Example 4
Correlation
If you want to see whether a correlation exist between two signals, you can use the SIMEQ function corr10:
>> corr10('Leading Tag' , 'Trailing Tag' , 'From' , 'To', DelaySeconds)
Suppose you do not know the transport time in
the sample lines for the preheater top gasanalyser.
The command
>> corr10('W8RPipePress','W8_S_O2Eval',now-1/24,now,300)
will calculate the correlation between the
riser pipe pressure drop and the O2 measurement in the preheater using
data from the last hour from now. The correlation coefficient is
calculated and plotted at each 10 sec between 0 and
the specified DelaySeconds and displayed as a function of the delay.
As can be seen, a strong correlation exist,
and the O2 signal is opposite proportional to the riser pipe pressure
with a delay of 90 seconds.
Since the gas retention time in the preheater
is only a few seconds, most of this delay is due to the transport time
in the sample pipes. In this kiln we know that the delay is 60 sec when
the sample flow through the analyzer is at
the nominal value of 60 l/h.
So we can use the measurement to check that during this hour the sample
flow has been 60l/h*60sec/90sec=40l/sec - somebody should make a
scientific expedition and adjust it.
The corresponding trend plot of the same hour
is shown below. It is not easy to see the correct delay.
As usual, correlations should be treated with
respect. The existence of a high correlation does not mean a
cause-effect
relationship exists.
The number of storks in Denmark and the
number of childbirths are highly correlated, but probably not an
expression of
a cause-effect relationship.
Mostly the correlation can be used to
estimate how much a given signal influences another if you already have
a
theoretical idea that correlation exists.