Low and critical battery warnings are triggered to soon

[julianoes]

I think we should either lower the thresholds that trigger low and critical batteries or instead filter much slower so that sudden peaks don't always trigger warnings.

[thomasgubler]

I agree the current solution is too sensitive against sudden throttle peaks.

[DrTon]

It can be tuned with BAT_V_EMPTY and BAT_V_FULL parameters. Default settings maybe really a bit sensitive. But the problem is that actual voltage under load depends on copter, battery model and battery temperature significantly. Look at typical discharge curve: http://www.rcheliresource.com/wp-content/uploads/2009/05/tp-5000mah-pro-power-40c-cell-discharge-curve-graph.jpg

Voltage range shouldn't be used for remaining charge estimation, but if battery capacity not set or no current sensor available it's better to warn user too early than too late.

BTW, battery configuration is documented part of setup: http://pixhawk.org/users/battery_config

[LorenzMeier]

@DrTon The discussion here is that transients trigger the alarm (which they shouldn't). Its not about the absolute charge level.

[DrTon]

It's exactly about charge level, because batt "warning" triggered on 25% remaining charge and "critical" on 10%.
Look: https://github.com/PX4/Firmware/blob/master/src/modules/commander/commander.cpp#L1023
And if you tune battery parameters as described in Wiki everything will work perfectly. I think you even not read it...

Example configuration for 3300mAh 3S battery + PIXHAWK + APM Power Module (voltage and current sensors):

BAT_V_SCALING 0.0082
BAT_C_SCALING 0.0124
BAT_CAPACITY 3300.0
BAT_V_EMPTY 3.3
BAT_V_FULL 3.5
BAT_N_CELLS 3

Compare this with your params. ;)

By default battery capacity is NOT configured, because we have no idea about it. But for known configurations (e.g. Iris) we can and should add this to autostart parameters, I missed it.

[LorenzMeier]

@DrTon Its not about what the system is capable of if set up correctly. Its about what it does as default. Since we don't know the battery capacity and need to operate voltage based, we need to fix the voltage based operation anyway so that the default works.
We then also need GUI support for the current based setup. But as long that's not there we need sensible defaults.

[DrTon]

Ok, but if one happy user will plug warm, very good battery on light copter he will get first batt warning at 3% remaining battery and the second on 1% (look discharge curve). Another user will still get warnings at 70% and 60% (old, cold battery under heavy load). Is it ok??? I got some dead bubbles instead from batteryes with this approach...

[DrTon]

Owners of copters that supported out-of-the box will have already configured battery capacity and voltage (we know exactly or almost exactly what battery and current sensor they have). Other users with custom setups will need to tune at least PID, and must have some basic knowledge about capacity, voltage etc. and should be able to set 2-3 parameters, isn't it?

[LorenzMeier]

In theory yes, but we still need the under voltage protection in that mode. If the user doesn't charge the battery between flights we will have a wrong estimate and need to fall back to voltage. And then we again need to reject short transients. No matter what we do, we need to fix them 8).

[DrTon]

I think you haven't reviewed my batt fixes PR :)
Yes, we need, and we already have:

remaining_charge = min(estimate_voltage, estimate_discharged)

But for this protection voltage range should be on the lower end of discharge curve (and I wrote about this in Wiki):

BAT_V_EMPTY 3.3
BAT_V_FULL 3.5

To be sure that on high voltages only discharged capacity will act, but leave voltage only as protection.

What about short transients, now lowpass filter is already much slower than before my fixes, we can, of course make it even more slow, but it may be dangerous. Not fully charged battery may go <9V very fast, while filtered value will be still ok. I have one idea to compensate internal resistance of battery, model in FlightPlot works great, but need one more parameter: effective internal resistance, that depends not only on battery model, but also on battery temperature. So I decided to not complicate things.

[LorenzMeier]

I'm aware of that - we still have the issue with the default setup 8). How parameter sensitive was your resistance model? I agree we should not rely on anything that makes us sensitive, just trying to understand if it could be viable.

[DrTon]

Look, blue is simply voltage-based estimate filtered at 0.5Hz. Yellow is resistance compensated estimate also filtered at 0.5Hz. Resistance in my case (3300mAh, cold weather, ~5C) was 0.04 Ohm, should be tuned with at least ~20% precision for good results.

[DrTon]

Green is discharged current based estimate, but capacity was set to 2200 instead of 3300, I forgot to change :)

[LorenzMeier]

Interesting - if you set the resistance significantly (50% off) too low or too high, how do the plots look like? I'm trying to understand if we can find a mode where we guard against wrong parameters / learn the right ones without needing excessive tuning.

[DrTon]

50% off in resistance value will give ~ average from blue and yellow.

[julianoes]

Addressed in #1211, let's test...