Thank you so much for the kind introduction.

As already mentioned, my name is Moritz Moss, scientist at the Institute of Particle Technology

from the Friedrich-Alexander University Erlangen-Nürnberg.

And I thought I also should thank the organizers for the beautiful congress, but then I realized

I'm also part of the organizational team, so good job.

Okay, before we dive deep into the topic of deep learning based synthesis of sedimentation

boundaries, I first want to make sure that you know the difference between AUC and AC.

So thanks to my previous speaker, I don't have to introduce the LumenSizer once more.

Now I just want to highlight or mention some differences.

So for the AC, for the LumenSizer we are using, we're typically working with rectangular-shaped

cells, while for the AUC, of course, we would use sector-shaped cells.

Also with respect to the rotational speed, we see some differences.

So for the AC, at least in our lab, we typically go up to 5000 RPM, while for the AUC, we can

go up to 50 or 60,000 RPM.

And therefore, we typically use, of course, the AUC for analyzing very small nanoparticles,

but for larger nanoparticles or particles in the submicrometer range, the AC is, at

least in my case, the method of choice.

Okay, therefore, when we're analyzing large particles, we can also neglect the effect

of diffusion during the sedimentation.

So here the sedimentation effect is much stronger than the diffusion effect.

And what we obtain from the LumenSizer are these sedimentation boundaries where we typically

see the transmission of the light instead of the absorption, and the radial position

starts typically at around 100 millimeter.

And here are some examples how the experimental results could look like, and we can differentiate

between different kinetics.

So on the first column, the typical case of sedimentation, so here the particles typically

move from the meniscus to the bottom.

We can also see the opposite trends that the particular particles move from the bottom

to the meniscus, so here we have a change in the density gradient, so the density of

the particles is much smaller than the density of the surrounding media.

We can also see both kinetics, like in the third column where we have sedimentation and

flotation in one experiment, or like in the fourth column, we don't see any kinetics.

And I'm pretty sure all of you are able to see some differences between those four classes,

but have you also realized that some of those examples are not realistic experiments, but

rather generated by an AI model?

So would you see any differences here between the realistic ones and the AI synthesized ones?

Hopefully not, because this is the aim of our work, to generate realistic sedimentation data.

But of course I will give you the solution, so the images in the reddish boxes are generated

by AI, while the other ones are from real experiments.

Okay why are we doing this?

So we saw that for not so experienced users, it's still a problem to differentiate between

the different kinetics within the experiments, and it's also still a problem to identify

the meniscus position correctly and therefore to analyze the particle size distribution

or the sedimentation coefficient correctly.

And therefore we would like to develop models, AI based models, which should support the

user by either classify their samples or analyze their samples.

Nevertheless, when you would work with AI, you need a lot of training data.

So whenever you would start working with AI, be aware that you have a lot of training data.

So the first step in our project was to generate a lot of training data.

And therefore the first choice was to use simply classical models to simulate the data.