On Feb 17, 6:25*pm, "Phil Allison" wrote:
"Robert Macy" = *one stubborn ****er

A cap does improve PF...

** But only with theoretical transformers - *NOT real ones.

First, use a Linear Model to represent the isolation transformer.

** Waste of ****ing time and effort *- *as it only repeats the same ****WIT
error you have been sprouting here all along.

The primaries of *REAL *commercially made E-core transformers are *NOT
linear inductors !!!!!!!!!!!

The off load primary current at rated voltage is *dominated by the third
harmonic * of the AC supply frequency.

The laminated iron core is then saturating, quite heavily.

JW's 1kVA iso tranny is a very typical example of this fact.

You will *NOT *find this information on webs sites that merely discuss
transformer basics.

You WILL *find this if you test a cross section of commercial E-core
transformers with the aid of a variac, RMS current meter and a scope
monitoring the current waveform.

This has NOTHING do with badly or well made transformers - *all makers do it
to save weight and cost.

BTW:

I happen to own a 1kVA transformer very similar to that described by the OP.

Tested as above, these are the figures:

VAC * A rms * *I peak

30 * * * 0.08 * * * 0.11
50 * * * 0.11 * * * 0.14
70 * * * 0.20 * * * 0.35

90 * * * 0.45 * * * 1.0
110 * * 1.0 * * * * 2.0
120 * * 1.4 * * * * 2.9
130 * * 2.2 * * * * 4.6

Up to 70 volts AC, the tranny is approximately linear with an effective
inductance of about 1.1 H.

At and above 90 volts AC it suddenly changes - *current starts to increase
exponentially and the wave becomes very peaky with a 1:2 ratio between rms
and peak values.

At 130 VAC input, effective primary inductance ( based on simplistic
calculations) *has dropped to less than 0.2H due to core saturation.

I must have tested hundreds of E-core trannys this way in the last 20 years
or so and ALL do much the same thing.

Toroidal and C- core types are different.

.... *Phil

For various reasons, it took MUCH longer than I anticipated to post
back here!

After more accurately modeling a REAL transformer based upon PA's
data, I am convinced that adding a cap in parallel will NOT improve
the PF.

The best intuitive way to explain is to simply say: in order to reduce
PF, a resonating cap is added in parallel to cancel the effect of the
inductance. The effect relies upon the inductance to be a CONSTANT
value, the core of a typical isolation transfomer during its operation
is NOT constant {reducing dramatically as the peak current flows. With
this changing value of inductance all the 'goodness' of adding the cap
completely disappears. Using the following simple model, I could not
even change the cap to some 'optimum' value. PF just stayed bad, did
not get worse, just stayed bad, and the cost of adding any cap was
wasted. .

Small discussion about the dataset. The slight increase in apparent
inductance going from 30 to 50 can be explained as being due to the
coercivity of the core material. At low currents, the BH curve loop
being followed is more horizontal than the BH curve being followed as
the voltage increases and current increases, the inductance then
starts dropping due to the saturation of the core. If you plot the
data set as Apk/(sqrt(2)*Arms) vs Vac, you'll see a strange shape to
the curve. plot as 20*log10(Apk/Apkcalc) vs Vac and it is VERY
interesting to notice a 'step' and then constant value. see the
undershoot, overshoot, and ringing of the data as Vac increases. Note:
The following model does NOT display this type of performance. I
wonder if it was caused by a 'two-step' saturation? In other words,
material saturated leaving another material that saturates at a higher
current value, like regions in the core?. I'm going to go back and
try 3 inductors in series, to see if I can get a 'better' fit - air
core inductor, inductor 1 saturating first, inductor 2 saturating 2nd.
Only mentioned as interesting, do not think a finer model will result
in a different conclusion, though.]

LTspice has a simple nonlinear model, called "Behavioural Model", for
an inductor. The inductor's flux is: Flux = tanh(x), where x is the
current through the inductor.

The model is supposed to follow the saturation curve fairly well, but
assumes ZERO coercivity, in other words, zero hysteresis. The model is
like following the 'centerline' of the hysteresis curve.

Several observations, I could NEVER get the model to fit the data
provided by Phil Allison, [which translates to still don't have a good
model]. However, after some 'adjustments'. the characteristics of the
model did fit the characteristics of the data [No time to EXACTLY
create/present the tables of comparison, will do later]

Suffice to say, as the input voltage increased; the rms current went
up faster than if the transformer's core were a constant inductor, the
peak current for the rms current went up at about the same ratio as
PA's data. At higher voltage, the peak current noticeably distorted
the current waveform into appearing to have severe 3rd harmonic
distortion. Actually, instead of sinusoidal, looks triangular.

Using this model, I calculated an appropriate cap, added it to the
circuit, and found NO EFFECT on PF !!!

I then started changing the cap's value, looking for some optimum, and
found none.

CONCLUSION: For "real" isolation transformers, it is NOT POSSIBLE to
add a cap to 'adjust' the PF for the load.

I defer to PA's experience with a multitude of manufactured
transformers, in defense of my comments, my experience was limited to
custom transformers [whose performance was always better than
commercially available] and my own transformers, which perform a bit
better. Example, 100 turns to get 1 Henry. No DC current is allowed,
but you do measure 1H inductance with only 100 turns. Coercivity is
about 1/100th silicon steel, which my understanding is usually used in
commercial transformers. So what I'm used to working with does
approach 'ideal' inductance.


The simple model is here for any interested:
name the file something ending with .cir LTspice will run the
simulation. You're on your own for changing values.
Note the Rcore value was added to represent the unloaded 40+W
dissipation, you will find its absence/presence does not affect the
conclusions.

TEST_ModelBehaviour - nonlinear inductor using behavioural Model
* for use on LTspice
*
..tran .1 20 19.95 .1m
..param k0=sqrt(2)
..param k1=120
Vac 1 0 AC={k1} SIN(0 {k0*k1} 60)
Racmains 1 IN 0.01
*Cc IN 0 30uF
Rpri IN 3 0.26
..param kk0=1
..param kk1=1.2/{kk0}
..param kk2=0.412
Lcore 3 0 Flux={kk2*kk1}*tanh(x/{kk2})
Rcore 3 0 350
..end