Using DOE optimizing a distillation process

[#1]

Hi Community,

I have been asked whether DOE can be used to optimize a distillation process where the feed composition varies from tank to tank. Factors could be Feed Rate, Reflux, Temperature but they are adjusted according to the feed composition and therefore not independent. Any ideas?

Thanks in advance

[#2 · Accepted solution]

What do you mean by "optimize a distillation process?" Reduce cost? Reduce time? Reduce by-products? Increase yield?

The composition varies, but can it be quantified for each tank? Is the variation due to set up or how the process proceeds over time?

What scale is the experiment to be? Can a small scale laboratory experiment be done?

What if you do not adjust the process factors despite the varying composition? You could then learn about the effects of composition and process factors.

[#3]

Hi Mark,

Our final goal with the experiments we are running currently is to demonstrate successful separation of the key components. We also need to build the knowledge to be confident handling the actual feed when it is available. Therefore, our aim is to find an optimum condition where we get high purity product (high composition) with high yield. So by optimizing the distillation process I mean reducing uncertainty and increasing yield.

The composition of the feed varies and it can be quantified for each tank. The variation is due to the changes upstream of the distillation column. During each batch the expectation is to have a constant feed composition though.

It is a pilot scale we are talking about and it is not possible to run small scale laboratory experiment (it could be possible but the smaller distillation column is fully occupied for the months to come for production of another product we need to deliver to the customer).

We can continue with the current parameters despite varying composition, but from batch to batch, we would lose either the yield or purity. But I agree it can be a good way of seeing how the column performs and how well of a separation we can get.

Thank you very much for your reply. My colleague got some food for thoughts and are happy. I think he will use historical data instead of running a DOE right now.

[#4 · Accepted solution]

Mark asks some excellent and pertinent questions.  I have some more:  What are the response variables? Why do feed compositions vary tank-to-tank? why and how much do you adjust variables in the current process? Have you identified ALL of the variables (controllable and noise)? Do you have any hypotheses as to which variables have the greatest effect?

To get clues about the consistency and factor effects and estimate measurement error, you could start with a directed sampling (nested/systematic...Tank-to-tank, within tank, measurement).  Then after focussing on the larger source of variability, you could certainly design experiments to understand causal relationships and develop models.

[#5 · Accepted solution]

@Sop In addition to all of @Mark_Bailey's questions I'll ask another few:

1. Do you have historical data from the production process in question that can be used for modeling purposes?

2. If the answer to 1. above is, no revert to @Mark_Bailey 's questions and thoughts. If 'yes', then:

2.a. You may want to torture that data from a modeling and process understanding point of view. Starting with an explanatory modeling approach, with an eye towards building a predictive model since you goal is 'optimization'...from whatever point of view or criteria, @Mark_Bailey lists many, you choose. Who knows? You may be able to solve the problem at hand without ever having to resort to DOE.