nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

AI for DOE - any experiences/knowledge?

martindemel

Hi a customer, and JMP fan, asked me if we know about a software which is using AI to design the next experiements. The offering company says they can build better and more efficient designed experiments. He asked the internal group looking int this software to challenge it with other professional software and suggested JMP. So any information would be helpful on that. (I know I have to find out more and will have a first call on this on Wednesday).

Has anyone came across this type of research or use case already? I researched a bit and it seems that two austrian universities (vienna and Innsbruck) has done some research on that and had some success. As it was on quantum experiments I guess this is rather a simulated or combinatorical experiment. Though there was a statement on machining processes. These were from early 2018 (https://www.uibk.ac.at/newsroom/artificial-agent-designs-quantum-experiments.html.en ) and 2015 (https://link.springer.com/chapter/10.1007/978-3-319-23838-8_3) .

Thanks for any feedback.

Martin

Accepted answers

martindemel

New Information:

Competitor is Citrine Informatics (https://citrine.io/), a quite new Silicon Valley start up which is very good in marketing and numbers throwing. However the idea behind is valid somehow.
It seems that they build up a huge database of materials information they use to find the most likely material addressing the objective in charge. To me this is kind of comparable what P&G did on their own with their chemical composition of a certain product they wanted to know instead of a plain product/material. With this they could react much more flexible and more independent from one supplier. They used within others PLS.
It also seems that they are trying to build up that database with experiments from customers to learn better. Not sure if everybody wants to participate in that. It's probably a matter of risk assessment, if I like the results of my innovation experiments to be included in the database to have a better algorithm also other companies then use to make a better materials choice.
This is still an iterative process, and the comanies using their service still have to conduct experiments, but on the materials they suggest. This seems to be better than some classical approaches like what Dysson did with his first product (>5000 trials), I still doubt this will reduce the burden of finding a good design plan.The advantage I can imagine is that they probably get a better starting point.Still there is the question if they suggest anything on the design itself.
Interestingly BASF has a pilot running (https://citrine.io/2018/06/21/basf-and-citrine-informatics-collaborate-to-use-artificial-intelligence-to-develop-new-catalyst-technology/) to accelerate the development of new environmental catalyst technologies.
Citrine has spoken on a DECHEMA congress in Cologne recently, where Bernd, Mahmoud and Alex where present as well, I found out. So please watch out if you hear something from other companies dealing with materials and chemicals.

May we can use this thread also to add information we all gather from conversations with customers where they mention Citrine (to be clear, if a customer asks if we have heard from that - we don't want to hype them by mentioning them ).

I will be keep you informed since I got new insights on their pilot at Lanxess (another chemical company). If you have customers doing a pilot too, please add your gained knowledge to this thread as well. This will help in the future as I believe they will raise more interest in the future and we will be challenged by our customers, though we probably don't need to fear the challenge.

ian_jmp

If nothing else, this thread shows that we should feature Silvio's thinking (presented in Prague) whenever the opportunity arises. He's given a complete description of a specific case (with everything you need to follow along), but I believe it's a very general message that we could promote more. FWIW, I think his was one of the best Discovery papers that I've seen, and of course the P&G brand is worth a lot.

All comments

MikeD_Anderson

Let’s pull a page from Goodnight’s playbook - can you define what the customer means by “AI” for us.

The article that mentions “statistical or artificial learning” discusses using those techniques for “optimization of the design matrix.” I’d argue that the holistic DOE workflow (data mining + optimal design) fits that description. We just chose not to use the current buzz words.

Maybe I’m wrong there, but that’s my first pass read based on the articles.

Best,

M

Dan_Obermiller

It also sounds very similar to Evolutionary Operations or Simplex Optimization, DOE techniques where you evolve to an optimum set of conditions. The algorithms may be newer and more powerful, but it is still an evolution to an optimum.

martindemel

New Information:

Competitor is Citrine Informatics (https://citrine.io/), a quite new Silicon Valley start up which is very good in marketing and numbers throwing. However the idea behind is valid somehow.
It seems that they build up a huge database of materials information they use to find the most likely material addressing the objective in charge. To me this is kind of comparable what P&G did on their own with their chemical composition of a certain product they wanted to know instead of a plain product/material. With this they could react much more flexible and more independent from one supplier. They used within others PLS.
It also seems that they are trying to build up that database with experiments from customers to learn better. Not sure if everybody wants to participate in that. It's probably a matter of risk assessment, if I like the results of my innovation experiments to be included in the database to have a better algorithm also other companies then use to make a better materials choice.
This is still an iterative process, and the comanies using their service still have to conduct experiments, but on the materials they suggest. This seems to be better than some classical approaches like what Dysson did with his first product (>5000 trials), I still doubt this will reduce the burden of finding a good design plan.The advantage I can imagine is that they probably get a better starting point.Still there is the question if they suggest anything on the design itself.
Interestingly BASF has a pilot running (https://citrine.io/2018/06/21/basf-and-citrine-informatics-collaborate-to-use-artificial-intelligence-to-develop-new-catalyst-technology/) to accelerate the development of new environmental catalyst technologies.
Citrine has spoken on a DECHEMA congress in Cologne recently, where Bernd, Mahmoud and Alex where present as well, I found out. So please watch out if you hear something from other companies dealing with materials and chemicals.

martindemel

@Dan_Obermiller Thanks, yes the articles in the links I found may act like these evolutionary (global) optimization algorithms. However, after I got new information I doubt Citrine does anything with the design itself, it seems they just try to inform the next design with the best materials they found in their database matching the objectives a new material should have. Nevertheless we might see more from these university studies in the future.

ian_jmp