[Nut-upsuser] BNT-1000AP

[Arjen de Korte]

> Well, I can tell you that the frequency should in fact be close to 50Hz
> (well, in theory it should be exactly, but we don't live in a perfect
> world).

One way to get at least the nominal value correct, is by averaging the
measured frequency (one reading every minute) over several days. That
should be exactly 50.00 Hz, since utility companies will keep the longterm
average at the nominal frequency (although the instananeous frequency may
drift somewhat).

> Changing the coefficients by trial and error I managed to get
> it close to what I would expect, and close enough to be good enough for
> knowing when we hit battery low:
>
> # upsc blackknight at localhost
> battery.charge: 93.0
> driver.name: powercom
> driver.parameter.batteryPercentage: {5.0000,0.3268,-825.00,0.93022,0}
> driver.parameter.frequency: {0.00041994,0.0}
> driver.parameter.manufacturer: BlackKnight
> driver.parameter.modelname: BNT-1000AP
> driver.parameter.port: /dev/ttyS1
> driver.parameter.type: KIN1500AP
> driver.version: 2.0.5
> driver.version.internal: $ Revision: 0.5 $
> input.frequency: 47.63
> input.voltage: 226.7
> input.voltage.nominal: 230
> output.frequency: 47.63
> output.voltage: 226.7
> ups.load: 23.8
> ups.mfr: BlackKnight
> ups.model: BNT-1000AP
> ups.model.type: KIN1500AP
> ups.serial: Unknown
> ups.status: OL
>
> Now, from experience, I've never actually seen a battery at 100 %, but
> usually between 91 and 98 % ... so I'm rather happy with that, and the
> frequency is close to 50Hz ... I'm not going to worry too much about
> that not being accurate - unless someone can tell me why I should be.

Well, it would be nice if the coeefficients would be correct, but that
would probably also require the results from multiple units to come to an
average for this model.

> Now, the values I've changed are as follows:
>
>      frequency = {0.00041994,0.0}
>      batteryPercentage = {5.0000,0.3268,-825.00,0.93022,0}
>
> The values are based on the defaults in the powercom man page, and I've
> doubled the first value for frequency, and I doubled the second last
> value in the batteryPercentage list.
>
> I haven't seen any documentation (haven't looked too hard yet either) as
> to what each of values mean, so the it was done entirely on trial and
> error.  What is the "scientific" way of approaching this?

Since the function that is used to convert the readings from the UPS to
the actual values is first order, you would need at least two points to be
able to get the right coefficients. To reduce the impact of the resolution
of the analog-digital conversion, you'd want two points that are
reasonable spaced apart.

Frequency calibration requires the ability to

  a) Change the input frequency to known values of (for instance) 47 and 53
     Hz.

   or

  b) An accurate frequency measuring device and enough variations on the
     mains frequency (by changing the engine speed on a generator set for
     instance).

Calibration of the battery charge is easier, for this you only need a
fixed load of about 50% of the nominal rating of your UPS (a set of
lightbulbs or an electric space heater for instance). Starting with a set
of fully charged batteries (so that you know the initial charge is 100%),
you'd run the UPS until it switches off automatically, while monitoring
the reported value of the charge at regular intervals. Since you know the
start- (100% charge) and endpoints (0% charge), you'd be able to
interpolate all the values in between and relate them to value reported by
the UPS. For the battery charge when charging it, you can follow the same
route. Starting with a flat battery, note the value reported immediately
after starting to charge it (0% charge). Then charge it for an hour and
then briefly remove the mains and immediately check what value is reported
now that it is running on battery with the same load as before. You can
then use the values determined in the discharge test to estimate the
battery charge you have at that time. Keep doing that until the reported
value doesn't really change anymore (should take somewhere between 6 to 12
hours). Last you need to calculate the coefficients for both the charging
and discharging states. Currently we use a first order function, but you
can switch to higher orders if needed/desireable.

> I've basically got access to a multi-meter, but I can't measure
> frequencies with that, so I don't know how much worth that is.

To calibrate battery charge, you only need a fixed load. The exact value
of the load is not very important, as long as it is constant.

> Also, how do you actually measure what the actual battery charge is?

See above.

> I reckon the above is good enough for my purposes.  Thanks for prodding
> me to fiddle with those coefficient values though.

I will add this UPS to the driver if I can find some time in the next few
days. Thanks for your efforts.

Best regards, Arjen