resistance range of the Flow is rather limited. This doesn't just limit the maximum gradients that can simulated, there is also a minimum resistance. This plot shows the range (in green) for a 70kg rider + 10kg bike with -c 10. |
|
Hi all. Sorry for not engaging earlier in this discussion... but here I am :-) We have a come a long way, leaving the power-tables [from antifier] behind us.
Note that the power-curve can be customized with -c -G -p (ref manual). I am quite open to adopt improvements and extensions, but I think this will not be one of them. So please let me know... |
|
Sorry, my intension was not to start a great discussion about the calculations in the software. I hoped the easy possibility to make own adjustments would help to end this kind of discussion. If this table is erased everything is changed to default. |
|
@decodeais I understand your intention and it is true that a look-up table gives a lot of flexibility. But in practice it is unfortunately a lot less useful than it might seem because you need to know the correct values. If you don't have a power meter all you can do is guess (and in all likelihood you will end up making things worse). If you do have a power meter it is still a lot of work and the correct values for my setup will be different from yours. And then if you pump up your tire tomorrow you might have to do it all over again. Our power model is not as bad as you might think. Across the three different trainers and power meters we tested it was usually better than 10% accurate across a wide resistance and speed range. The main problem is that this is only true if you know the proper calibration value for your bike and trainer and right now there is no good way to find it without a power meter. But replacing 1 unknown parameter with 36 is not really going to fix that. There is a proper solution (spin-down calibration) and I am working on that. In the process I will probably collect an additional high-quality power meter dataset as well. But as you can imagine probably imagine after your own experiments that is a lot of work and I don't have a lot of time right now so it will probably take a while. In the mean time the best advice I can give is this:
About the tcx files: Rouvy uses its own physics simulation internally using only the power value. So if the speeds in particular are different that is not surprising. If the power or cadence values are different something is wrong. |
|
I am sure you won't believe both graphics show the same ride. The first shows the uploaded tcx-file the second was from rouvy.
!!Attention corrected the pictures, it were so many tests. |
|
I do believe you.
|
|
I created the tcx-file in Rouvy and found a mismatch when i compared the date. But the right one shows the same problem. |
|
@decodais, @switchabl I would suggest not to integrate the power table. |
|
It is OK. My wife wants to make some tests when the weather is not so good. If she insists on this table I will make a special version for her. |
|
My wife also has her own Fortius. For now closed; if you need support just pop-in again. Reason to close that I want to keep issues and pull-request actual, to be able to keep track. |
For me the original power measurement on the Tacx Flow worked ok, but in the rouvy tracks of my wife it was obvious that there was much power uphill but very low power downhill. I looked to the power table for the Tacx Flow in the Manual an found a lon linarity at high speed and low power. This looked not very important, but when I compensated it, downhill looked much more realistic for my wife.
This bilinear interpolation could make sense on other trainers too.
Please have a look on it