Hybrid Car – More Fun with Less Gas

Filtering harmonic distortion on a 2.8KV 50Hz 230v Petrol Generator to satisfy fussy UPS

register ::  Login Password  :: Lost Password?
Posted by Johnny B Good on August 3, 2009, 8:30 pm
 Hi everyone.

 I've done some more research into the generator / UPS compatability
issue and can now say that I was totally on the wrong in my original

 It took quite a bit of experimentation (and some serendipity) to
discover that the UPS isn't particularly concerned about the harmonic
content of the incoming supply voltage waveform.

 The main parameters of concern to a UPS are the frequency (the
requirement being quite a loose tolerance of +/- 5% on either a 50 or
60Hz supply) and the voltage remaining within predefined set limits.

 In my case, although the generator was running just beyond the upper
frequency limit as initially set up, this was quite easily adjusted, and
the real problem I was having turned out to a consequent of capacitive
loading on the generator's output.

 Now, despite the fact that the generator design involved a conventional
2 pole rotating field energised via sliprings from an AVR module, it is
still susceptable to the self excitation effect derived from the leading
current due to the capacitor loading presented in the form of a pair of
4.7uF capacitors in the UPS's mains input circuit.

 Effectively, this capacitive self excitation was hijacking the AVR
control by sending the output voltage right up to 280v, some 50 volts
above the AVR mediated setting.

 What was happening was that the UPS would see the reappearance of the
genset supplied mains voltage at acceptable voltage and frequency and
attempt to switch from battery power back to mains, whereupon the
9.4uF's worth of capacitive loading would appear across the supply
causing the voltage to jump from its nice steady 230v to somewhere in
the region of 270 to 280v, way beyond the buck boost regulation range of
the UPS, forcing an immediate return to battery power allowing the
generator to once more stabilise at its AVR mediated 230 volt level
which then initiated another attempt by the UPS to switch back to mains
power and a repeat performance.

 Now one way to mitigate this effect is to preload the generator output
with a suitably inductive load (an inductor wired across the output).
This will work but a 230v 50Hz rated 500mH (or lower) inductor is a far
from standard item. The best I could do was to parallel a bunch of 4H
transformer primaries (some 8 400VA transformers in all). This allowed
the unloaded UPS to switch back to generator power and remain in that
state but the loaded state would cause it to cycle between battery and
mains as before.

 In this case, the problem is almost certainly due to the fact the load
actually consists of yet another UPS providing a second level of
protection to my computers and I suspect that this UPS, an Upsonic
UPS600, also has its own bunch of mains input capactors to further
agravate the capacitive loading problem. Unfortunately, I don't have the
circuit diagram for this model so can only surmise at the existance of
the extra capacitance.

 I haven't, as yet, removed the redundent Upsonic UPS600 from the power
supply chain to test this theory but, when I do, I'm going to fit an
autotransformer in its place to reduce the voltage from a nominal 230
down to around the 190 to 200v mark to minimise damage to the VDR spike
protection components in the PC's PSUs.

 Whilst this might provide a solution, I think the real culprits  behind
this universally experienced UPS compatabilty issue are the cheapjack
penny pinching genset manufacturers who seem to have elected to use an
AVR circuit derived from automotive alternator design practice.

 The automotive alternator doesn't have to contend with the possibility
of self excitation effect from capacitive loading due to the 3 phase
stator windings being directly connected to a 3 phase bridge rectifier
pack, thus allowing an  AVR module that only needs to control excitation
current via a single series pass power transistor (effectively, half an
output stage) in just one direction only.

 When the AVR has to contend with capactively induced self excitation,
it seems to me that it needs to be able to counter this effect and this
requires that the AVR is able to drive excitation current in _both_
directions. This can only be achieved with an output that can both
source and sink current which requires a push/pull output stage of two
series control power transistors as per a single ended audio power amp.

 A 'drop in' enhanced regulator module capable of using the exisitng
excitation supply voltage and field assembly would need to be a bridged
output design (i.e. four output transitors). Although it might be argued
that this raises complexity and costs, the additional cost becomes
rather swamped out by the costs of the gross parts of the genset itself.
The complexity issue is a non- issue once a new module design has been
committed to mass production anyway.

 However, the idea of using such an 'improved' AVR module might have
some fundamental flaw which I've not been able to discern (but which an
experienced genset designer might be able to point out), so I'd be quite
happy to listen to anyone who can put me straight on this matter.

 So, for anyone else, whose dream of providing extended UPS runtimes
courtesy of a cheap petrol genset, have been well and truly shattered by
the genset/UPS compatabilty issue, there are two options that can be

 The first being the fitting of a suitable 250mH mains voltage rated
inductor (230v 2 to 4KVA case) across the genset's output (inductive
loading is not a voltage stability issue) or else, figure out a drop-in
AVR module design capable of bucking the self excitation effect from the
capacitive loading of the UPS(es). This last _might_ be a non-starter
but the first does definitely work.

 I have considered yet another means of addressing the issue. Since the
waveshape of the supply isn't particularly critical, fitting clipping
diodes across the genset's output to clip at 350v peak for a 230v supply
(and 180v peak for a 115/120v suply) might prove effective.

 In practice, a stepdown transformer to allow 'amplified zenner diodes'
operating at lower voltages/higher currents would be a more workable
solution. the transformer leakage inductance helping to take the sting
out of the current spikes in the clipping diodes as well as countering
the capacitive effect.

 That's three possible workarounds. Of the three, unless I _have_ missed
a fundamental flaw, the improved AVR module is the most elegant


Regards, John.

 Please remove the "ohggcyht" before replying.
The address has been munged to reject Spam-bots.

Posted by vaughn on April 11, 2009, 11:46 am

   The "Ferups" UPS that the OP is trying to use is actually sort of a third
type.  It uses a huge, lossy transformer with a capacitor to form a
brute-force tuned circuit which stores enough energy to carry the load
through a dropped cycle or two while the inverter has a chance to wake up
and do its job.  They are very inefficient.  (just feel the heat being lost
in that core).  I have broken several of them up for the scrapper.

I have a Sola computer power filter here that works on the same principle.
It wastes so much energy that on a cold evening I have been known to use it
as a foot warmer, but it normally sits cold and unused on the shelf.

I would not ever consider using a Ferups UPS in my home.   The last thing I
need is a 24/7 power vampire.


Posted by Jim Wilkins on April 11, 2009, 12:28 pm

The output of those isn't a pure sine wave, the peaks are rounded off.
I never put one on a spectrum analyzer so I don't know the harmonic
content but it's quite visible on a scope.

I wonder if a small motor load, like a fan with the blades removed,
would clean up the line adequately.

Posted by Johnny B Good on April 12, 2009, 12:44 am
  My newsreader has finally shown me my second posting but no sign of any
further replies :-(

 I notice, using my web browser to access groups.google, there were
actually quite a few replies (damn my killfile settings!), so, working
from the web browser display, I'll respond to each contributer by name:

Martin Riddle:

 The very first thing I thought might have been causing trouble was the
use of a floating generator supply but when I converted a mains filter
block to allow the earth and neutral of the genny to be tied to the
mains earth and neutral lines, I was rather dismayed to discover that it
had no effect whatsoever. That's not to say I won't be tying the earths
and neutrals when I complete the project (one way or another).

 54Hz is +8% which is outside the UPS's +/-5% tolerance range which is
why I adjusted the generator to 52Hz no load, dropping to 49Hz at 2500W
full continuous rated load (absolute max load being 2800W). I thought
I'd already covered this.

 What is an ISObar? I've googled but, apart from atmospheric pressure,
the only electrically related hits involve some sort of mcb distribution
panel which I can't see being any help in filtering the noise components
out of my generator supply.

 It's not a common mode filter problem.

 I've measured the winding inductances of a 400VA 120 +120 primary with
a 34-0-34 volt secondary. I estimated the 120v winding to be 187mH, the
full secondary to be 44mH with half secondary being 11mH. The only
usable winding for a filter inductor expected to carry up to 10A at 50Hz
is the 11mH one.

 Incidently, the split primary will still be 187mH when both windings
are paralelled in phase for 120v mains. It's only when they're wired in
series for 240v mains that the inductance will quadruple to 748mH (
giving a magnetization current of just over 1A on a 240v 50Hz supply)


Eeyore :

 I can't see how an ultra isolation transformer would help in this instance.


Tim Jackson :

 Your reply is the most comprehensive so far, so I've pasted it here:

 I agree, although it's possible to get a useful 12 db attenuation with
a 5 pole Chebyev 3db filter and 53Hz turnover, but as you pointed out,
the required inductance (and capacitance!) values start to go a little
over the top. :-(

 I'm also unsure whether the 75Hz component is causing the problem but
since this component actually increases on-load by 11db (from 36db down
to only 25db down on the fundamental) I suspect it might be. Also the
third harmonic is even worse at -29db no load rising to -12db full load.
The fifth harmonic seems to be just below the turnover point for the
response tilt with load, remaining close to -28db with the higher
products above the 7th reducing with applied load. The distortion
product spectrum reaches a peak around the 1600Hz mark at -26db no load,
which plummets to -50db on load.

 To be fair, the frequency plot for mains power is no oil painting
either (it just looks a little less ugly is all). For comparison, the
mains voltage odd harmonic levels are, starting from the third, -32,
-31, -48, -48, -59, -56, -52, and -61db for the 17th, thereafter
dropping away at about 12db per octave into the noise floor around
-100db at 7KHz.

 There doesn't appear to be any 75Hz component but there is a 100Hz one
at -64db. The only significant harmonics are the third and fifth some 30
db down on the fundamental. Since the UPS is quite happy with this
quality of mains voltage even on its most sensitive setting, I'm
guessing the problem is going to be the significantly higher third and
fifth harmonic levels in the generator's output. At just 12 down on the
fundamental, the third harmonic at full load would represent some 20% of
the total power coming out of the generator!

 The plots were obtained using a low voltage wallwart ac transformer via
a resistor attenuator network to feed the line in on a laptop using Cool
Edit Pro one mid afternoon last year.

 Thanks for the suggestions. I'll keep them in mind.

 Although the SmartUPS 2000 is of the mains pass through switching type,
it is a cut above those dinky 500VA types that rely on the limited
battery capacity to save their grossly undersized transformers from
burning out.

 The runtime chart for a range of loads shows 450 minutes for a 35 watt
load going down to 8 minutes for a 1400 watt load. That's a 7 1/2 hour
run time on the lightest load which suggests they've avoided the sin of
running the transformer cores into significant saturation (a major
source of overheating in the transformers of the cheaper SoHo UPSes).

 Although I've been tempted to consider uprating the battery pack from
its original 18AH 48v size to something like 100AH to extend the run
time, I decided the better option would be to add a standby petrol
generator instead. Mind you, I'm using four 25AH 12v dryfits along with
another eight 7AH dryfits as it is.

 The 7AH dryfits are spares for the smaller UPSes taking advantage of a
'Free Ride' to keep them conditioned (although it doesn't do any harm to
the run time of the big UPS in the event of an outage ;-). However, I
don't plan to grossly uprate the capacity, just compensating for loss of
capacity due to ageing.

 You do have a point but I'm considering a way to simplify the process
and avoid having to power cycle any running computers. A logic
controlled changeover relay would allow that without the risk of a lapse
of procedure resulting in the "Yanked Plug" syndrome.

 Well, what I'm really trying to do is bypass the failure of the UPS to
accept the quality of the generator supply as being sufficiently good
for the load it's supposedly protecting.

 I agree, except for the fact that if the UPS could have been programmed
to accept the generator power as being good enough, I'd have retained
the benefit of UPS protection in the event of the generator supply
failing for any reason. Bypassing the UPS in this fashion removes that
benefit. Once the UPS is shutdown, in the absence of mains input, I
don't think it can be brought back up fast enough to avoid a break in
supply, which is why I referred to it as a "Workaround" solution.

 I don't think so, TBH, I think it's more like the way I want my own
setup to work (except for the automatic generator startup and changeover
from mains to generator power).

 The sequence you were witnessing would most likely be:

 Incoming mains supply compromised, UPSes kick in immediately this was
sensed (that first flicker on the building's mains supply). The Diesel
Generator control unit would monitor the situation for a few seconds
before deciding it was time to invest in a genset startup to conserve
the UPS battery (perhaps 5 seconds?). Then, and only then, would a
startup be initiated. The controller would monitor the genset's vital
stats and confirm all was good before operating the transfer switch to
restore power to the building's incoming mains supply (possibly another
half minute or so for the runup phase) shortly after which there would
have been that second flicker when the UPS starting seeing power on its
mains input and had synchronised its output sufficiently to minimise the
switchthrough transient.

 If the UPS was of the type that converted the input power to dc to
float charge and power an always on inverter, you'd have never witnessed
any such flicker events. However, you may well have heard (or even felt)
the diesel genset startup in response to the outage.



 I believe I am the OP (I know I'm sure I am ;-). I never mentioned a
Ferrups (although I do have a small example of one stowed away in my
basement), but I know what you mean by "Power Vampire" ;-) However, most
UPSes take a surprisingly large amount of 'maintaining' power (a figure
of demerit the manufacturers are reluctant to advertise,,, anywhere).

 For example, did you know that an APC SmartUPS 700 demands 20W without
load on a fully charged battery (or with battery pack removed, even!)
yet a SmartUPS 2000 with three times the load capacity only demands some
30 to 34 watts (still a lot but about half what you'd need for the three
SmartUPS 700s required to do the job of one SmartUPS 2000).

 The best UPS in this regard (maintaining power per VA of protection)
that I have is, curiously, another APC example. This is the Backups500.
It only takes 3W for its 500VA's worth of protection. I have this
protecting my FreeNAS box against loss of power on the protected mains
supply (the one protected by the SmartUPS2000, - the very same one that
I can't persuade to play nicely with my chosen standby power generator).


Jim Wilkins:

 I'm afraid the answer to your question is no.


 Ok everyone, and thanks for your contributions. I've just refreshed the
googlegroups page and  can see that my repeat posting has appeared with
no further contributory postings so it's back to normal News Reader
service for me.

Regards, John.

 Please remove the "ohggcyht" before replying.
The address has been munged to reject Spam-bots.

Posted by Johnny B Good on April 23, 2009, 2:51 am
 Update on generator filter testing:

Well, I've been busy making up a simple "3rd Order LPF" on a breadboard
using the Pi configuration (my .5 amp/100mH to 5 amp/50mH choke with all
ten of those 30 microfarad 600v caps- 5 each side of the choke).
Luckily, I thought to test from a 15 volt AC supply before simply
connecting it up to the genny supply or I'd have blasted the output caps
to bits with voltage magnification!

 The output voltage behaviour of such filters[1] under load impedances
that depart from the design impedance can be quite wild, making them
totally unsuitable as a solution for this particular problem. However,
this prompted me to take a closer look at the generator itself to
determine exactly what class it was (there are usually three types, a
squirrel cage induction motor with phase excitation capacitor, a
brushless alternator (effectively two alternators back to back on the
same shaft - an exciter feeding the rotor field of the main alternator)
and the more conventional slip ring fed rotating field type). My example
is of the latter type with an AVR module (meaning no rotating capacitor
and diode to worry about;-).

 I've been assuming that the 1500Hz noise components were due to an
artifact of the AVR circuit (wherever and however that was implemented)
but testing by substituting the AVR supplied field current source with a
battery source reveals this is not the culprit at all. Interestingly,
the recorded voltage waveform shows the same no load / on load frequency
plot characteristics (aside from a lack of voltage regulation on load).

 Under no load, the voltage waveform does appear to be a sinewave but
with horrendous levels of 1500Hz superimposed (presumably a cogging
artifact from the multislotted ironwork of the stator designed to
achieve a sinusoidal output from the winding pattern on the stator).
When loaded with the 21 ohms of the kettle heating element, the 1500Hz
component almost completely vanishes but the waveform departs radically
from a sinewave shape.

 A quick calculation suggests that a 5 microfarad capacitor could
provide the 20 ohm loading at 1500Hz and suppress the generator's
cogging products. I'm thinking of fitting 30 microfarad caps across each
of the two 115v stator windings (effectively 15 microfarads across the
230v output). I've, therefore, rewired my Pi network filter so that it's
now merely two of the ten 30 microfarad caps in series across the supply
so I can test its effect. The imaginary current loading this creates at
the funadamental frequency (50Hz) is about 8% of the generator's maximum
so shouldn't add any significantly adverse loading on the generator.

 It's quite possible that the 1500Hz noise component is contributing
significantly to the UPS mains quality sensitivity problem (but I don't
expect it to be the whole problem), so this may go a good part of the
way to resolving this issue.

 The other main unwanted component is the relatively high level of third
harmonic (150Hz) and I think my best strategy here is to use a simple
parallel tuned circuit in series with the output to notch this out. This
has the merit of avoiding voltage magnification issues at the supply
frequency and predictable voltage levels in the 150Hz parallel tuned
circuit (around 60v). The 5th, 7th and higher odd order components are
around the -30db mark or lower so don't seem to be a contributor (the
capacitors should offer some further reduction of these components
anyway leaving me to concentrate on that troublesome 3rd harmonic).

 As for the 25Hz sideband components, that's almost certainly a
consequence of the prime mover's characteristic rotational behaviour due
to it being a single cylinder four stroke petroleum (gasoline) engine
(rotational acceleration every other revolution on each power stroke
with associated deceleration on the exhaust, induction and compression

 A nice heavy flywheel would undoubtedly mitigate this effect (or,
better still, a parallel twin cylinder prime mover of similar capacity -
200cc in this case, supposedly a 6.5HP engine). However, there's not
much, if anything, that I can do about this so fingers crossed that the
25Hz sidebands aren't significant to my current problem.

 I've got the test filter ready to go, but since it's now 02:42 BST,
I'll have to 'sleep on it' and do my next bit of testing later this
morning or the early afternoon (assuming the weather remains dry - not a
given in the North West of the UK).

 I'm hoping that the UPS will at least accept the filtered supply under
no load conditions and not keep cycling between 'mains' and battery as
it did when I last tried this test several months back. If it accepts
the filtered supply, then the 150Hz notch filter may solve the on-load
issue and be worth pursuing.

 At least an air cored 3.763mH inductor is a do-able proposition - but
I'll be 60 microfarads shy of the 300 required to resonate the inductor
at 150Hz but a few more turns will bring the inductor up to the 4.7mH
required to resonate 240 microfarads at 150Hz. I'd prefer a larger C to
L ratio to reduce the volt drop at 50 Hz but the 4.7mH test filter
should provide 'proof of concept' before investing any further effort in
tracking down another supply of capacitors.

 With a bit of luck, I might just have found a solution to my dilemma
(but I won't be getting my hopes up - this just might be one of those
'intractable' problems wherein the real solution lies in getting the UPS
to a fully functioning state as far as programming it to accept 'really
bad quality' power - or else find a similar unit that _will_ work under
these conditions).

 I'll report back on my results sometime in the next day or so.

[1]  It's one thing to plug various parameters into the on-line filter
calculator and admire the resulting frequency plots but another, totally
different experience, when working so close to the wanted frequency when
the filter output is left open circuit and fed by a voltage source of
close to zero impedance.

 Since there's no way to guarantee the loading impedance placed on such
a mains voltage filter, this represents a really serious drawback to
their practical use, so much so that I've had to totally abandon the LPF
idea and restrict the filtering to a simple capacitive shunt on the
generator output windings to suppress the high frequency noise
components and, if necessary, augment this with a less contentious
series notch filter for the rather troubling levels of third harmonic
which appear at full loading.

 Adding, if deemed necessary, additional series notch filters for the
5th and 7th harmonics would be a trivial exercise by comparison but the
25Hz sideband components may well turn out to be the real showstopper
and, for this, I don't have a 'practical solution'.

 If it comes down to this, my only option would be to replace the
offending UPS (an alternative petrol genset in a similar price bracket,
i.e affordable, seems doomed to suffer the same shortcomings).

Regards, John.

 Please remove the "ohggcyht" before replying.
The address has been munged to reject Spam-bots.

This Thread
Bookmark this thread:
  • Subject
  • Author
  • Date
please rate this thread