Štěpán Kopřiva (Blindspot Solutions): when a person reports being quarantined in the morning, we can reschedule shifts in seconds thanks to artificial intelligence
Where human brains are no longer enough, artificial intelligence will help with the logistics of the pallet. "The algorithm can take into account a huge number of combinations. But we want it to whittle those combinations as quickly as possible, leaving us with only the most efficient ones. We're looking at somewhere between 30 seconds and a minute to calculate for one container," Jakub Novák from the logistics planning department at ŠKODA AUTO tells the ADASTRA podcast about the OPTIKON solution.
Ivana Karhanová: Shipping containers filled to the max. Stacks of papers with calculations and drawings on how to arrange different types of pallets so that not even a mouse slips through. Yet the human brain had reached its imaginary edge, and artificial intelligence had to be involved. And the result? In six months, ŠKODA AUTO was able to ship 151 fewer containers and save 80 tonnes of carbon dioxide emissions. I'll be talking to Jakub Novák from the Logistics Planning Department about how artificial intelligence is being used in logistics at ŠKODA AUTO. Hello.
Ivana Karhanová: Jakub, let's give the audience a brief description of what you needed to solve at ŠKODA. What was the problem?
Jakub Novák: We deal with the shipment of containers to our plants abroad, where they make cars from individual parts. We dealt primarily with Russia, where we send multi-packages, which have a slightly more complex structure and construction, and therefore it is more difficult to fold them into containers. It is more demanding. Over the years, of course, we wanted to get more and more material into the containers. This is, of course, related to costs to environmental protection. Gradually we came up with combinations that helped us fill each container more and more. But of course, after a while, when we had four hundred and five hundred kinds of pallets, and we also had to keep track of how busy each container was, whether there was material for the proper production week, we were no longer able to come up with more combinations to keep moving forward. In short, we had the container so full that we didn't know how to fill it up even more and use the last of what was in it.
Ivana Karhanová: I guess we should say that there are - as you mentioned - several hundred of these types of pallets, and basically, you are mixing fragile goods, you are mixing different types of packaging that must not be damaged by being placed, for example, at the bottom.
Jakub Novák: Exactly.
Ivana Karhanová: You have to fill that container as much as possible and as fast as possible.
Jakub Novák: That's right. It's about figuring out the right combination and getting it in there quickly and accurately. It's about the fact that there are a large number of pallets.
Ivana Karhanová: Just a lot of different kinds.
Jakub Novák: Yes, there's just a lot of them, and they're plastic, wooden, metal... And you have to make sure that the material doesn't damage each other. Some of the parts are very heavy, some are light, sometimes there are chemicals involved, and all of this together makes it difficult for us to work.
Ivana Karhanová: How did you fill the container before? What were your procedures based on?
Jakub Novák: The procedures were based on the fact that we have a lot of experienced workers who have been with us for a long time, and they came up with many of the combinations themselves. New workers watched their more experienced colleagues and learned how to choose the right palette, know if something fits or doesn't fit, and balance it all. So we were doing it purely based on experience and putting our insight and know-how into how to roughly put everything together.
Ivana Karhanová: How long did it take those people to learn how to stack the containers themselves?
Jakub Novák: It's in the order of months.
Ivana Karhanová: I found on your website that you gradually encountered several problems. That at first, you loaded the types of pallets that were most suitable, which makes sense, but when the container filled up quickly, you ended up with exactly those types of pallets and goods that were difficult to combine and didn't fit anywhere. How much more can you fill the container now?
Jakub Novák: Basically, we assume that the container we are sending has a volume of about 78 cubic meters. We initially filled it to 71 before we started looking for some other solutions, mainly concerning costs and the environment. We wanted to fill it to at least 72 because every cubic meter costs us a lot of money, and every cubic meter that remains empty is there for nothing.
Ivana Karhanová: So you have put algorithms and calculations on it. Yet the system learns from what your employees have already figured out. Why didn't you start from scratch? Why are they learning from what you already have?
Jakub Novák: It's a particular thing, and we didn't want to throw away the know-how of our employees who have invented a lot of combinations. And also, to avoid that, for some reason, the algorithm might not be able to come up with these combinations, or we have to teach it in a complicated way. So basically, we ported the existing varieties. At the same time, we asked the program to find us more combinations because there was already a massive bunch of them, dozens, maybe hundreds of pages of combinations. So don't think of those palettes as boxes that you stack up exactly how you want.
Ivana Karhanová: Well, I imagine, and I can see the square container. So it looks pretty easy to me from the outside.
Jakub Novák: It's easy in the case of sending material to India, where the one-way packaging goes. That means we pack the goods in a carton that we don't want to return to. Whereas in Russia, we want those pallets to come back to us because they are specific pallets for molded parts, for example. That means that the pallet looks more like a cage, and it only has four fixed points, both at the top and at the bottom. And those are the only support points that we can use, so we can't put a cardboard box on it, for example, in the middle. So that's why we needed the AI to be a little bit pre-trained and know the combinations we have there, which are somewhat specific. We put three pallets down, then put two pallets on top of them that kind of overlap them, and then we put maybe one pallet on top that overlaps the whole floor plan of what we're building. And the AI probably wouldn't have come up with that on its own because we would have to have each pallet specify whether it's just a solid corner or solid all the way around.
Ivana Karhanová: I'll add that what's going on there is non-linear combinatorial optimization. If we explain - again, this is from your website - this method can calculate the most efficient way to fill the container for shipping. This is because it contains an algorithm that can select the best option from all possible combinations of pallets - usually around 1040 in a single container. Going through all of these combinations is hugely time and computationally intensive, so the algorithm is equipped with logic that efficiently searches among these options for only the ones that lead quickly to the desired result, thus meeting the specified computation time limit. So how long does it take to calculate?
Jakub Novák: The brief was to get it done in a minute. This means that the algorithm, as you say, can take into account a vast number of combinations. But we want it to cut down those combinations as quickly as possible, and we're only left with the most efficient ones. At the same time, the algorithm has to consider that the container it builds is not the only one. And, of course, as you said, it can't have lousy material left over, which is problematic to get into that container. That means that at the moment, we are somewhere between thirty seconds and a minute to calculate for one container.
Ivana Karhanová: How long did it take you at ŠKODA to develop and deploy this solution?
Jakub Novák: It was quite fast. We wanted to fill the containers more and more, so we wanted to develop the solution quickly. For example, ŠKODA AUTO DigiLab helped us find service providers who designed the algorithm and helped us develop and deploy the application. From the first idea to implementation, it took us a year. After that year, we deployed the first version of the application, which of course, was still being tweaked in some way because one thing was that we did a proof-of-concept, where we debugged whether the artificial intelligence can sort the stacks better than we can sort them at the moment, which was a criterion. In short, we had an overview of how we loaded the container ourselves, and then we compared how the app would handle that volume of material. We started to develop when we got to the point that AI could help us solve that. And as I said, we started with a concept at the beginning of the year and deployed the first version at the end of the year.
Ivana Karhanová: ŠKODA is digitizing quite a lot. Still, have you encountered anything during the development process that you hadn't counted on or that surprised you?
Jakub Novák: Basically, it was mainly IT security because we wanted the solution to be cloud-based so that it could be accessed from anywhere. We use it on 10-inch tablets that connect to the internet, and the app has a web interface. That means we want to connect from anywhere, and we don't want to install it on that device because that would be extra work for us again. That was a considerable specificity because back in the development period, the vendor was already running the application on their cloud-first so that we could get to it. Only then did they move the application to our cloud, which involves quite complex issues in terms of IT security, firewalls to get us where we need to go, but nothing that shouldn't be getting to us.
Ivana Karhanová: On the other hand, I suppose equally because of IT security, it's impossible to develop the application directly in your cloud.
Jakub Novák: I suppose it would be theoretically possible, but it would delay the initial development phase considerably because first, we would have to create the cloud environment at home, and only then could the supplier start developing. This way, we agreed that he would begin to development in-house. In the meantime, our cloud environment would be prepared, and then the solution would be transferred.
Ivana Karhanová: Can those who send you containers to ŠKODA use the system if it's running in the cloud? Or is that not how it was intended at all?
Jakub Novák: They could, but at the moment, did not intend to be like that.
Ivana Karhanová: With digitalization and artificial intelligence, there is often a specific resistance or perhaps distrust of the people who are supposed to work with the results. How did your people who load containers react when you told them that you needed help with the algorithm? Did they take it in stride, or did you have to think through some concepts of working with them?
Jakub Novák: I'll answer with a bit of a rhetorical question. Imagine if I put side wheels on your bike. Would you like to have them there?
Ivana Karhanová: Like, maybe in some corners, they wouldn't be wrong, but...
Jakub Novák: Don't take this pallet, take this one, because then the other containers will work out better. So, of course, this was a bit tricky to work with, and at the same time, I have to say that we were also working under time pressure - not only in terms of the deployment itself but also in testing because there was no room to test outside. So in short, it had to be tested on-site. So that means that it was very challenging for those people at that point to keep up with their work and, at the same time to be able to help us with the testing and somehow make it all come together.
Ivana Karhanová: What made them finally, and now in a good way, accept the solution? Well, I suppose they eventually took it as their own...
Jakub Novák: It brings benefits, which will show up in things that can happen even to the experienced ones. That means typically unbalancing the container.
Ivana Karhanová: On the one hand, it's more complicated.
Jakub Novák: Exactly. Of course, when you load it on the ground, you don't know it. But then, when we take it with the stacker and hand it over to the railway carriage that takes it to Russia, it is already on the stacker that we find out that it is poorly balanced because it will swing on the stacker. And if we put such a container on a railway wagon, the railway would refuse to let us do it because it would wear the wagon unevenly, and there would be a risk of damage. So I think that is also a consideration that maybe helps our workers to accept that solution.
Ivana Karhanova: When we talked about the fact that it takes quite a lot of work for new people to learn the processes, this is now falling away.
Jakub Novák: And it's much better because we also have feedback from new people who can drive a forklift, which is a primary criterion. But it's one thing to drive a forklift, and it's another thing to choose the correct setup from all those pallets. And that, of course, is a problem for the people who come to us at that point. So now, when they get a tool that tells them: Take this pallet and this pallet, they know how to grab it and how to load it. And thanks to us, they also have the know-how to take it.
Ivana Karhanová: What was the initial impetus for the demand for such a solution in the first place?
Jakub Novák: It was the will to reduce costs and maximize space use. As part of the green logistics strategy that ŠKODA supports, we wanted to make the most of the container's potential. That means not carrying air in it but carrying material in it. By deploying this solution, we could save five trainsets in 2020 that did not have to leave, which is a considerable amount of carbon dioxide. It is around 160 tonnes, and I think that is undoubtedly one of the new solution's excellent results.
Ivana Karhanová: You said that from the 78 cubic meters, you were able to load about 71 cubic meters before. That means some 90 percent. Then you were not able to figure out how to translate it differently? Or what prevented you from just filling the other meters?
Jakub Novák: It's a combination of two factors. One is that the palettes are more and more diverse because new projects are coming up. For example, we're producing a new Octavia in Russia or another new model, and as a result, the palette in which the parts are placed might be slightly different. And this happens periodically. That's one of the things, and the other is time because we have to load a certain number of containers per shift at the same time so that we don't stop production in the foreign factories, figuratively speaking, so that everything gets there when it's supposed to get there. And that's why there's not so much time during loading to think about whether we should take out the pallet that we've already put in and try to put it in a little bit differently or go and get another one. So, in short, we also need to maintain speed.
Ivana Karhanová: And did you run into anything, for example, on the data side or the process side when you were implementing all this - besides security?
Jakub Novák: We ran into that. I think that the first important thing was the input data, that is, the entire status of the warehouse of the dispatch area - the pallet data, the dimensions, the weight, what material is there, when the material is to be exported, and so on - is stored in the pricing subsystem, and as soon as we need to receive something from the SAP system, which is not only for ŠKODA AUTO but for the entire Volkswagen Group, then, of course, more and more things enter into it, just in terms of security, but also in terms of the performance of those systems. Because if we want to receive data periodically, for example, or if we're going to receive it online in our application, that puts a strain on the system. And not only for ŠKODA AUTO, but because it is a group system for all brands. So this means that we had to work very hard with our colleagues to make sure that we were getting the correct data at the right time and at the same time not overburdening the system that everyone needed to use.
Ivana Karhanová: So you end up pulling them? How?
Jakub Novák: We initially started by exporting once every 15 minutes, which in the early stages of testing proved to be sufficient. But we knew that for a production deployment, that's a problem because when you need to schedule a container at a certain point, it's hard to tell: Wait five more minutes for the actual data to load. So the way we're doing it now is that the data comes in about three to five minutes, and we get it in a way that our SAP system talks to the SAP system in Germany.
Ivana Karhanová: Bude OPTIKON, jak se systém ve Škodovce jmenuje, zanesen i do dalších procesů, další práce? Nebo jaké s ním vůbec máte plány, když už je vlastně nasazen a funguje?
Jakub Novák: Byli jsme osloveni kolegy z jiných značek. Zajímají se o naše řešení, protože nejen ŠKODA AUTO posílá CKD, tedy rozložené vozy, do zahraničních závodů a odesílá je, ale i ostatní značky koncernu, které se potýkají se stejnými problémy. Takže toto je pro ně způsob, jak si v tomto ohledu pomoci. V podstatě na tom s nimi spolupracujeme a zkoumáme možnosti nasazení. Zároveň máme možnost nasadit toto řešení i pro prázdné palety, protože řešíme i jejich oběh, samozřejmě buď v rámci závodu, nebo v rámci toku palet od nás k dodavatelům prázdných palet, které oni zase plní materiálem pro nás. Zatím bychom ale asi chtěli, aby OPTIKON využívala především společnost ŠKODA AUTO a koncernové značky, takže jej externím partnerům k dispozici nedáváme.
Ivana Karhanová: Jakub Novák z oddělení Plánování logistiky nám vypráví o optimalizaci logistiky ve ŠKODA AUTO. Děkujeme za zajímavé povídání, že jste přišel do dnešního podcastu. Tak zase někdy na viděnou.
Jakub Novák: Děkuji za pozvání.