Eng-Tips is the largest engineering community on the Internet

Intelligent Work Forums for Engineering Professionals

Register Log in
  • Congratulations waross on being selected by the Eng-Tips community for having the most helpful posts in the forums last week. Way to Go!

Multivariate Analysis?

Status
Not open for further replies.
ManTicora

ManTicora

Chemical
Jan 5, 2006
25
CA
I have have a flow of reagent (Methanol) entering 21 process users across my plant. Some users uses the Methanol as fuel others as solvent and some as reagents...
Could one suggest the best way of knowing/classifying these users are causing High methanol usage during a given period?
I'm thinking of using PCA but not sure if that is over-killing the problem... A simple Pareto Chart maybe?
Thank you for your suggestions.
 
Replies continue below

Recommended for you

Sort by date Sort by votes
As opposed to just measuring the usage?

TTFN
faq731-376
7ofakss

Need help writing a question or understanding a reply? forum1529


Of course I can. I can do anything. I can do absolutely anything. I'm an expert!
There is a homework forum hosted by engineering.com:
 
Upvote 0 Downvote
IRStuff " Just measuring the usage" gives discrete flow rates but I'm not sure how "consumption pattern" can be revealed by time series of flow rates.
Thanks
 
Upvote 0 Downvote
??The integral of flow rate is the consumption, is it not?

TTFN
faq731-376
7ofakss

Need help writing a question or understanding a reply? forum1529


Of course I can. I can do anything. I can do absolutely anything. I'm an expert!
There is a homework forum hosted by engineering.com:
 
Upvote 0 Downvote
Integrating the flow will not reveal the frequency at which a given user is affecting the overall flow!?
At best total ("integral") flows per users will help build a Pareto Chart which was mentioned in the Original Post.
Is PCA the best way of classifying user while determining the pattern at which users are affecting the overall methanol consumption?
Thank you for your hints.
 
Upvote 0 Downvote
I'm stupid, because I thought you understood that I meant that you measure the 21 individual rates. Otherwise, you need to find out if each process has measurements, which they probably do, or you have to camp out in each process and see what they're doing at any given time.

Math is not going to tell you which of 21 aggregated users is using what.

TTFN
faq731-376
7ofakss

Need help writing a question or understanding a reply? forum1529


Of course I can. I can do anything. I can do absolutely anything. I'm an expert!
There is a homework forum hosted by engineering.com:
 
Upvote 0 Downvote
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
@IRstuff (Aerospace) wordings tells a lot about I.Q. but we're digressing here...
Let's bring the discussing back on track...
+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++

Can anyone suggest means to monitor users consumption pattern in a network of 21 users.
The goal is to know at a given time period, which user is inducing more variability in plant total Methanol consumption.
Trends of flow rates to these 21 users is available. But it's not trivial to screen 21 trends...
Could Principal Component analysis help here or just using a Pareto Plot? Or any other insightful methods?
Regards,
 
Upvote 0 Downvote
"Trends of flow rates to these 21 users is available."

Huh? I thought you didn't have measurements? If you have the measurements, you have the consumption.

TTFN
faq731-376
7ofakss

Need help writing a question or understanding a reply? forum1529


Of course I can. I can do anything. I can do absolutely anything. I'm an expert!
There is a homework forum hosted by engineering.com:
 
Upvote 0 Downvote
Trends are data and consumption patterns are "information".
A sample of "real life" trend is attached. That's what we, Engineers, are tuning into executive information...
Once again "If you have the measurements, you "DON'T" have the consumption."
 
 http://files.engineering.com/getfile.aspx?folder=a68ace53-a053-4495-9c54-dd4627ebf967&file=trends.png
Upvote 0 Downvote
OK, fine, whatever. Not worth my time to sort this out for you.

TTFN
faq731-376
7ofakss

Need help writing a question or understanding a reply? forum1529


Of course I can. I can do anything. I can do absolutely anything. I'm an expert!
There is a homework forum hosted by engineering.com:
 
Upvote 0 Downvote
@GregLocock (Automotive) You will have fun practicing your talent for aggressiveness at your home. As stated earlier wording hints about I.Q. and emotional maturity

This post is meant to enhance the knowledge sharing of this forum.

Here is the problem statement:
As shown on attached trends(above), I have a set of raw process data of which I would like to extract information.
For instance, identify among the 21 users which one is inducing more variability in the main flow to the methanol header.
I would like a peer "ENGINEER" with knowledge in statistics/data mining to suggest which approach will be appropriate.
ANOVA, PCA, FA or Pareto Plot?
Enjoy you evening!
 
Upvote 0 Downvote
Well MantiCora I hope your little hissy fit helped you, it sure as hell amused me, and seems to have annoyed the only person trying to help you directly.

So, what problem are you actually trying to solve? Until you know this then you are lost.

One you know that the following questions may be applicable.

What measure of variability is acceptable to you?

Are you only interested in periods where total demand is close to maximum? How close?

What time period do you want to average over?

Are you really only interested in the variability of each users demand, or are you also interested in the average demand of each user in the time period of interest?

Can you assume that each user's requirements are independent of the others?











Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376
 
Upvote 0 Downvote
I'd ASSume that the peak flow is of primary interest, as this is when the maximum variation in true flow rates (due to manifold losses) would occur. But, then, it's not yet stated what the problem is, as Greg has pointed out. State the problem more clearly, please, so that we could help you.

One major issue with ANOVA is that it really isn't suited for nonlinear processes like flow systems. IMO.

You might think about getting pressure independent flow control valves (a company called FCI makes a valve that would work for methanol) for the users, to seperate their flow rate from what appears to be a variable system pressure. But then, again, I'm guessing as to what the real problem is.
 
Upvote 0 Downvote
In an idle few minutes while watching a gruesome but predictable horror movie I had a bit of a re-read of this equally disastrous thread and I think I may have a problem statement:

Variable demands by each user result in a variable total flow required, averaged over a time period T. We'd like to know (1) which users are contributing most to the variability, and (2) if we can explain the peak demands by looking at a subset of the users.

Peak demand is defined as a total flow rate greater than P.



For T I'd guess using something like total system capacity/P/10 as a first guess

So, as IRStuff suggested right at the top, integrate each user flow over period T. Call these U1 to U21, and the total flow F= sum(U1..U21). This gives less data to handle and eliminates possibly spurious high frequency peaks.

Then examine the data (essential step in any statistics). In this case I suggest plotting a scatter plot for each U versus F. At that point hopefully some relationships will stand out. If they don't, no amount of statistical fiddling will produce a reliable model.

(1) is pretty easy, just the standard deviation or 90%ile range for each U
(2) will depend on what you've seen in the scatter plots.





Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376
 
Upvote 0 Downvote
Thanks Gentlemen for your kind contribution.
@GregLocock (Automotive) Your problem statement is exact and quite elegant as you uses fewer word! Thanks.
(1) is there a way of coupling the standard deviation with the flow rate... A kind of "weighted" standard deviation?
(2) I will share the findings tomorrow after implementing these scatter plots ( U versus F). But I anticipate noisy and non-linear plots.
@btrueblood (Mechanical) Interesting to know that ANOVA doesn't suit nonlinear processes. Could you please give a reference of the FCI valve you mentioned?
Cheers :)
 
Upvote 0 Downvote
(1) is there a way of coupling the standard deviation with the flow rate... A kind of "weighted" standard deviation?

Yes, divide SD by mean...


It is not as useful as you might hoped for your purposes where the main issue is the sum of all the users, not the relative instability of any one process.

Cheers

Greg Locock


New here? Try reading these, they might help FAQ731-376
 
Upvote 0 Downvote
Status
Not open for further replies.

Similar threads

henriquedpires
  • Locked
  • Question
Problems with Inventor Nastran (non-linear analysis)
Replies
4
Views
44
henriquedpires
henriquedpires
GregLocock
  • Locked
  • Helpful tip
Analysis method for non stationary time history signals EMD or Hilbert Huang Transform 1
Replies
2
Views
11
GregLocock
GregLocock
Indy
  • Locked
  • Question
Fault tree analysis
Replies
0
Views
9
Indy
Indy
davet1948
  • Locked
  • Question
Right Censored Data Analysis in MS Excel
Replies
0
Views
13
davet1948
davet1948
Rentik
  • Locked
  • Helpful tip
Simplified FMEA Approach vs. MTA (Maintenance Task Analysis)
Replies
0
Views
24
Rentik
Rentik

Part and Inventory Search

Sponsor

Back
Top