## Featured Articles

Check out the latest featured articles.

## New Article

Product Viscosity vs. Shear

## Featured File

Vertical Tank Selection

## New Blog Entry

Low Flow in Pipes- posted in Ankur's blog

2

# Data Plant Stability Analysis

data stability plant analysys

5 replies to this topic
|

### #1 A_D_M_MII

A_D_M_MII

Junior Member

• Members
• 22 posts

Posted 21 July 2021 - 12:56 PM

Hi everyone,

I am doing a university project and I have a question:

I have a lot of plant data with the measurements of the gauges and indicators of a plant, these data covers a big period of time (4 days), and I have to choose the most stable period of 5 hours in order to simulate it.

Visual inspection by graphics is hard work because there are a lot of gauges and it will also include my human error.

I have thought about using a statistic tool like the variation coefficient in excel, but i don't know if it is mathematically correct.

I would like to be mathematically correct because it is a university project and i would like to know how to justify every step i do.

Have you any idea about that?

Thank you all of you!

### #2 latexman

latexman

Gold Member

• 1,166 posts

Posted 21 July 2021 - 01:06 PM

Can you calculate a running 5-hour standard deviation against a running 5-hour average?  That sounds like it may be useful.  Remember, MS Excel is your friend!

### #3 Pilesar

Pilesar

Gold Member

• Members
• 795 posts

Posted 21 July 2021 - 01:09 PM

In my opinion, visual inspection will work best for you. Steady state is subjective. Try to pick important parameters like feed characteristics (feed flow, conditions, feed composition) which are 'must haves' for steady state. It will take some time for the steady state to propagate downstream, so take the last five hours of the steady feed period. Statistical tools take more effort than you expect and can sometimes be misleading. Statistical calculation results should be verified and are not to be trusted if they contradict visual inspection of the data. So why not just go to the real authority in the first place and bypass the statistical work? My answer might be different if you were trying to automate the procedure for on-line application, but this is just a single data set and does not require automated results.

### #4 A_D_M_MII

A_D_M_MII

Junior Member

• Members
• 22 posts

Posted 21 July 2021 - 01:19 PM

Can you calculate a running 5-hour standard deviation against a running 5-hour average?  That sounds like it may be useful.  Remember, MS Excel is your friend!

Yes, i have done it, and i have the variation coefficient (Standar deviation/average) by consecutive sets of 5-hour, so, in the next step i guess that i have to choose the period of 5 hour with the lowest coefficient for each gauge, but it is also a "handwork", i don't know if there is other way wich allows me choose the best option by maths (I have 150 gauges approximately)

### #5 A_D_M_MII

A_D_M_MII

Junior Member

• Members
• 22 posts

Posted 21 July 2021 - 01:24 PM

In my opinion, visual inspection will work best for you. Steady state is subjective. Try to pick important parameters like feed characteristics (feed flow, conditions, feed composition) which are 'must haves' for steady state. It will take some time for the steady state to propagate downstream, so take the last five hours of the steady feed period. Statistical tools take more effort than you expect and can sometimes be misleading. Statistical calculation results should be verified and are not to be trusted if they contradict visual inspection of the data. So why not just go to the real authority in the first place and bypass the statistical work? My answer might be different if you were trying to automate the procedure for on-line application, but this is just a single data set and does not require automated results.

Yes, I know what you mean, and i have already done it, but by visual inspection of the most important parameters i have about 4-5 candidates, so, in order to justify why i have choosen one of them i would like to use a stadistic method for the other parameters.

Thank you

### #6 Pilesar

Pilesar

Gold Member

• Members
• 795 posts

Posted 21 July 2021 - 04:04 PM

It may be that your candidates are equivalently acceptable. Consider choosing two steady state periods. If you tune the simulation with one data period, then the other steady state period should come pretty close to matching also. If the other set of data confirms the model, then that would seem to be the proof you seek. This would work best if there were significant differences in data between the two steady state periods. Your simulation model will likely require material balance adjustments to your data anyway and there is where you will need to spend the analysis effort.