Boost Manufacturing Agility and Flexibility with SAP HANA

Boost Manufacturing Agility and Flexibility with SAP HANA

Webinar On-Demand

Meet the Experts

Digital transformation is a goal that many manufacturers have been working toward as a means of optimizing operations and improving manufacturing productivity. But the last year has increased the urgency of this initiative as it has become clear to manufacturers that digital transformation can help maintain business continuity and increase agility when introducing new products, address supply chain constraints, and meet fluctuating demand. But what is required to achieve this agility and optimize plant operations, supply chain management, and pricing and warranty planning? And how can SAP HANA support these goals?

Hear from SAPinsider’s own Senior Analyst, Robert Holland, as he engages an A-Team of thought leaders in this engaging ‘fireside chat’ about the state manufacturing agility and flexibility with SAP HANA.  Guests include Sebastien Boria (Computing Architect, Airbus), Dr. Tom Bradicich (VP, HPE), and David Austin (Sr Engineer, Intel).

View this webinar to:

  • Learn more about these use cases and their criticality in manufacturing today
  • Understand how a flexible infrastructure can help free up investment for your core business
  • Explore how SAP HANA can help boost manufacturing agility
  • Discover how this environment can help accelerate your transition to SAP S/4HANA



Read the webinar transcript below.


We say there are three zeros in manufacturing. There’s zero downtime, zero quality issues, and zero harm to the environment. And manufacturers really need zero downtime if they’re going to be facing demand. Everyone wants to ship high quality products. And what’s very much in discussion today, and certainly it was a feature in SAP’s. Christian Klein’s keynote at Sapphire this year is sustainability and environmental governance, but it’s also cultural, social, and morale harm to the company. So, Sebastian, how is Airbus addressing these challenges and solving the goals that those challenges represent? 


So, what you need to understand from our industry is that we are governed by change and modification all along the processes because we have even huge number of aircraft that we are building, not mentioning drone satellites and so on. At the end of the day, we are far from having a mass production system. On the other end we have high quality product, meaning that to some extent they cost a lot during each production stage. So, we can’t imagine to have defects that goes to the decision of crushing them. We need by any means to avoid having these kind of determinants. So, we are more in governing the process flow changes to behavior, way of working, meaning that the next step is never well known at 100% because it depends on the tree and the characteristics that have been produced before. Which means that to some extent each aircraft, each system, each products are pretty unique in their process flow and the way operators operate on top of the product. And this leads to very strict problematics of computational power that we need to put at the edge because when we go to that direction, that means the system has to evolve in front of the viability that we could have trying to follow that process. 


It’s not as rigid as for example automotive industry or mass markets. We are speaking about cheap industry on that one. We are at the level where we can’t imagine to go in that direction. The main challenge that we have is to be sure that at the end the environment that we perceive from a system is the one that is still manageable and for that we need a bunch of sensors, we need a bunch of computational power, but we need on top mass scheduler something that could help to synchronize actions and to verify eventually quality at each step to stabilize it. That’s why when I discussed the first year with that term, I completely subscribed to one of the rules to answer the 30 which was a free. C is basically compute, communicate and control.  


And this is the most important thing that we have to do. We need to deliver. At the end of the day, we don’t just need to have good products, we need to deliver them on time and quality affecting the environment, that’s for sure. 


I pick up on that because it’s close to my heart. As Sebastian said, the three C’s, as he pointed out, connect compute and control and the application that I was excited to work on many years ago, actually, that prompted that triad of terminology here. The three C’s had to do with a wonderful application going on where there was visual recognition of some of the holes that are in the body of the aircraft that have to be filled with rivets or with bolts as well. But if you apply those three C’s And you have the thing in the IoT very near to David’s heart here, the T is things right in the IoT. If the thing is the Rivet and the tool and it’s connected to the internet, that’s IoT and you’re able to then more efficiently select the right part as opposed to relying on human memory or human observation. But the computer system is helping with the visualization is taking place. So as that connectivity takes place, there’s another type of connectivity, the first which is to the supply chain. So, imagine one of the potential use cases here is if the part breaks, it is immediately connected to the supply chain and can alert the supplier that, hey, these bolts are beginning to fail in this particular application. 


And are you seeing that in your manufacturing site so that connectivity takes place there only because the computation took place that decided what was being viewed and being worked on? And then the third C control is, well, let’s stop working on it or let’s get new parts or let’s order new things and take control quickly. Now contrast that to a manual approach to this problem where by something breaks, something stops, and you ask what happened and say, well, when Susie was working on it, it broke. Well, where is she? Well, she’s out to lunch and you’re losing all this time or she’s on vacation. We have to wait till they come back to find out what broke, when and how things can be repaired better with the supply chain. And when you have that SAP software, you have the compute power of Intel Coupled with Hewlett Packard enterprise, you can have this end to end, connect compute and control system out at the edge that would enhance the productivity and drive that downtime to zero, which is one of the three zeros. 


Yeah. And as Sebastian was talking about some of the quality thinking and systems, it struck me that Airbus is working on the manufacturing scale of probably, maybe nothing bigger on the planet. Large scale planes feature sizes on the order of meters. At intel, we work on the order of nanometers in some of our processes, but the similarities and the quality systems. 


Very analogous. We have a manufacturing philosophy at intel that says you can’t inspect quality into a product, which means just by looking at your problems, looking at your systems, that doesn’t fix anything. What fixes things is when you have a tie back into your vehicle control systems, your quality systems to drive, the defectivity to drive the root causes down to zero. Only then can you really have a truly streamlined manufacturing process. One interesting that happens when you put these types of quality systems into your manufacturing process, your yield actually goes down, which is not a desirable thing. Most factory managers that I know of don’t like to invest money into something that’s going to drop their yield. But what it’s doing is actually exposing where the problems are. And then, of course, the challenge is, well, how do we take that knowledge? How do we take that learning? How do we take that information, that information and drive back through to the root cause of our manufacturing process so we can get down to the zeros that we’re talking about here? So, I really like what Sebastian is talking about that resonates, whether you’re talking on the order of meters or nanometers. 


Well, it sounds like a lot of what we’ve been talking about here is not just problems that are necessarily challenges that Edge Bells is addressing, but the challenges that are facing high tech manufacturers in general today, whether, as you say, you’re on the scale of meters and some of the biggest deliverables that you can possibly have in an A. 380 or something like that, versus an intel chip, which is obviously much smaller in scale. Are there any other challenges apart from the ones that we sort of discussed that might be facing high tech manufacturers today, Sebastian, and your interactions with other organizations that you work with or within Airbus that you haven’t mentioned previously? 


Yeah, for sure. It is directly linked to the ecosystem that we need to put in place in order to solve the problem that has been mentioned already. If you imagine that you have all these problems to tackle, what you need for sure is a digital chain that is able to do so. And in order to do so, even if you have the best software on the planet, what you need to be sure is that at the end of the day, you can integrate them. So, one of the main parts of the difficulties after finding a way to assemble the right bricks is to be sure that the right brick can be assembled. It means that, for example, when we go to interaction between hardware and software or software to software, we need to be sure that we have the level of openness that we are looking for, meaning we have the right relationships with the right supplier at the right time on one side and on the technical one that we are sure that we have a decent level of documentation and APIs in order to be able to connect systems. Nowadays everything is digitally connected. 


We need to take care about that. We can’t imagine to have Silos in time of Square that works perfectly. But in the Navali tower we need to be sure that everything is connected. And that’s why I was thinking about adding Edge doing the task. But connected to NES and ERP, we are using SAP also for sure regarding that task, and it is not an easy task to be sure that at the end of the Apr anyway, it is one of the big challenges that lead us in the supply chain between the make or buy strategies that we have to do, especially in software environment. We need to be sure that at the end everything that we buy will be glued to something that we could make and vice versa if we take back our functions about the classes and all the systems that we have just discussed regarding the pattern or condition of the home. One of the main challenges we have to face so far was to be sure that it will be constant, meaning that electronics evolve far more than aircraft. And at the end of the day, we don’t want to be disrupted because of the six month challenges that face the supplier of the chip. 


On the other end, we want to be sure that next generation of chip, next generation of plastic, and next generation of software will still be interoperable with the legacy system. We are building aircraft for 50 years, not one or two. So, we need to be sure that our industry system can cope with that too. 


Robert, you mentioned something, as I think about what pardon me, Sebastian said and what you said, and you noted that David comes from a world at intel at the Nano meter level. Sebastian comes from a world at grandiose level. I mean, have you even seen a jet engine, which is not the whole aircraft? That’s huge. And I come from the manufacturing world of in between per se, where we’re working on boxes all the way from the size of a shoe box to a pizza box to the size of a refrigerator as well. And the challenges are very similar. Not identical, but very similar. We in our world of manufacturing systems, for example, servers and storage had a problem driving our defects down to zero that are out in the field. We don’t want product to escape out in the field. Obviously, that is defective. No manufacturer wants that. It is costly. On the warranty side, it also damages brand and customer satisfaction. So, we employed a visualization and it’s fascinating. I know David appreciates this too, because with AI algorithms, with cameras looking at the product as it’s going by the conveyor belt, it can be inspected for quality issues much more thoroughly and much faster than the human eye. 


And applying that to boxes, we’ve also applied it to printed circuit boards with Seagate manufacturing, which I know makes storage and hard drives. And that capability is quantifiably, dropping inspection time from minutes to seconds. Also, with the AI, it’s ramping up the ability to know what is an error because the human has to go to class and train. Oh, this means an error. This means not. Whereas the AI functionality allows that to happen very quickly and then defects are dropped. We’re measuring as high quantitatively as 25% less happened immediately because of this application of video and quality assurance. It’s absolutely fascinating to see this. If a picture is worth a thousand words, then a video is worth a billion words, and we can have that capability applied to the quality of all of our products, all the way from the lowest level chips to the highest size things like jet engines and aircraft. 


Absolutely. Would you have a similar experience, David? 


Yeah, I’d say so. I’ll tie together a couple of ideas that we’re talking about. Robert, as you’re talking about, what are some of these manufacturing challenges that we deal with? And Tom’s, bringing in these advanced analytic workloads like AI, there’s two things that stand out to me as real challenges that we have to solve as an industry. One is what I would describe as this lab to Fab scale challenge where we have a problem. We have an inspection system that we want to put in place so we can get it working on a small scale on a lab bench somewhere as a proof of concept. But then to scale it out into a manufacturing environment, let’s say on the scale of an Airbus, that’s a big challenge for AI to solve because there’s so many variables that come into play when you put it in a live manufacturing environment versus what a small data science team may be able to do on a smaller scale. So that’s one that we’ve got to figure out how to overcome to really allow the true benefit of AI to come to life. And the other is that this one is kind of near and dear to my heart is with these advanced analytic workloads that we’re talking about here, largely those are in the hands of the data scientist today. 


And while this pains me to say a bit, as a data scientist, we have to start removing data scientists from the loop of every one of these solutions. My vision is subject matter experts. Right. The manufacturing expert who’s been on the Airbus factory floor for 10, 15, 20 years, who really knows and understands the systems they can start putting together these AI and advanced solutions so you don’t have to have these data scientists who are scarce and expensive putting together those types of solutions that can land in a useful environment. So those are two problems that I think if we can overcome those, we’re ready to take the next step function up in our ability to bring these advanced analytics solutions to life. 


Sorry, Sebastian, was I interrupting you there? 


I was just seeking to validate that the use of AI is everywhere now. It’d be great to hear Sebastian’s point of view on that as well. It’s not a new concept. Obviously. It’s a term from maybe the. But innovation is either something new or something nobody remembers. But the point is the advances in the algorithms and the learning combined with the compute power, because recompiling machine learning models is very compute intensive. So, they will enjoy that from the intel company combined with what we do with our high performance computing servers, it’s a beautiful match to promote AI in almost every dimension of manufacturing. 


In fact, I think we have a follow up question that Director Sebastian little later on, specifically on AI and manufacturing. But let me just sort of talk a little bit about you can just continue with what we’re sort of talking about more with challenges before we sort of move on to some of the technology solutions that Airbus is potentially using to address some of these challenges. But I know one of the things that Sebastian mentioned is integration, and it doesn’t matter whether it’s with the software you’re using to make sure that it all connects and works together. But it very much has to connect to a process and something that Airbus is going to be delivering. So, they don’t want to invest in anything. Now, just as important from an integration standpoint is supply chain. We’ve seen very much over the last year that supply chains have been disrupted. They’re no longer integrated. Organizations have had to look in many different places to address supply chain. What are the supply chain challenges that Airbus faces that other organizations can face, particularly around the topic of make versus buy? I think that was something that Tom and I had talked about before this conversation. 


So, Sebastian, what are you seeing? 


I would say there is a product by itself which is challenging regarding this supply chain problematic, that’s for sure. But there is another one which is less visible but indeed very complex to tackle. When we speak about manufacturing operations, it is the quality of the system that handles the industrial system on its own, meaning that as I started to mention a few minutes ago, we need to be sure that at the end of the day we are able to control the systems and it is very important to know where we are, where we are standing in front of viability of the process versus the process as it has been designed within the tool that we are using. Let me give you an example to make it more concrete. Sometimes we have the best first class citizen software in order to manage a particular part of the process. But we need to be sure that at the end we are able to interface with the quality system that has to go with this measurement and so on. In order to be able to do that, we need to be sure that we are not inventing the wheel. 


It’s not an option. We can’t lose that time. We need to be sure that at the end of the day the suppliers that will provide you as tools or software will be on time, because if not, it is as dramatic as a part which is coming late to assemble the aircraft. Not having the right tool will lead to infringing the freezer targets that we were just mentioning. And on top of that, we need to be sure that we are able to control the future of something that we buy. Therefore, the strategy that we adopted inside the tool and processes Department in the company was to be sure that every time we buy things, we are able to glue them with make things, meaning that to some extent, again, the problem of API and it comes into the equation, we can’t only buy, we can’t only make it’s a kind of equilibrium. It’s a momentum between both approaches. 


Fantastic. Now, Tom, I know you’ve at HPE have helped many organizations face some of these and solve some of these supply chain challenges. What have you been seeing over the last twelve to 18 months, particularly that you can talk about? 


I think few times in history have the supply chain challenge has been as severe with respect to what has happened with the Pandemic as well and being able to plan the ability for intercepting as the parts and the sub systems arrive as well, such that you can create the product and then hit an announcer a launch date. In Sebastian’s case, it would be presenting the finished aircraft to the customer, in David’s case, the technology to the customer. In David’s case, it’s very high volume as well. Now at the end of the day, though, there’s another thing I like to talk about that is very important. It has to do with a pricing strategy and a warranty strategy. How does that connect to the supply chain? Well, as you approach the time you have to put a price on your product and many times it can be done beforehand or at the point of announcement or you’re seeking to take a pricing action, you have to figure out what your costs are. Well, sometimes the costs are related to obviously the availability of the supply and many of them are price elastic with respect to volume. 


In other words, if you can’t get it, it’s usually more expensive as well, but also the warranty analysis because that can be very expensive. If one has many warranty claims, it can be extremely expensive. So therefore, replacement parts combined with the main production parts are very key in the analysis. And if you think about it, one has to go through many scenarios. I’m going to tie it back to high performance computing because the problem, as you’re asking Robert, is how do I know? I’ve analyzed all the different variables of when the supply arrives, is it of the right quality? How much do I have to send back? When is it delayed? What is the cost? Because now it’s in short supply as well. And analyzing all those variables, if you think through it, you have to change one and hold the others, change this one and hold the others, and then your launch date or your delivery date comes out, and then you have to fulfill different types of backup supplies that have to be stocked in different geographic locations. All that has to be computed. And if one has more high performance computing systems to do it, one can actually get the right pricing and warranty strategy prior to the launch of the product because there’s a little bit of a horse race going on. 


You don’t have forever to figure out what your pricing is as well. When you have to deliver, especially when you’re delivering volume and you’re bidding even in more real time, like David’s company would where Sebastian is, usually a lot of pre ordering takes place. There’s a little more stability there and what the pricing and warranty would be. But with David, when you’re seeking to get design wins. And correct me if I’m wrong, David, but you could be surprised with a very large order because your sales team has done a great job working with your engineers. And then to be able to revamp that, that’s where computing systems come involved connected to the supply chain. It’s extremely important because we’re all suppliers here and we must in order to survive, protect profit, obviously. 




And make money. That’s the goal of the Corporation. But at the same time, give our customers the best opportunity and the best service and the best deals. And that’s a balance. And there’s things in calculus that have to do with minimaxes and all types of mathematics that go around in the high performance computing systems that serve. That one sliver of the process, Robert, that you asked about, Robert. 


I think I know, David, that you obviously have had challenges with supply chain at intel over the last year. I think most people, even those who aren’t particularly following the high tech sector, have heard about chip shortages impacting a lot of different things. How is intel managing some of these supply chain challenges and what are some of the ones that you’re facing that other organizations might face? 


Yeah, a lot of those challenges come down to difficulties with the forecasting. The supply and demand of both raw materials, like unexpected orders, come in large ones that can disrupt the whole planning part of the supply chain. So, these are things that we’ve been dealing with for many decades, and they do tend well towards advanced analytic solutions to help solve and model some of those. I’d say the current time, though, there’s never been more variation and variability in the system and that’s leading to some challenges and our ability to forecast. But I’ll also say this is where AI can also come into play. Ai forecasting systems can far exceed both what humans can do and what some of our previous analytics solutions can do. So, we’re starting to apply AI actually to some of these supply chain challenges. Ai isn’t just for predicting what Netflix video that you’re going to watch next or how to inspect this part for anomalies or defects. So, all of these solutions really do kind of tied together in some way. That the world is trending towards more advanced analytics and AI and supply chain is another challenge that we can solve there. 


Absolutely. Now I think we’ve talked quite a lot about the challenges that organizations can face. Let’s talk a little bit about some of the technology solutions. I know AI has come up a couple of times. Sebastian has mentioned that. Tom has mentioned that. You just mentioned it again. David, what role does AI play in manufacturing at Airbus, Sebastian? And how is it solving some of the problems that you’ve faced? 


So, to our mind, we strongly believe that AI will be the two days to solve the problem that we are not able to solve a few decades before us, but not in a way that it is a systematic answer to everything, that to some extent we don’t have enough data for everything. It’s where we divert from mass production industry. The quantity of that available is not as huge as it should be. And this is pretty interesting because then what we need to have when we are speaking about data analytics is to be sure that we are tuning it in a way that it could become frugal. So, there is a part of algorithm which is well known, everything which is related to supervised machine learning, which is now standard, I would say in manufacturing for a huge number of tasks, for example, locating the center of the home while analyzing the contour of the parts that have been drilled. This is very standard, and a lot of people use that all day long. But when we go to quality with limited quantity of data, some supervised learning, reinforced learning or perhaps even unsupervised or energy based machine learning will help us in the future. 


And this is areas where we have colleagues that are specialists about this domain. And we need to be sure that at the end of the day we are covered regarding our industrial system. Later on, after the first boom of AI links to this machine learning thing, I believe that we will have to work on infrastructure and architecture of AI algorithm that will fit in that domain. Now, most of the people I am working with are joining analytics, standard data analytics with behavioral modeling in order to create a kind of trade between both type of algorithms. This leads us to have perhaps a good solution in between the two world. So, the standard data analytic world and advance AI techniques link behavior. And this is pretty important because when we have to manage on the first side, but also people that do not have all the answers to supervise the results of the algorithm, we need to have other views and other ways to do things. So, there are a lot of friends regarding AI, which is a fantastic area to grow even internally inside the company. We need for sure to invest more on that, not to forget something which I always say to my guys, all models are useful, some are wrong. 


You know, to some extent it is very important to be sure that keep the sentence in the opposite meaning it was telling, which works also to see the optimistic way of doing it for sure. And what is important with this sentence is that to some extent, if you consider your industrial system in the face of having AI topic on top of it, you need to be sure that you are taking the right KPIs. You need to be sure that you have the right computation power. So, there is no magic, one with one formula and one algorithm that will fit all everything is a kind of giant ecosystem where unit partners that are at the age shoot first class citizen into their tech to be sure that you assemble the right technology, among others. It’s also right for AI. And. 


I think we might be having some minor bandwidth issues there, Sebastian, but I definitely think we got the gist of what you were talking about. Tom, you had talked about analytics AI before. What role are you seeing at play in solving some of these technology challenges that people face? 


It’s quite fascinating, I think, for your question about what role it’s interesting to think about AI. I believe it has achieved celebrity status today and it’s everywhere. And I don’t mean that frivolously because it’s quite valuable and it should. It’s everywhere. But I might liken it metaphorically to the cranberry. And as Brian Reagan would say, the cranberry is everywhere. It’s obviously in cranberry juice, but it’s also in grape juice. It’s in Apple juice, it’s in pineapple juice. The cranberry is in cereal, the cranberry is in little cakes, the cranberry is in liquor, the cranberry is in gum. The cranberry is everywhere. So, AI is everywhere and should be because it’s extremely valuable. But from its role point of view, I think it’s instructive to bifurcate the dimensions of AI. And I like to call one dimension of AI infrastructure AI. And that’s very important to a company like mine because we provide infrastructure, networks, servers and storage. And one can apply AI, and we do in our info site and introspect offerings that actually analyze to find failures, to predict failures. That’s a big dimension in AI is the probability dimension that can predict things and also to learn behaviors and also to work with cyber security. 


So, to secure your networks, to ensure the data integrity on the hard drives or the solid state storage, as well as the compute capability AI can be applied there. Now, you notice that it has nothing to do with directly with the second dimension called mission AI. And that’s where Sebastian cares about. How does the AI apply to my mission? For example, I may have stresses associated with the wings of the aircraft, and I have to use AI to analyze that. I may need AI, like I talked about in my supply chain to make predictions on what pricing would be and what consumer demand would be. I may need AI to analyze the video images to find either issues that are related to quality assurance. And that’s in the domain of the mission because that’s the reason why Sebastian would deploy. So, they both have roles, they both have significant roles, and many of the algorithms and the applications are applicable and repeatable, especially some of the fundamentals with respect to learning and the fundamentals with Bayesian inference and the ability to predict based on limited amounts of data. But of course, those predictions become surer with more and more data collected as well. 


So, I see it in two ways to sum up really in the mission and how it’s going to apply to manufacturing the product and the processes. And then secondly, that underlying infrastructure which must be reliable, because if it fails, then we don’t hit the zero downtime that we’re after as well. And if it fails also, it can be harmful to perhaps the environment or two people safety. So, it hits all those zeros. 


Yeah, I would agree with that, Tom. And I would say AI is a tool, and like any tool, you have to apply the right tool at the right place at the right time for it to be effective. And we have to temper our enthusiasm a little bit and really look at it in that lens and not just think AI is a tool to solve any problem, but where it’s solving really important problems. In manufacturing today, I see with the partners that I work with is where we can replace repetitive human tasks with AI. We really can do something special. So, one of those is, of course, replacing eyeballs with cameras. This is a great example of a repetitive human task. That’s a strain on the eyes. It’s not very exciting or valuable for the workers in a lot of cases. And quite frankly, AI can do a much better job on some of these tasks. So, I think that’s a very good use of AI where we’re advancing. And Sebastian, I think there’s something very important is the thinking here is starting to evolve and change and being what I would call much more manufacturing friendly as we’re using more advanced systems like unsupervised or semi supervised learning where we can train these systems, only looking at what say good looks like rather than giving these supervised AI systems, what they traditionally like is give me equal number of good and bad and of this defect then and that defect, Ben, that’s not a very manufacturing friendly solution in a lot of cases and we can do a lot more with these new types of learning. 


So, I would really encourage people to start thinking outside of what traditional AI has looked like in the past. And some of these more advanced techniques I think are actually much more suitable to manufacturing. And this is why I’m saying right tool, right place at the right time. We really have to be smart about how we apply these technologies and if we are there’s just almost unlimited benefit that we can receive from them, that’s for sure. 


I could not agree more. This is very important and one of the main challenges that we will have to face with these advanced techniques anyway in the future would be transferring within AI will be the right challenge that we need to face, and we need to unlock it. Even from the academic perspective. It’s not yet fully done. Profitability regarding algorithm will be there and we need to certify everything. The certifiable ENS will be the key assets that we need to deploy at manufacturing level and even at product level. That’s for sure. 


Absolutely. Now I know that’s something that I think we’ve sort of talked about as part of I can’t speak we’ve talked about as part of AI as to a certain extent is that of video analytics. And I sort of want to ask this question because I think it follows on well from the conversation and particularly the answer that David just gave, what role does and will video analytics play in manufacturing? Are you using video analytics connected to AI, Sebastian? 


Yeah, for sure. Inside machines we are using advanced math techniques, whatever they are directly related to AI or more known with awful mathematical names that are quite consistent. Anyway, the most usage that we have which is completely implicit will be the location of all for example for that all that act as reference in order to be able to drink the others. We are also using exclusion kind of algorithms that enable us to understand the quality of all, especially in sandwich of materials and for sure even on the tools themselves, we are using AI in order to recalibrate tools on the fly thanks to the best optimal trade off that we could find. At the end of the day, I think this is where it’s very important. We need to find solution out of the box that fits the answer. Ai, I like the idea of AI is just a tool. It’s a wonderful tool, but it’s just a tool, and it is one we are for sure using in analytics regarding paternal condition video in various applications. 


I can’t contain myself. I have to elaborate on that application that Sebastian was talking about. It’s absolutely fascinating. Picture a handheld tool like a wrench or a ratchet, a handheld mechanical tool, but it’s smart, meaning it is connected to a network. So, it’s a thing in the IoT. But using AI, depending on what the purpose of that immediate at the moment use of that tool is, the mechanism can be calibrated or measured. So, there are hundreds and hundreds of tools and hundreds of different calibrations and measurements. Some are physical size, others are torque, which is the amount of stress on the tool. But imagine being able to have the computer intelligence do it for you, as opposed to the human trying to figure out, should I put it on this? Should I turn this switch? Should I move it this way? But it’s automatically happening as well. And the efficiency, the reduction of error and the reduction of time to actually use the tools in putting rivets and bolts and nuts together in an aircraft is absolutely fascinating. And many times, a human has to still do it because it’s a little difficult to get a robotic arm in some of the tighter spaces within a particular product, such as an aircraft. 


So, the dimension of AI as a tool applied to handheld tools, to me, is among one of the most fascinating in manufacturing. 


Yeah, I totally agree. And there’s almost limitless use cases that we talk about here with video analytics and the role of AI. Something that I spend a lot of time thinking about is AI in manufacturing from the standpoint of how do we scale it, how do we go faster in manufacturing? We have what I would describe as this long tail problem where every use case is somewhat unique. And if you understand AI systems, you understand that AI systems are very good at learning and telling us about a certain data space. But once we get outside of that data space, they don’t generalize very well. So, we have this problem where almost every solution needs to have its own training system, its own learning system, its own inferencing system. So how do we make that more general? It is a problem that I work on quite a bit. I could go get scalable compute from HPE today. We’ve solved that part of the scalability problem. It’s really more around the algorithms and solutions. But if we can crack that problem, and I truly believe that some of these new learning systems like Sebastian is talking about or one of the ways that we’re going to do that, then we’re really going to start to realize the true benefit. 


Absolutely. Now, I guess something we’ve talked a lot about AI. We’ve talked about video analytics, and what sort of role the simulations and analysis play in manufacturing. I think that’s just as important. There’s a few other topics here we can talk about in the last few minutes before we wrap up. Sebastian, how is Airbus using simulations, simulations and analysis. 


Simulation and analysis, numerical analysis at a lot of stage of the product of the manufacturing design and so on and so forth. Here I think we can put the Joshua sentence in the right order. Meaning all models are wrong, some are useful, meaning that for analysis simulation helps us to prepare to reduce the potentiality of scenarios that could be weird. Anyway, nothing replaced reality. We don’t have an exhaustive view of all the simulations that could be done. So again, it’s a play between simulating a lot of scenarios to avoid us losing our time to practice them in reality. But at the end of the day, it’s condemned them with reality. Then we need to have a proof of concepts that are completely valid into real world in order to validate the assumptions that we made at the simulation. We are not trusting the simulation enough to say oh yeah, because it’s green. In terms of simulation, everything is valid for sure. We need to consider that it will drastically reduce the assumptions that we need to take, which is already a benefit. That’s for sure. 


What about robotics? I don’t know that we’ve really talked about robotics that much so far in what we’ve covered. Do you see that as a technology solution that Airbus is using and leveraging and what role does it play for you? 


Yeah. So, this is where it’s a bit funny. We are a company which is managing production of the aircraft mainly by end. We have a lot of machines for sure, but a huge part of the assembly faces are realized through tools. On some processes it can go up to 80% doing manually for sure. We still have a robotics that’s for sure entering into the game because more and more that’s clear regarding the level of quality that we want to achieve versus time to deliver. So, automation the first rule of automation, like supervised learning system. At the end, if you don’t know your process, you will fail. So, in order to be able to succeed in a robotic story, you need first to master your process in the way that you will be able to design the right robotic application. And here again the capability of processing data sent to AI enables part of the robotic process for perception, which is one of the biggest problems that we have to face in a non-static environment or environment. As I mentioned at the very beginning is linked to linked to modification of the process on the flow. 


So, we need robots that have this capability of adopting this kind of behavior, which is completely different from the production system that is usually on the market where they are performing the tasks repeatedly in the exact same way. So, the planning is fixed and just re executed. So, at the end of the day, robotics will be the component of all the technology we have just described. We need computation, we need communication. We need perception. We need capability of analysis. We need an AI on top and good models in order to make it real. Without all of these ingredients perfectly managed, it will probably fade. So, robotics is a trend by itself. 


All right. Well, it’s been a fantastic conversation, gentlemen. Thank you, Sebastian, Tom, David, for all your insights here. I think we’re going to pivot to a Q and a portion now. So what we’ll do is, if you have a question, just as a reminder, please type it into the Q and a text box and hit submit. And we would love to answer your question. 


All right. That was fantastic. Thank you, gentlemen, for your conversation here. We do have a couple of questions that have come in. We have a few minutes here where we can potentially address those questions. Why don’t we start with just a general question here? 


What are you kind of seeing? 


Maybe we’ll start with you, Tom and David, what are you seeing as the customer trends in It infrastructures that are sort of deployed in one or more of three places, the remote cloud, on premise data center or at the edge? What are you sort of seeing happening with that, particularly as it sort of connects to manufacturing? 


Yes. Hi, it’s Tom Bradish here from HPE. And thanks for listening to our webinar here. Yes, please. I’ll go first and then David, I’m sure we can comment. You talked about a trifurcation. Effectively, three domains. There’s that remote cloud, and then there’s the on premises data center. And then there’s the edge or the far edge, which can be a retail store, which can be a manufacturing floor, a battlefield and oil rig, many other places that are not the data center and not the remote cloud data center. And what we’re seeing is a couple of things. Number one is continued growth in the remote cloud that is happening. However, what is very profound is the continued growth of the on premises data center. But yet that customer wants the cloud experience, even though that data center is local, meaning not remote. And that cloud experience, of course, abstracts all the complexities and the costs and the skills required to managing It. So effectively, bringing the cloud to you or bringing that cloud experience to you in the data center is a very brisk and growing trend. And we can impute that same model. At the edge, there’s more and more intelligent compute and It infrastructure being put on, for example, manufacturing floors being put in retail stores such as large home improvement stores and grocery stores. 


And that, too, is looked at in a major way as being procured as a service. And that, again, prevents having to have It skills at the particular edge, at the retail store, on the manufacturing floor, and also frees up the investment that the company might have to be able to invest in their own products. Because most of our customers. 


Obviously, the dominant majority. 


Are not in the It business, and therefore running It as a service makes a lot of sense to them. So, in summary, in addition to that remote cloud growth we’re seeing on premises data center growth with the cloud experience, and then tremendous It complexities being put at the edge. 




And I’ll add on to that. 


Specifically in the ads. 


The trend that I see is how many streams can we get per box? How much compute power can we pack in? And how can I get my advanced algorithms to run at high performance, which is driving some need to offload some of that AI workload to custom accelerators. So essentially, it’s a story of compute power and performance all happening at the edge. There’s a lot of investment going on in that area, and there’s a few innovations that are driving some of those. Some of it is algorithm advances in the AI industry and some of our advancements on moving from lab to Fab for deployments and trying to get as many of those deployments fed to a box as possible, which is great news. This is bread and butter for us at intel, but it’s also the challenge if somebody wants to implement 10, 15, 20 streams on a single box. 


All running different AI models. 


Well, that’s a pretty big problem to solve, but those are the primary trends that I see driving at the edge. 


Sebastian, maybe if you have any thoughts, how is Airbus using different It infrastructures to support a lot of what you’re doing from a data center perspective? And maybe we just have a couple of minutes left, so maybe we’ll wrap up with this. 


So, I believe, as was mentioned, it’s just a compound of things. You can’t have one solution that fits all this is really key there and probably orchestrating the right tech for the right algorithm at the right place, at the right place is key. At least it’s what we are trying to achieve, meaning that we have jumped from manufacturing side, where everything was linked to the lowest level, to the Plc, to something which is kind of environment, and so on. And now we are discovering the direct connection to cloud, putting the fog in between, meaning that at the end of the day, as it was mentioned, computational power is key, connectivity is key, and the right tradeoffs give you the most successful stories at the end. 


More Resources

See All Related Content