Stastistical Techniques in Chemical Industry

A

Atul Khandekar

Can someone provide any pointers to what statistical techniques are used in chemical / pharma industries for both batch and continuous processing? Is there anything different?

Thanx
 

Marc

Fully vaccinated are you?
Leader
There is a difference but I admit I cannot explain details.

I remember back about 5 years ago this came up with a client. There was a difference in how they chose samples among othe things. Sorry I can't help personally. Maybe a CQE here can help.
 
V

venkat

I was with pharma industry for a brief period.
In batch process each batch is identified bya separate tag and identified at the end in terms of yield, purity and overall quality of the product.
In case of continuous process the output of one step is the input for another and as such samples are withdrawn and checked and sampling technique is used based on the quantity of the material involved.
 
A

Atul Khandekar

Bringing the topic up again......
Any specific pointers / comments anyone?

Thanx.
 
D

Darius

Control Charting on Contiuos Process

In may experience, the problem with chemical continuos processing and control charts,:confused: is that some of the references recommend the use of individual and moving range, but something is left out, the autocorrelation between samples.
Some sites reccomend the use of other charts like EWMA or CUSUM, I tink they are excelent tools, but the objetive of them is to detect small changes in the mean (likely to be important in farmaceutical but not in chemical industries in general). :frust:
In Advanced topics on SPC from Donald Wheeler and some articles published on statistics say something about the autocorrelation coefficient, but there is not much information about, I used it and works fine for me.
There are some cases that even this coefficient does not work, when the quality is not affected by changes between specification limits, in that cases, some operators work at one setup, others on on a different ones, so in that cases precontrol charts work fine. :truce: I tink, the use of precontrol is against many SPC preformers, but it´s my belief, I hope this may help.
 
A

Atul Khandekar

Re: Control Charting on Contiuos Process

Darius said:
In Advanced topics on SPC from Donald Wheeler and some articles published on statistics say something about the autocorrelation coefficient, but there is not much information about, I used it and works fine for me.

There are some cases that even this coefficient does not work, when the quality is not affected by changes between specification limits, in that cases, some operators work at one setup, others on on a different ones, so in that cases precontrol charts work fine. :truce: I tink, the use of precontrol is against many SPC preformers, but it´s my belief, I hope this may help.
Darius,
Could you please elaborate / provide some website links where I can read more about this. Any examples of use of pre-control?
Thanx,
_Atul.
 
D

Darius

About autocorrelated process, there is just a few references

(broken link removed)

and a book

"Advanced topics in Statistical Process Control", the power of Shewhart's Charts by Donald J. Wheeler. SPC press, 1995
Chapter 12, Control Charts for Autocorrelated Data

It's a shame that Wheeler wrote many articles in quality magazine (that you can see on the net), but any about the subject.

In the book, he said that the autocorrelation coefficient does not affect when there is low autocorrelation, so, it's safe to use it anytime, and such is my experience.

:bigwave:

PreControl has it's uses, as Steiner said,

"in the final paragraph on control charts of Ishikawa ’s (1982)famous guide,where he wrote “Control charts are easy to construct so are widely used.But there are surprisingly few really useful charts ”

(broken link removed)

But one example that came into my mind is: when the process is drifting (going up or down) very slowly, of course SPC rules could detect it but as a run up or down, but in a chemical process the run up or down could not be a special cause, because most of the time is what is expected.

What, if you can not change anything (reduce variation of the variable), but you want to keep an eye upon the behaviur of the variable, to understand the relationship between variables, or to know what to expect of the process if a parameter is changing.

Darius

:smokin:
 
Last edited by a moderator:
D

Darius

About autocorrelated process, I like to explain a little more:

There are two ways is wich autocorrelation affects periodically collected data, excessive autocorrelation will have visible impact upon:

1. the running record (bu is something that most of the times happen in a continuos process without being a special cause condition)
2. the control limits calculated according to the usual formulas, the "contimated limits" will be narrower than they should be in order to properly characterize the process.

Calculation:

Autocorrelation_Coefficient=1/(1-r^2) ^0.5

being r the sample correlation coefficient, calculated using

v(1),v(2)...v(n)

x(1)=v(1), y(1)=v(2)
x(2)=v(2), y(2)=v(3), and so on


Sigma = Range_Average/d2*Autocorrelation_Coefficient

Darius.
 
Top Bottom