COSMOFACTORY

This transcript may contain errors, please refer to the audio in the podcast to ensure accuracy.

[00:00:00]
Deanna: Welcome to Cosmo Factory, a podcast by Cosmoprof Worldwide Bologna, where we explore the entire cosmetics, personal care, and fragrance industry supply chain. I'm your host, Deanna Utroske. And here at Cosmo Factory, we look beyond the trends to discover the ideas, initiatives, and innovations that are truly advancing beauty.
Let's get started. In this episode, our topics include artificial intelligence, industrial automation, vision systems and neural networks, track and trace technology, and more. Joining me today on the Cosmo Factory podcast is
Francesco Ringressi, Cosmetic Business Development Manager at SEA Vision.
Francesco: Thank you so much for the invitation today to this talk. I'm very excited to explain you our technology. Thank
Deanna: you. I want to mention that we're actually recording our conversation [00:01:00] today here on the show floor at Cosmopack in Bologna, Italy. Um, where we can actually see here at the show, say a vision technology in action.
So let's start there. Francesco, tell us about the lipstick inspection system that your team has developed.
Francesco: Yes. So for to this exhibition, we prepared a very special machine for you to see. that basically have the artificial intelligence technology embedded to a conveyor belt that represented the production factory that you may have in your company, uh, where basically multiple cameras with AI algorithm are working together to make the control as equal as the human eye.
I would like to say even better because our technology is far superior in terms of the defect catching.
Deanna: Excellent. Excellent. And tell me why you started with lipstick. Why not a product that might be a little bit less complicated?
Francesco: Um, because we thought that was [00:02:00] easy. So for the reason we start from that, no, uh, jokes apart.
Uh, the truth is that the vision system Needs to rely and solve a problem that you have on your common packaging line. So for this reason, the concept to have a machine to inspect the lipstick was one of the first tasks, because at the moment the job on the production line is done using human eye. So it's a very tiresome job as well.
Keep in mind that you have people that are looking at defects that are tiny, has a fraction of millimeter, uh, eight hours a day, maybe in three shifts. time with a special light condition. And so it's a very stressful job. And also it's normal. The attention span of each person. Tends to decay during the day and so was the control was not objective.
So we thought that was a good moment during COVID pandemic where the makeup market was not so strong in demand of lipstick [00:03:00] to start thinking about this kind of control. And I want to stress out that we can do the control now because the technology that is available is make us capable to do so in terms of artificial intelligence.
Deanna: And tell us more about what this system is actually looking for. How did you teach it to see lipstick? What can it notice? What can it tell us?
Francesco: Um, so, uh, the computer vision is, uh, the concept of, uh, identify what is inside an image. Um, there are different level of, uh, image recognition. You can identify. Uh, inside an image that, uh, there is an object.
So there is an a lipstick. You can identify this position, so it's in the middle of the image. Then you can start creating boundaries around the different part, such as the body, the tip, uh, the neck, uh, or the mechanism itself. And then there is the last part where artificial intelligence really came, uh, on, uh, that is the semantic segmentation [00:04:00] strategy.
So basically we train the system to, uh, automatically, uh. Apply a label for each pixel that, uh, the image is composed of. This is done automatically and autonomously. So in order to train the neural network, we create a huge set of images, have a real data set with the real, uh, lipstick, fake lipstick that we create using 3D technology and other lipstick that we creating, using another artificial intelligence to generate and based on the data set, then we train the system in order to perform some task.
automatically and we reward the algorithm in case that, uh, uh, the result were in line with our expectation. That process is, seems to be complex from maybe the people hearing us, uh, but the truth is that it's quite normal and this is the structure of mainly all the artificial intelligence that are today.
Deanna: And tell me about some of the details that you've trained the system to [00:05:00] identify in terms of color variation, those sorts of nuances.
Francesco: In my idea, the fact that now the industry is moving as well. into industry 5. 0 paradigm. Uh, I'm here today also to make, uh, more understandable, the artificial intelligence technology to the people.
And we train using the human expertise because, uh, we want to replicate, uh, at the same, uh, image processing that have the human eye. Uh, against the lipstick. So we work with the industry, first of all, and with the operator inside the industry that, that perform this task every day in order to, uh, perform the same control in terms of, uh, uh, holes that could be around the tip, for example, color that is miscoloration due to the fusion that is not correct, or the wax is inside and the oils, um, the problem that is quite common that the lipstick itself is not insert corrected inside of the, yeah.
Uh, the mechanism, so the bullet is damaged [00:06:00] and all the other control that are quite normal, such as a scratches, um, uh, marks or logo, not perfectly printed or some strange, uh, texture or working with AI give us the capability to perform all this type of control automatically as a human person will do, uh, expanding and surpassing the fact that for each type of product, you need a specific recipe.
Deanna: If I understand the technology correctly, what you've created is not something that will do quality of checks on one brand of lipstick. Any lipstick could show up in front of this vision system. It will not only recognize that it is a lipstick, it will be able to identify which bits are the packaging, which bits are the product, um, all the elements there, um, and then be able to run these quality control checks on a product it's effectively never seen before.[00:07:00]
Tell me more about that.
Francesco: That is the fascinating part, uh, because, uh, we trained the neural network in order to, and the algorithm that we embedded in the vision system in order to be, uh, to perform the task of assign each pixel automatically. So
Deanna: there have to be a lot of pixels.
Francesco: There are, there are a lot of pixel.
I mean, keep in mind that, uh, uh, we are using five megapixel camera, for example, for some of the controls. So there are a huge amount of pixel and a huge amount of information because then an image is not the same image that we see on our screen, but in our, uh, technology, an image is composed is a 3D image of three different color, and then that is analyzed.
But the system is performing these, let me say, pixel assignment automatically and in real time. And, uh, also for me, it's fascinating to make it work. So I'm very happy that this technology now is developed in this way. Um, And again, [00:08:00] the system itself will recognize the different part and create the different, uh, assignment and label on the pixel, but then is always the operator and our customer to select if what is seen by the system is small enough or big enough to be considered as a reject, for example.
Deanna: Tell me more about those parameters. How are your customers setting them? What sorts of things might they be looking for? And what decisions are they asking the machine to make for them?
Francesco: Well, um, the customer at the moment that we work with ask us to control, uh, for example, the, with particular interest that the area around the tip, because of course, is the first one is very eye catching.
Uh, so the controller we perform over there is that the shape is okay, that there are no holes on the back or on the top of the tip that no bubbles are formed all around the bullet and the setup is quite simple because the [00:09:00] system as complex as it could sound and Promise is not that much and needs to be easy to interface with from the human operation and daily routine point of view.
So we set for each camera a set of standard control that we perform on the body, on the insertion part, on the tip, where all the artificial intelligence is running and by simply setting some threshold with a slider on our HMI, you can then accept or reject a specific defect.
Deanna: Um, and so it's clear that this vision system is helping with quality control, but I'm imagining there are other advantages maybe in terms of production speed or, or other, other sorts of benefits to using this technology.
Help me think about that.
Francesco: Well, um, as a vision, uh, the company work with, uh, that is part of the markets in a group. We run machines and lines that are much, much faster than the lipstick, let me say, casting [00:10:00] process. Because casting lipstick is time consuming, because you need to hot, you need to make hotter the lipstick.
And then you need to froze it down in order to able to extract. So it takes time, and so the machine are not that fast. In terms of productivity, uh, usually human operation are the bottleneck. And so, removing the human, uh, control, the machine, could run faster. The truth is that, uh, now the formulation always tend to have ingredients that are always a little bit more complex to, to use and complex in terms that they lower the average speed of production.
Um, but our controller is set up to run much more faster. Probably 10 X time than what we are showcasing here at the exhibition because the technology can be scale up. The memory can be scale up as well. And also because, uh, we are, as a [00:11:00] vision, we work in other industry where the output is He's even 600 pieces per minute.
So it's not a problem.
Deanna: Yes. And you and I spoke several days or even a week or so ago by zoom. And I think you told me that the technology is actually helping manufacturers troubleshoot some of their production errors. How is that?
Francesco: Well, um, you know that Italy is composed by third party manufacturer or in general, a cosmetic business is scattered among different players.
And our technology rely on cameras and the cameras then have threshold. And this kind of Uh, parameters can create what I called an electronic range board on an electronic panoply that you can share and agree with your customer. So during the industrialization phase or before the production, you can both agree that the quality, uh, it's maintained all around the batch.
First of all, [00:12:00] because we perform 100 percent of control and based on the threshold that you set inside the system, you can set an agreement between the two. two party working together. So, uh, in this sense, it helps avoiding recalls, avoiding, uh, I don't want to say legal battle because I don't think, but the reprocessing time that could happen when you have AQL quality control or sampling quality control.
We perform 100 percent of the best quality control. So
Deanna: you can start to understand Based on what the vision system sees, what step in the process might need to be remedied if there's consistent laws?
Francesco: Yeah, I agree completely. Uh, thanks for noticing that. Uh, the idea is that, of course, it's a reject is categorized.
With the counter, the counter can have a trend in time, we can catch the trend and raise an awareness of the operator running the machine that maybe some key parameters such as temperature, pressure [00:13:00] or other things that are quite common, that the that there is a the usage of the molds that are inside the machine.
Maybe you need to change it so you can anticipate and, uh, avoid to waste a lot of material and the reprocessing time. So, uh, our system do not reduce. waste because waste is always produced, let me say, but of course it can reduce the reprocessing time and avoid maybe using again some of the equipment or new parts.
So in this sense, we reduce the waste in production.
Deanna: I guess another question I have is if our industry is ready for this technology, like, I often get to tour different facilities in the ingredient space or in the manufacturing space. And the next time I go into a manufacturing facility, am I, should I be expecting to see this sort of AI inspection system?
Francesco: First of all, I hope so because we are here at the trade show. But I think that sooner or later, this [00:14:00] technology or this kind of technology will enter inside the company. Cosmetic industry at the moment is not so exposed to this kind of automation control. Only a few players have already set and used visual system, mainly maybe for controller related to print quality or code reading, etc.
But this is something that is a little far beyond that. And also, when we designed the system, we thought that the best. design was to be as big as the same space the young man can fit on the line. So it means that the form factor of the system is the same form factor of having a person standing on the line.
So, um, again, we don't want to replace the operator, but we want to give them a more rewarding job inside the company.
Deanna: Excellent. I think you mentioned when we spoke earlier [00:15:00] that You've noticed in some product manufacturing situations, there are, you know, non disclosure agreements in place that limit the use of cameras or photography, and I know this again from touring facilities, they'll have me leave my phone somewhere before I get to go into the facility.
Um, but I actually, um, got to write an article recently about. Independent brands. So some of the smaller and emerging brands in the startup space here in beauty and how they're working with contract manufacturers. Um, and the team at and karate cosmetics told me that they've introduced what they call a filming in factory service.
Um, and that not only helps these independent brands sort of showcase the production process and Post videos on social media about how their lipstick gets made and this sort of thing, but they have found that some of their larger customers, um, you know, more conventional beauty makers. are interested in now showing the production process.
So I'm hoping this is good news for your technology.
Francesco: Well, um, I, I think so because, uh, even, uh, [00:16:00] this morning when the exhibition starts, we have a lot of interest from young people and in general, maybe even people that do not work directly in lipstick manufacturing that they were amazed by the technology.
Um, I think the open, the beauty world needs to be more transparent, not only in terms of raw ingredient and packaging and how it's made, but also from the processing point of view, because, uh, there are always machine and people behind a product. And we can, uh, you can, if you can open your doors, I'm sure that all the people will, uh, be very happy to see how the product are made in a fair way.
Deanna: Yeah. Yeah. Excellent. Let's, um, broaden our perspective a bit on say a vision as a company. Beauty is just one division within say a vision. Can you help us understand the company's full range of capabilities? Okay.
Francesco: Well, we work in the life science, uh, and. Cosmetics still is not part of life science business, but makes our [00:17:00] life better.
So in some way it's similar. The company was born around 30 years ago right now. And we were focused on pharmaceutical packaging business as first, always with visual system. But now after so many years, the range of product in our portfolio is quite big. And also we scattered all around the world. Um, at the moment, there are around 400 people, more or less, working in Italy and in different subsidiary worldwide.
Uh, we start from, uh, uh, from a garage, not a real garage, but a house in the countryside that, uh, a casino to say in Italian. And from that, now we have a very beautiful at quarter in Pavia and, uh, the range of the product, I think the most interesting one also looking and how, uh, How the world is moving is a traceability in terms of production, uh, that it, uh, in some area as pharmaceutical is already [00:18:00] present in other area, uh, for different reasons, such as fashion is moving or being produced while we speak.
Uh, because That leverage in case it's not only safety, but also, uh, waste reduction and sustainability. And, uh, I know that there is a huge effort from all the player in the beauty market to push on forward on sustainability and recycling and green deal in general. So I think that these will be the next, uh, impact in the area.
Deanna: That's helpful. And since you mentioned the company's history, can we hear a little bit about how, or maybe when the transition happened from. a basic vision system, you know, more sort of simplified automation into this neural network technology where the machine is working, like I said, with objects that might be unfamiliar with, or in some ways making decisions.
How did, how did that technology transfer, um, excuse me, how did that technology advancement, [00:19:00] um, affect what your company can do?
Francesco: Well, um, the answer is quite hard, but easy to understand. Technology is always moving. So, you know, As you can see, we transition from, uh, black and white camera, for example, to color camera.
And then there were analogical camera with a lot of cable to today that we have wireless camera. So we always chasing new technology in terms of hardware. But as well, technology is moving in terms of software. Uh, the fact that now on the market is possible to find, um, graphic card with, uh, very advanced chip that can perform real task.
And also that there are some scientists or data scientists that are able to generate a neural network to perform task and to write the code. So there is starting at this kind of new field that is artificial intelligence for everybody. Uh, we blend all together. So the transition was that. Since the technology is [00:20:00] available, we needed to incorporate it because we want to be ahead of the game and deliver product based on this kind of solution to the market.
Deanna: Yeah, excellent. I'm wondering if we can think about maybe some other ways that vision technology could or maybe is being used in the cosmetics and personal care industry. I mean, the sale vision work in track and trace.
Francesco: Uh, yes. So, uh, track and trace is, um, maybe a little bit misleading for the people hearing us because now at the level of a regulatory or by law, you need to trace the batch production.
So it means that you have a unique code for the batch, but the traceability that we are proposing is that we trace each product all along the supply chain. So it means that each product became unique. We, uh, give them, uh, What somebody like to say a digital product password or digital product, uh, identification that basically in can be used, we can give them using [00:21:00] different data carriers.
Um, now, nowadays is more common that people pay with their mobile, for example. So we can use NFC technology, uh, QR code are quite common after COVID situation that we know all around. My mother's know what is a QR code now. Uh, and, um, And on our simple data matrix, basically, we trace the product from the production to the final distribution.
But not only that, we can also link the information that are related to the raw material that are used, the sub batches that are needed for that specific production, and put it all together. So to have a full transparency of your product and on that, then you can start calculating, for example, the combo carbon footprint based on the distance of your, your raw material or, uh, calculate the cost related to shipping and distribution.
Uh, the cool fact that with a product that have this kind of feature [00:22:00] of traceability, customer and consumer can be directly linked. And also speaking always with AI, that is the topic of today. If you think the huge amount of data that you can gather. For example, then, uh, IE strategy based on this big data set to hyper target customer, uh, with marketing probably will have a cost for engagement that is far, far less than the one that we have today.
Deanna: Yeah. No, that's excellent. Mentioning the link to the end consumer. And I think when we spoke before you use the expression that technology is very powerful for the brand owner. Can you say a little bit more about. About what the potential is there?
Francesco: Well, of course, there are different function inside the company or inside the brand owner that want to understand how their production is going.
So you know exactly, for example, tracing the outbound and inbound product. how much time it stays inside the warehouse because it's unsold, or how many products you're sold are sold in real [00:23:00] time through e commerce or retailers. So, um, and maybe this is from the sales team, but from the operational point of view, somebody wants to know exactly how many products are inside the warehouse in order to plan the production for a restock.
So with this kind of, uh, technology, you always have real time data, uh, where are your product are. And of course, uh, And one of the bigger benefit is that if you can control your supply chain, because knowing exactly what kind of product you are selling to your distribution or to your, uh, third party logistic, for example, you know, you know exactly that this product needs to be within that market and not diverging somewhere else that, um, we know that this is an issue that is affecting the market and affecting the brand.
So, um, there are. Also, other powerful tool in terms of transparency again, because as you are mentioning that video for the company opening [00:24:00] the raw material and ingredients and with the safety of the customer in mind, you go always giving the customer a real product on the shelf that represents the safety, the safety maximum, in my opinion, for that.
Deanna: Yeah, excellent. And you've mentioned raw materials a few times. Are you working with any ingredient suppliers specifically, or could you use vision technology in that sector?
Francesco: Generally speaking, we had some technology that's using a near infrared hyperspectral analysis in order to make chemical analysis compound.
Um, in our portfolio. Of course the raw material. For, for us, probably the most important thing is to, uh, link to the production order, the, to the batches that are produced. So we have as well that kind of platform that use the, uh, that monitor the usages of raw material and the inch code and all the, all the production in all the production steps [00:25:00] from the raw material sourcing until the reaching to your plant.
Sub batches, a packaging batch and a filling batch, for example.
Deanna: Are there other ways that I haven't yet imagined that your technology can support transparency along the supply chain?
Francesco: Well, no, but we can have a talk later on, maybe we find it. You can
Deanna: dream something up. Exactly. Yes, lovely. And what about, um, questions that I should have asked you?
What, what sort of questions do you get from your customers? What sort of questions do you expect to answer here at the trade show? What do people want to know?
Francesco: Well, uh, of course this word, for example, what, uh, many customer ask me how we manage, uh, data security, for example, please, uh, because we are taking some images, but I want to reassure you that no images is going outta the company without, uh, the consent of the company.
And in generally, we are not, uh, extracting data. So, uh, for us. Protecting [00:26:00] the, uh, intellectual property is a must and we do this already with so many companies that for us, for us is normal. Um, then another topic that I have that much is that people are still scary about a vision system because, uh, there is still not so many knowledge on that.
Uh, as a vision. Uh, we are, um, doing some kind of job working together with the industry in order to increase the understanding. Uh, so we are creating a hosting also some, uh, meeting with AI specific team in order to make the hour. A customer a little bit more in to detail on that. And of course, customer ask us about the pricing because this is other thing.
So I'm, I'm quite sure that everybody's going to ask the price of our technology. Uh, the return of investment of this kind of, uh, product. It's quite [00:27:00] low, uh, also keep in mind that the total cost of ownership, uh, needs to be taken into consideration. So it's not just the one time cost, but it's the benefit that can bring you in the long run.
And, um, from what we see in the long run with the main medium speed, uh, on the line increasing, et cetera, et cetera. Uh, then of course it will benefit more. So the, the early adapter, in my opinion, we'll have the most, the most skill to spend on the market.
Deanna: Yes, yes. Oh, that's a very important point. Thank you.
I'm wondering as well. What happens if someone loves your technology and chooses to buy it? How does the installation work on the line? What about training the person who's going to be setting the parameters or servicing the system? There's some concern. Talk about that piece.
Francesco: Well, um, say vision usually have an approach that with our project management team, we work together in order to create a gun to related to the project that we needed to perform.
So there is an evaluation phase of [00:28:00] the project. Problematic of the customer, not only speaking about lipstick and based on that, then we create an action plan that is agreed. Uh, of course, nobody can stop them, the existing machine. So doing things on new machine is much more easy, but if we are speaking about retrofit, what we call on the existing line, uh, then we find that the space constraint, uh, we find the place that the control could happen because customer have a lot of dreams sometimes and I'm, I'm.
I'm sorry, but not everything you, you think is possible to do can be done in the way. Um, and based on that, that the service, the system is installed and commissioned with training, uh, remote connection in case of problems. So we give support. And since we are a multinational with offices branches all around the globe, we can offer a 24 hour assistance in case it's necessary.
Deanna: Yes. Yes. Excellent. Oh, that's very helpful. Thank you. Can you tell me about any future technologies? Are there any, any, um, [00:29:00] vision technologies or track and trace initiatives or, or projects that say a vision is working on that you can mention?
Francesco: Yeah, well, um, always speaking about, uh, cutting edge technology.
That is something that, uh, maybe, uh, in a beauty industry will arrive, but in a later stage, um, A line cleaning process is a problem sometimes, especially when you need to guarantee that there is no mix up or a residual product on the line from the previous batches, or in general is very time consuming because you need sometimes to change the machine, a lot of tooling, and the machine have hidden spots where that you cannot reach.
So we have another technology that we Showcase in the next exhibition, but more mostly in the life science, but I'm more than open to explain you because we have a demonstration line in Pavia that we can perform a demonstration that using camera strategically. [00:30:00] placed inside any kind of machine and using, uh, anomaly detection strategy with AI, we can perform in real time, uh, analysis.
If the machine is making broken parts or there are parts in a place of the machine that should not be broken. that are not supposed to be, that the tooling installed is okay, that the registration on the machine are done correctly. Again, it's a tool to help the operator perform a task that is the one for the line clearance.
So we have the kind of technology that could be applied to any type of machine, or even the one that are in exhibition today. Maybe it's something that could be interesting in a beauty world Uh, because batches tend to be in a smaller size, so you need always to be sure that everything is very clean.
And of course, uh, uh, in, in pharmaceutical world, for example, it's much more critical that a product that the, that is not supposed to be, is [00:31:00] remaining from the previous batch. So over there is an healthy issue in here. Instead, I think it's more related to, uh, probably to faster change time.
Deanna: Yes. Oh, that's super cool.
Can you, can you tell me what the device or system looks like? How, what is going in? to do the looking around.
Francesco: Uh, so, um, at the moment, my colleagues are developing different kind of approach. Uh, one approach is to use a stand, uh, standalone camera. So fixed in the inside, the same one that you can imagine, uh, looking at traffic light.
Um, there is a more advanced approach to use a cobot. That is, or an AGV that is moving on the line and performing the task by himself using a robotic arm and perform a deeper inspection. Uh, these are the things that we work daily in order to be ahead of the game a little bit and propose to the market solution that solves the real issue.
Maybe in one year [00:32:00] time, next exhibition, I can tell you about new technology that we are developing.
Deanna: I love it. Well, Francesco, this has been a very informative conversation. I thank you so much for joining me today on the Cosmo Factory podcast.
Francesco: It was a pleasure and, uh, was also my first podcast, so thank you so much for inviting me.
Deanna: Thank you. Thank you so much for listening. If you find the Cosmo Factory podcast useful, please take a moment to leave us a five star review and share your thoughts. So even more cosmetic industry professionals can discover the Cosmo Factory. I'm Deanna Utroske, please join me again next Tuesday for a new episode of the Cosmo Factory podcast.
We'll see you
then. Bye.