Deep Dive #3 - The detection of water in food

00:00:00: [MUSIC PLAYING]

00:00:02: Deep Dive.

00:00:04: Hello, and welcome to another episode of Deep Dive

00:00:08: into Applied Science.

00:00:09: My name is Martin Zeng.

00:00:11: I'm your host, and today I have a very special guest for you.

00:00:14: And this guest can introduce himself over to you.

00:00:20: Hi, my name is Maximilian Staudt.

00:00:22: I'm currently the PhD student here

00:00:25: in the Department of Food Engineering.

00:00:28: And I hope we have a nice discussion about my topic

00:00:31: and bring you the concepts of my thesis a bit closer.

00:00:36: So before we dive deeper, please describe your project

00:00:40: in one sentence, if possible.

00:00:43: Yeah, that was a really great question, Martin.

00:00:47: Thank you.

00:00:48: I thought about this quite a lot,

00:00:50: and I go with the possibility to reduce energy usage

00:00:55: while trying.

00:00:56: So the whole title of your project

00:01:00: is the determination of depth profiles

00:01:02: using angle-resolved near-infrared spectroscopy.

00:01:07: That's a very complex title.

00:01:10: So what exactly is behind this title?

00:01:14: In first glance, you can imagine it like you or I

00:01:19: try to find a way to determine water content

00:01:25: in a very close surface area of a sample.

00:01:29: In my case, food stuff, like pasta blades, for example,

00:01:34: if you want to make lasagne.

00:01:36: To a certain extent, you go and go maybe with powders as well.

00:01:40: Powders is a huge part, if you think about protein powders.

00:01:43: And most of the ingredients used in our modern foods

00:01:49: come in powder.

00:01:51: So that's the main focus to find the water.

00:01:58: And the goal is to reduce, with the knowledge of where

00:02:01: and how much water is inside the food,

00:02:04: to reduce the drying times by checking exactly

00:02:08: where's the water and then applying the perfect drying

00:02:17: technique for the point where the water is.

00:02:23: Before we come to the drying of the water,

00:02:26: why is the location of water such an important topic?

00:02:30: Why do I have to know where water is in food?

00:02:34: Usually, it's like you try something.

00:02:37: And a lot of energy and knowledge

00:02:42: goes into how to increase shelf life of food.

00:02:47: So drying is a very important process step

00:02:51: in increasing shelf life.

00:02:53: But it's also very intense in energy usage.

00:02:57: So usually, to stay safe, you try longer than needed.

00:03:03: So that's where I come in place, while showing, OK,

00:03:07: there and there is that much water.

00:03:09: We can determine the right time, like the perfect time

00:03:14: for the drying process, so that we don't need this extra time,

00:03:18: like the safe time.

00:03:20: And this time, the safe which in time,

00:03:23: also means a savage in energy use.

00:03:26: Because drying is not like--

00:03:31: with more time you dry, the harder it

00:03:34: gets to get the stuff dry, the more energy you need.

00:03:37: So it's very important to keep it,

00:03:39: to keep the shine as short as necessary.

00:03:42: That's the goal.

00:03:43: And you also can use it not only for food,

00:03:45: but also in plastic or even in clothing.

00:03:49: So everywhere we're trying is applied.

00:03:51: But one limitation is the penetration

00:03:54: depth of the mechanism behind the angle

00:03:59: resolved NIR spectroscopy.

00:04:03: Your research, does it focus on both aspects,

00:04:06: on the drying and on the location?

00:04:08: Less on the drying, more on the location.

00:04:10: OK, OK.

00:04:11: And can you recognize water in any type of food?

00:04:15: Or is there only very specific types?

00:04:18: It's depending less on the food, more of the penetration depth.

00:04:23: Like if you have a product that is very, very huge, very big,

00:04:28: that it's very, very unlikely.

00:04:31: The method I'm using just penetrates,

00:04:34: the light penetrates the sample to a few millimeters.

00:04:39: So there is a limitation where you can use it.

00:04:42: So like pasta plates is like the best way to understand it

00:04:46: because you have very tiny, very thin blades.

00:04:49: They are not thick and then you pretty much

00:04:51: can use this method very, very easy.

00:04:54: So you try to locate the water with a so-called

00:04:58: NIR infrared spectroscopy.

00:05:00: What exactly is this?

00:05:02: Let's start with the spectroscopy part,

00:05:04: just for better understanding.

00:05:08: Generally spectroscopy means you measure

00:05:11: the interaction of matter with light.

00:05:15: So in my case, using NIR, that's a short version.

00:05:22: Yeah, for near infrared, maybe to better understanding,

00:05:28: to see light as a kleptomagnetic wave from two.

00:05:34: And we humans see just in a certain range.

00:05:38: That's the visible part.

00:05:39: And right next to this visible part

00:05:41: is the NIR infrared light, or the infrared light.

00:05:44: And you can separate it in near, mid, and far infrared.

00:05:48: And I'm using NIR infrared.

00:05:50: So I'm using the NIR infrared part of light

00:05:53: and see the interaction with the matter.

00:05:57: And after this interaction with the matter,

00:06:00: there are certain informations hidden

00:06:02: in this changes in light.

00:06:04: Then I can use to interpret it certain informations.

00:06:09: Generally used for the total water content, total fat content,

00:06:14: or total protein content.

00:06:16: So it's a technique that's around for like 70 years.

00:06:20: Mostly used, or mainly used by American companies.

00:06:25: Like there's more to focus in this regard.

00:06:28: It's a very easy way, a fast way to find

00:06:31: like this macro food contents.

00:06:38: And what I'm doing, I'm not trying to find the whole water.

00:06:42: Like if we have a sample, by now it's

00:06:44: used to find in the sample the entire water content.

00:06:48: And I try to look for the fine locations of the water.

00:06:54: So now that we know what NIR is and how it works,

00:06:58: how exactly do you visualize the location of water?

00:07:01: There is a neural network, very helpful.

00:07:05: A neural network is great for pattern recognition.

00:07:09: Imagining you have your favorite dish from your grandma.

00:07:14: Like you know the ingredients and you know how it tastes.

00:07:17: But you don't know how she cooks it.

00:07:20: When do she mix stuff together?

00:07:25: And just how the process is that she

00:07:29: makes to get your favorite dish.

00:07:32: And so you know the ingredients and how it tastes.

00:07:37: Then the neural network is crazed for this

00:07:41: to find a way from your ingredients to your dish

00:07:45: by trying, simply by trying around and adjusting things

00:07:50: inside the neural network.

00:07:53: And then you got the result.

00:07:55: So now it's the thing with the neural network.

00:07:58: You have more than one result.

00:08:00: You need more than just one result.

00:08:05: So then you bring other stuff inside.

00:08:08: Other foods like your favorite pizza, your favorite burger,

00:08:11: everything.

00:08:12: You know the ingredients.

00:08:13: You know the result.

00:08:14: And by giving all this information,

00:08:16: the neural network will find the pattern.

00:08:19: And then you have a cooking network in the end.

00:08:23: Because it now finds out how to cook certain ingredients,

00:08:26: what suits together.

00:08:28: But therefore, you need a whole lot of information.

00:08:33: And my information, my ingredients, so to speak,

00:08:36: are like the crafts I receive from the NIR.

00:08:40: The water content, where I know where it is.

00:08:44: Because I use a 3D printer.

00:08:47: So I have a 3D image of my sample, so to speak.

00:08:52: And I know where the water is and how much is inside.

00:08:55: And with this information, I try to visualize the stuff.

00:09:02: Because I know I got the result and I got the information.

00:09:06: So I model the neural network to go from my information

00:09:10: to my result.

00:09:12: That's a very tough part.

00:09:13: And by now, I don't really know which type of models I can use.

00:09:19: I can try.

00:09:20: Or is it one model?

00:09:21: Is it more?

00:09:22: How much information do I need?

00:09:24: That it's like the next step that will be coming.

00:09:28: Sounds like a very complex process overall.

00:09:31: You mentioned your neural network.

00:09:34: And that it needs a lot of data.

00:09:37: How much data exactly?

00:09:40: Can you specify this?

00:09:42: Before I started, I can't really tell you how much data I need.

00:09:45: I just see there will be a point where it's like,

00:09:49: don't getting better.

00:09:51: So if you have your neural network and you trained it,

00:09:56: then you go there with an unknown result, unknown things.

00:10:01: And then you see how it responds to this new information.

00:10:05: And with this validation, you know how good your network is.

00:10:10: And then with a few mathematical things,

00:10:15: you can get a percentage of how good your neural network is.

00:10:20: And then it's deciding not only the information is important,

00:10:23: also the way you model your neural network.

00:10:29: So it's a very--

00:10:31: I can't give a direct number by now.

00:10:34: I will see it later on.

00:10:36: And it's also depending on how good my results are

00:10:39: from the neural network.

00:10:43: You already mentioned that you are working with a 3D printer.

00:10:47: So the Department of Food Engineering and Technology

00:10:50: is by far not the only department that's using such printers.

00:10:54: But how do you exactly use a 3D printer in your project?

00:10:58: Using the 3D printer, there are two ideas behind that.

00:11:02: First of all, during my measurements,

00:11:06: sure that my samples are stable, so that there are no dynamics, no changes, like water

00:11:13: could try, and while I'm measuring, that could change the results, so that's one thing, like

00:11:25: I create with the 3D printer a stable sample, and on the other hand, I have in the program

00:11:32: I got my set for the special visualization hardware, and I need this because it's already

00:11:50: made in a computing environment, so neural network doesn't know a picture, those concepts

00:11:58: are strange to it, they don't understand it, need to translate it in a way that the computer

00:12:04: can find patterns, so giving it a picture won't mean much anything, but in a 3D printed area,

00:12:11: you also will have it in an algorithm or something like that, in a program, and a neural network

00:12:17: can work with this program, and in this program already are my information, where are the spots,

00:12:25: where the water later on is supposed to be, so I got my result in this information, and

00:12:32: because I use a neural network and I train it, I need the result in the end, so that

00:12:37: I can use this pattern recognition method, and that's the reason why one method to train

00:12:45: my neural network was the idea with the 3D printer.

00:12:49: Once your project sounds very complex, not only with the 3D printer, also with the neural

00:12:55: network, do you work alone or do you work in a team?

00:13:00: The first year I worked alone, now I have an internship from Italy, we are working on

00:13:07: a second method for a second result if you want to, so now my results for my neural network

00:13:13: came from the 3D printer, and we use another technology called NMR mouse, it's another

00:13:22: kind of spectroscopy that don't bring us like a visualization, because if you ever use the

00:13:30: CAD program to create something into a 3D printer environment, you see the stuff you create.

00:13:37: There we have more like an abstract way, we have a graph, not a problem, because you also

00:13:43: can try to find a pattern between your information and a graph.

00:13:49: The advantage of this idea is, in the industry you don't usually know your stuff, like if

00:13:55: you have your pasta plate, you don't know where it is, this way it's just, if it's working

00:14:02: with the 3D printer, I assume it would work with other methods, like the other method is

00:14:07: the NMR mouse, there are plenty of methods, but that's one we have here, so then we can

00:14:13: actually create a method that could be directly used in the industry.

00:14:20: That's a very good point, because once you finish the project, what are possible fields

00:14:27: of application for your project, in what areas can your research be used, who can profit

00:14:33: from it?

00:14:34: Pretty much everywhere we are trying is used to a certain extent, I mean I got the limitation

00:14:38: of this, just the light penetrates like a few millimeters inside the surface, so I'm

00:14:46: limited by trying products that are just thin.

00:14:51: The good thing is, with drying usually you have thin products, like plastic, pasta, protein

00:14:57: powder, other powders, they're very thin, so pretty much in a lot of drying environments.

00:15:06: Maybe even in other situations where you can't think of, maybe even here if you say, we have

00:15:13: a nice desk here, wood and clays on top, and you want to see if there's clays in the water,

00:15:20: then you can say okay, maybe this won't dry as good, so you could use it in such an approach

00:15:26: as well, but maybe it's meant for food.

00:15:29: So you're not limited to the food industry, but well, you focus on it.

00:15:35: So we are here in the food department, but the usage outside the food.

00:15:39: And please correct me if I'm wrong, but it sounds like almost everyone in the food industry

00:15:44: could possibly profit from your project.

00:15:48: If they have this tiny fraction or these thin samples, then they could, yes.

00:15:54: And the good thing is with NIR, NIR is a cheap way, it's widely spread, and you can

00:15:59: do it while you're making the process, it's online.

00:16:03: So one thing I ask myself since I've actually never heard of this project, are there similar

00:16:09: projects or are you like forerunner in your field?

00:16:12: So and so, like the neural network part and using an NIR is something that has been used

00:16:19: before, there are quite a few papers regarding about this topic.

00:16:25: Using the anchor-result method is something new, that's the different approach in this

00:16:32: entire topic.

00:16:35: Maybe we come to this anchor-result where why it's that important.

00:16:39: As I said earlier for neural network, you need quite a lot of data, quite a lot of measurements.

00:16:44: Now if you are in the food environment, you have like 100 samples, you measure them, it's

00:16:48: not enough to make a neural network.

00:16:51: The anchor-result approach means the measurement is taken from different anchors.

00:16:56: So instead of taking one measurement per sample, you take like 100.

00:17:02: So the amount of information you got to train your neural network increase.

00:17:08: That's the main part and the main idea behind this anchor-result thing.

00:17:14: When exactly did you start this whole PhD project?

00:17:19: Pretty much one year ago, on the 3rd of April.

00:17:24: So it's really, really close like a year ago.

00:17:27: And you plan on finishing when exactly?

00:17:30: It should be done in two years from now.

00:17:32: I don't expect the perfect deadline, don't worry.

00:17:35: Okay, interesting.

00:17:38: Also a question that really interests me.

00:17:40: Which obstacles did you face so far?

00:17:43: What are the challenges of your project?

00:17:45: Especially in the first year and I talked with other PhD students about it, what I didn't

00:17:48: expect is like getting this stuff together, working with this bureaucratic environment

00:17:58: is something you get used to and takes quite a lot of time.

00:18:02: And then if you go off this more into the science stuff, the first samples I created,

00:18:11: they needed to be stable.

00:18:13: And for being stable, I tried a lot of things with my samples and noticed, okay, no, that's

00:18:20: not the approach.

00:18:22: So basically trial and error.

00:18:24: Trial and error to find a way to have my samples stable.

00:18:28: For example, the samples I got now, I shouldn't, I won't use them for my measurements.

00:18:36: Like the new addition that I'm making, they should be stable.

00:18:40: I mean, I obviously will know what after results by more confident because all the informations

00:18:44: I got, working with them flew into the construction of the new ones.

00:18:50: So I assume that with the newer samples, I can create my information, my measurements.

00:19:00: Another thing that was hard because I'm working with neural networks, I'm working with CAD

00:19:04: Broadcoms to create my samples, working with an IR and also the ankle resolved positioning

00:19:12: system.

00:19:13: So that's like four types of systems I'm working with.

00:19:17: And you, with all of them, you get need to get used to and to a very certain extent that

00:19:23: takes quite a lot of time and especially I'm a food engineer, not an informatic.

00:19:33: So I'm not an IT guy, I have from the IT department.

00:19:36: So getting into the neural network, the programming and so took quite a lot of time as well.

00:19:42: Yes, that's what like the biggest obstacle.

00:19:44: Since we're already talking about Tria University of Applied Sciences, you already completed

00:19:49: your bachelor's and your master's thesis here at university.

00:19:53: So why did you choose to do your PhD also here?

00:19:57: Why didn't you choose to go, I don't know, to Luxembourg or somewhere else?

00:20:00: Yeah, I had this, I made my master's thesis in Luxembourg and I had the chance to work

00:20:07: in this company where I wrote my thesis.

00:20:11: The problem was getting to the company, first of all, second of all, I know the Hochschule

00:20:18: because I made bachelor's here, I know Tria, I love Tria, that's the reason why I wanted

00:20:24: to stay close and I hope that I can bring something to the portfolio of the Hochschule

00:20:30: Tria in general.

00:20:33: And so it's like it's a mixture between I know where I am, I know the people, so I

00:20:38: hoped it was like a plug-and-play situation.

00:20:41: I expected that I don't need much time to get into the, to work, then I noticed a few

00:20:50: obstacles that we mentioned before.

00:20:52: Well, sounds like you're adding a lot to the portfolio of Tria University of Applied

00:20:57: Sciences.

00:20:58: So if people are interested in your work, in your whole project, how can they contact

00:21:03: you?

00:21:04: Yeah, they always can contact me by mail or on the Hochschule website where there's the

00:21:10: phone number, the mail address, and if you wanted to see a bit more, there is a colloquium

00:21:15: of projects from the Hochschule Tria University of Applied Sciences and there you can find

00:21:22: my project as well and by time the information will be updated on the website as well.

00:21:28: Perfect.

00:21:29: Thanks for letting us know.

00:21:30: So it sounds like an overall very interesting project, even though very complex, especially

00:21:35: the overall title of your project, but still I think it's a really interesting work and

00:21:40: it looks like you have a lot of future fields of application for this.

00:21:45: So thank you very much for presenting it.

00:21:47: Thank you for having me here.

00:21:49: And I hope our listeners or viewers had a fun time.

00:21:53: And if you still have questions for Maximilian, feel free to ask them, feel free to contact

00:21:58: them and we will see each other in the next episode of Deep Dive.