Reference Manual
748384-C
September 2003
Rosemount Analytical Inc. A Division of Emerson Process Management
Introduction 1-19
NGA2000 Reference
Normalize the readings such that fullscale
becomes equal to unity (divide the readings
by the fullscale value). If your highest value
gas is less than 100% of the fullscale, you
must either change the fall scale value to
that of the span gas or get a gas of the
fullscale value. Trying to extrapolate to
fullscale is not normally practical, and re-
sults in significant errors.
Using a least squares fit, calculate the poly-
nomial coefficients. You may want to
weight the errors so as to make the fit more
accurate at the lower readings by making
the error weight inversely proportional to
that point's percent of range.
c. Details
The polynomial coefficients are stored in a
series of array variables called LINAO_[l..4],
LINA1_[I..4], LINA2_[1..4], LINA3_[1..4],
LINIA4_ [1..4]. These are the four zeroth
coefficients (element 1 corresponding to the
zeroth coefficient for range 1, and so on).
They may be found on the linearization co-
efficient screens, where they can be edited,
but it is usually more convenient to use a
PC to download them into the analyzer.
Since the analyzer normalizes its signal,
such that 0 corresponds to the range low
end and 1 corresponds to the range high
end, the linearization curve is constrained to
go through the points 0,0 and 1,1, meaning
that the sum of the coefficients should be
1.0. In fact, because of the allowable er-
rors, the sum must be between 0.98 and
1.02 if a 2% error is allowed.
The curve must be monotonic. If it isn't, the
zero or span algorithm may fail, resulting in
an offset to the zero or span point. This is
equivalent to saying that the first differential
of the polynomial has to be positive over the
entire –0.05 to 1.1 range.
The linearization should be performed at a
specific temperature, and this has to be the
same as the base temperature for the tem-
perature compensation algorithm. Nor-
mally, 25
°
C is chosen for this, and minor
errors will have little effect.
If it is desired to achieve accuracies of 2%
of point, and 1 % of range, whichever is less
(above 10% of range), fig s which are quite
practical in most cases, great care must be
taken to verify that the dilution system em-
ployed, and the settling time allowed, are
adequate. Errors of 0.1% of fullscale at the
zero end will make it impossible to achieve
this kind of result. You may have to wait
several minutes for the zero reading to "get
there", once you have put in a span gas.
It is essential that the gases used cover the
entirety of the range. Don't try to extrapo-
late from an 80% gas to the 100% point, an
error in your extrapolation will throw off the
entire curve.
Network variables: LINA0_, LINA1_,
LINA2_, LINA3_, LINA4_, LINSTAT, LIN-
FORRANGE, LINRNGHI, LIN_OVER,
LIN_UNDER
d. Troubleshooting
I use the self linearization screens, but
nothing happens…
Version 2.2 software does not support self-
linearization. You have to enter gases, take
readings and figure out the coefficients ex-
ternally.
I did a linearization, but my results don't
check out with reference bottles…
Verify that you used a 100% gas, that the
linearization ran-e matches or exceeds the
fullscale range, that you have enabled the
linearization at all and that you are using the
correct set of coefficients (all of these are in
the linearization setup screens under Expert
controls). Verify that you have entered the
coefficients correctly, and that their sum is
close to 1.0.
If all of the above is OK, there is probably
an error in your gas readings. Verify that
the analyzer is correctly zeroed and
spanned on the range you are using, and
that you are using a long enough stabiliza-
tion time. You may have to increase the
analyzer's t90 time to get better signal to
noise ratio.
