Re: Need Advice on a Simple Stat Question

"wki1234@xxxxxxxxxxx" <wki1234@xxxxxxxxxxx> wrote in news:09a6e82f-
740f-4be8-9058-58d178d1cb79@xxxxxxxxxxxxxxxxxxxxxxxxxx:

Thanks to all for your replies. I'll be a provide a little more
detail on the data, which is coming from a meter monitoring
electrical variables on a 3 phase AC electric arc furnace.

The 3 phases each provide ~33% of the power input. It is a violent
and sometimes unstable process, but the daily mean over 2500 data
points is typically 33 and the within day variance is 4. The
between day variance is usually 0.6.

Is that because someone fixes any imbalances at the beginning of each
day?

The balance between phases can shift for various reasons, but once
this occurs, the balance will not normally return to 33 without
human intervention. I want to detect if a shift has occurred. I
chose the most recent 2 days rather than 1 to try and avoid false
warnings.

If the balance of one phase shifts then one or both of the others
also will shift, since they are dependent on each other.

For each phase, my VB program is currently calculating upper and
lower limits for a 2 day average from the daily means and sigma
between days as avg +- 2 * s/sqrt(2) for 95% confidence. Is this
correct? Would a t-test or some other test be better?

Maybe I should only issue a warning if 2 of three phases are outside
the limits?


You want a test that give a "signal" when there has been a level shift
or a significant trend away from the desired level. And you have two
independent variables: line1 and line2, since:
line3 = 1.00-(line1+line2)

You should advise us why this test is useful, because the sensible way
to to design a procedure is to appropriately balance the relative costs
of a false positive vs. a false negative. It is not clear to me why you
are willing to wait to the end of the day, but perhaps your answer to
the "why" question will make that more clear.

--
David Winsemius
.

Relevant Pages

  • Re: Need Advice on a Simple Stat Question
    ... The balance between phases can shift for various reasons, ... I want to detect if a shift has occurred. ... most recent 2 days rather than 1 to try and avoid false warnings. ... limits for a 2 day average from the daily means and sigma between days ...
    (sci.stat.math)
  • Re: Play Want Bin, You Knows It
    ... traditionally brought around a shift in the balance of power. ... that mean Nintendo's due for a rise back to the top? ...
    (uk.games.video.misc)
  • Re: Bicycling with a Backpack
    ... belt, it should be okay, what you don't want to have happen is for the load to be able to shift side to side, as that could cause balance issues. ... the handlebars with or without backpack. ...
    (rec.bicycles.misc)
  • Re: agile development and frameworks
    ... That reduces the risk of getting a solution that is worse than the known ... finding that balance isn't trivial. ... One way to shift the balance is to find ways to make experimentation more ... include spikes (controlled, time boxed experiments), having Slack, keeping ...
    (comp.object)
  • Re: agile development and frameworks
    ... >> That reduces the risk of getting a solution that is worse than the ... >> be balanced, and I think finding that balance isn't trivial. ... One way to shift the balance is to find ways to make ... >> to do that include spikes (controlled, time boxed experiments), ...
    (comp.object)