Revolutionizing Gigafactories with a data-driven approach with Siemens, Voltaiq and AWS
Show notes
It is the last episode of The Battery Podcast’s first season, and it is also the second half to a conversation of industry experts from Siemens, AWS, and Voltaiq. In Part One, we covered the critical need for data driven manufacturing in the battery industry as well as the potential development timeline acceleration such an approach brings. In Part Two, we are diving into how businesses can integrate a data-driven approach into their operations.
Your host is Puneet Sinha of Siemens Digital Industries and our guests for this final episode are Hari Gopalakrishna Head of Strategy, Sustainability, and EV at AWS and Tal Sholklapper, CEO of Voltaiq. Let’s dive into how data-driven manufacturing can bring long-term success to a sometimes-uncertain industry.
What you’ll learn:
[00:30] Building out data-driven manufacturing starts with a foundation in IT-OT convergence and the digital twin
[06:46] Applying data and intelligence to the common problems of battery manufacturing
[09:51] The role of AI in data-driven manufacturing
[13:50] How the partnership of Siemens, AWS, and Voltaiq can help battery manufacturers
Transcript
Welcome to the battery podcast from Siemens Digital-Industries. We're picking up our conversation with Punit Sinha, Hari Gopal Krishna, and Tal Schlapper on making battery manufacturing a data-driven endeavor. In the last episode, we covered the trends and benefits, making this a worthwhile evolution, but for this one, we're diving into how businesses can integrate a data-driven approach into their operations. Take it away, Pune. Perfect. So let, let me peel the next layer of onion, and then we talk about, as you said, we have 3 pillars to our data-driven manufacturing practice. We'll talk about 2 ITOT convergence and digital twin. As I said earlier, digital twin is. It, it, it is there to, to help you do a lot of things in the digital world. You can look at the, not only just the self product performance, but also how the machines are, uh, developed, how they need to, all of these machines needs to connect, how all of these different machines needs to operate. And of course, ITOD convergence is coming at the factory floor of how you're bringing MES and how you're connecting, uh, with, with the, the right machines and factories together. Um, there is one example I want to bring, which is one of my favorite, is many times people think of digital twin that happens before you get in a factory, you're in the engineering side, you're sitting in front of a computer and do that, and you somehow throw that across the wall and things come in the, in the, in the factory. And when the things get real, things start breaking down. So I want to bring an example to elaborate that these things are not siloed, and that example is. And, um, I'll talk about a specific problem. So if you look at the winding. This, this operation happens at a at a very high speed, and, and the tension needs to be correct for the during the winding too loose, you are going to have a bad winding, too tight of tension, foil can break. And if foil breaks, of course, you have to stop the production. There are time and costs associated with it. This is a great example where we are bringing digital twin, which typically people think of something that you do when you're sitting in a in a cubicle. We take that digital twin off that machine of that winding operation, creating an executable version that is running on industrial edge in a factory. And the power that is as the operation is running on the left hand side. The digital representation of the same operation is running on the right-hand side. And that is allowing you to see, hey, is the tension, stress or strain getting too much at any point in time in that foil, something you can't measure. And as that digital twin explores that, it is working as a virtual sensor in a certain way. And then through ITOT integration, we can feed through PLCs how the motor torque speed needs to be changed, if it is getting it, you know, if the tension is too high, how to bring it down. If the tension is getting too low, how to speed, how to bring it up. This, this is an example of how the digital twins and ITOT integration, all of that come together. And it allows you to. Avoid the problem before it happens. Once the problem has happened, then you have to find a solution. It takes time. So this is one of my favorite examples of showing the power of digital and ITOD convergence, and they don't need to happen one in cubicle, one on a factory floor. So Hari, I'll come to you and help us understand how. AWS is making this happen because there's a lot of data and a lot of architecture there. Um, yeah, so. So what you're seeing right now on the screen is an industrial, a purpose-built industrial architecture. So, here, you know, what you're seeing also is a hybrid architecture. You have the cloud, you have the edge, and you have in between, right? There are some things that are better done in the cloud, and there are some things that are better done at the edge. Things that are latency sensitive, that are going to be needed with short-term, like really quick inference needs that are going to happen at the edge. So what you're seeing here it's a great example, by the way. You have an AWS IOT client running on Siemens Industrial Edge, which is connected to the equipment. Which it has the ability to connect to diverse set of equipment and process variables. You can collect all kinds of data. Once this data is collected, it's contextualized and it's sent to the cloud where that data could be combined with other types of data from MES, CRP, CMMS systems, all these different skater systems. And so that's where you would build your models, build your digital twins, and once you have that, you would then deploy that at the edge and it will run at the edge because they are, you know, context sensitive, they are latency sensitive, and so this is the kind of fundamental architecture that we operate within the industrial facilities at scale now as an example. As you scale from a lab to a gig gigafactory, the, the volume of data is also very high, uh, for one of our customers, it's almost 1 petabyte of data per day. So in order for you to make sense of this data and to be able to truly drive insights, you have to know how to handle that data, how to really contextualize, process it at the right places, and also, you know, really connect with other types of data. So you know you have the overall view of what problems are you solving, because it's ultimately outcome centric, right? You're driving an outcome. So Absolutely. And, and as you said, it's a data processing at the right places very critical because you cannot. Download the data from one computer and in in your USB drive and take to the other. So that is a real um practical problem. So I'm, I'm gonna. I shift the gear and come to scrap rate and quality issue. Earlier I mentioned that as we are deploying this, our data-driven manufacturing practices with our collective solutions, I mentioned that at a customer there is a 10% reduction in scrap rate we were able to achieve from their baseline. So, Tal, I would love for you to explain that, elaborate on, on that, that customer challenges. How did we solve that? And what, what are the borders that we are able to solve together? Yeah, of course, so it's a handful of these scenarios we've seen across customers, you know, the, there's that initial ramp up as you're getting up to, you know, increasing production volumes, you're, you know, decreasing those scrap rates, and a lot of customers we've seen this like stubborn, you know, we can get up to 60, 70, 80%, and, uh, taking that next step and further decreasing scrap rates, uh, from whatever that baseline is, is, is very challenging. And even once you get into production, there's still continuous uh challenges that will pop up and deviations and uh excursions that will occur in the production process, so. Uh, this customer, uh, in particular, uh, they started initially to see some early issues with, uh, very early warranty returns from their customer, um, their end customer, which was putting it into devices, and what they, you know, realized that they were able to catch some of these with a little tighter, uh, tolerances, uh, for their OCVs and voltage drops, uh, for self-discharge from the cells at the end of the production process. Uh, the first thing we did with them was actually start looking, uh, a little deeper into their existing formation data, and we're able to identify, uh, the ability to actually catch many of these defects within the first, uh, couple of hours of formation, those initial charge-ups, and be able to identify, uh, a number of those, uh, deviant cells, uh, very early in the process. The next thing we did was, it was very rudimentary. They did not have the full, you know, genealogy of their upstream process steps, uh, but we were able to actually take in at least the granular data, what. You know, piece of equipment that this come through. What a lot of, you know, slurry was mixed, what was the batches of material, and we're actually able to go through and tie this to the storage rooms and the humidity of their materials that were coming into the facility. And so when they didn't have proper materials handling and keep the material in the right environment, uh, we had these excursions that were leading to moisture depositing in the material. And, you know, It was great that we're able to, you know, shift left, catch those issues, you know, in formation early on, just looking at things like the differential capacity for the cells, but really the key, you know, thing that I was able to save them on the scrap race was, uh, you know, identifying the root cause which was actually in the materials handling and the moisture of the materials in this case, uh, but. Tremendous value all from things and data that already exists at the customer and just really putting that together in a way so that you could, you know, shift left the identification of issues, start remedying those, uh, looking to remediate those as quickly as possible and then having a way to drill down and really understand where that is coming from rather than having to go with thumb drives or disparate data systems internally, uh, and all in a simple, simple system. Perfect. So, I have an obvious question. So we, we talked a lot about data, collecting data and, and processing data. So we get this question a lot and I, I would love for us to answer that is, if it is all about data processing and extracting intelligence out, out of data, we are in the AI age. So why can't AI teams or data analytics teams in, in companies handle this? Yeah, it's something we get asked very often today, you know, someone can go in with some, uh, AI tools and very quickly build a dashboard. It's, it's, uh, if anything, it's easier than ever, which is great. Our teams take advantage of it as well, um, but the real challenge here is in order to really understand this domain specific, uh, you know, problem, you need to have a couple of things. You have to have the data well structured in a way that you can actually, you know, learn from large amounts of it. You need a lot of data, you need to be cleaned, and you don't just need the raw data most of the times, you actually need the features and metrics extracted from it that provide you insight onto what's going on in those devices. Um, and so. You know, really what we have been focused on is trying to create the infrastructure and platforms to, you know, do that to, you know, if you look across the battery industry, there's a whole range of AI solutions you'll see out there. They're trained off like 3 to 50 data points. It's, you know, that's not AI, that's, you know, you could do regression, you could do better most of the time just in Excel. Um, and so really the key here is, you know, you want to implement these tools. These tools are great, we're super excited about them, but really getting the data pipelines and infrastructure in place to bring this data at scale into, uh, solutions where you can leverage these advanced techniques is really the key in the battery industry and getting that infrastructure and then also consultatively helping customers figure out what data to bring in, you know, there's, there's a lot of potential data you can collect and. What data do you bring in and what resolution and how do you actually extract that meaningful insight so that you can layer in these advanced tools is really key. What about you, Audie? You have something to add? No, I think that he's covered it really well. One thing that I would add is, you know, Voltig as an example is a a solution that's built on AWS. We have a number of services that are targeted at security, data governance, and, um, a key management services, identity identity management, um, encryption services and things like that. And so, you know, if there are fear about he going to do this in the cloud because I have, you know, concerns about IP and, you know, IP and data protection and that sort of thing, uh, this is, you know, we have a suite of services to assist with that. And generally when these solutions are built on AWS we make sure that it's built correctly and so we have our own foundational technical review process to ensure let's say a full solution like Voltaic is leveraging the right type of security data management practices and it's built for scale so you know. There shouldn't be any concerns around that, you know, using cloud and that sort of stuff, yeah, yeah, and one of the fun things we've seen is that, you know, a lot of these early, you know, exec driven or internally driven, like, hey, we want to implement AI, we want to get something going, and, uh, at the end of the day, you know, these people, you'll spend 9 months doing essentially a POC and then they try to roll it out. It doesn't work because you haven't trained on enough data, you don't have the right infrastructure to enable these and. It's really the, the coming together of solutions like ours that they try to sort of hack together for those projects, um. You know, they, they, they end up really trying to pull us in and we become those key partners helping them enable these advanced tools. Absolutely. You know, guys, what, what I say is the difference between dashboarding and data intelligence. Dashboarding is knowing that tomato is a fruit, and, and data intelligence is knowing that you don't put tomato in your fruit salad. Um, so, I know we, our time is coming to an end. Uh, but before we end, I would love to hear what, what, how will you communicate to our audience? What is our joint value propositions? How, um, our audience, if they want to work with us, how they can approach us and initiate this journey. Howdy, do you want to start? Sure, so battery manufacturing is complex on many fronts from materials, R&D to manufacturing. And there's different domains of expertise in that, right? And what we are bringing, you know, from AWS is that data expertise, uh, AI machine learning. Large language models, generative AI, that sort of things, advanced analytics, and the right type of industrial data architecture to handle the data and the volume from the plan to the cloud, right? But that's only one domain. We need other domains like yourself, Siemens bring the expertise around industrial er connectivity. Uh, automation, software, PLM, MES, and decades and, you know, over 100 years of experience. So, same thing with Voltaic, the level of battery knowledge and expertise that they bring to the table and really making sense and, you know, driving insights out of all of this. That's why, you know, bringing all three of us together. And bringing an integrated set of tools and platform really helps accelerate uh our, our customers' business, right? The innovation and, and in time to market absolutely tell. Yeah, it comes down to, you know, the time frames, you know. Whether you've got an existing factory or you're putting up new ones, the cost pressures are here, the time pressures, you need to get these factories up and running, uh, you need to, you know, ensure the quality of the cells coming out of there because. You know, the investments, uh, you know, periods of free, low, you know, low interest rate money is gone. You're gonna have to do this with, you know, smaller staffs. You're not gonna have the leeway of years to try to get these factories up and running, uh, the high scrap rates, you know, the, with the cost models and contracts now for sales and prices, you know, you gotta work those quality issues out of the production process as quickly as you can, and so. Really what we're focused on is, you know, trying to deliver a complete solution here to customers and, you know, from the PLCs to the controls to the, you know, analytics to the infrastructure to store and host all this data, um, try to deliver that in a complete solution as quickly as possible so that you can hit the ground running and your team could be more efficient, uh, you know, you're not getting back that extra headcount and, you know, allow us to come in and partner with you to help you, you know, achieve those goals faster with the, you know, limitations we have right now. Thank you, guys. And um I just want to leave you guys with this image again, that data-driven manufacturing is achievable. We are doing it together by bringing digital twin, ITOT convergence, and data intelligence together. And one last point, if I can leave you with this, you don't have to wait till your production becomes at a high, large scale. This is something you can implement. Early. If you are getting up at the R&D scale, even at a pilot scale, this is a great place to start because the longer you wait, the bigger the problem it becomes. It's a lot easier to, to optimize your operation, extract the knowledge from the systems early in the cycle. So if there is one action I can leave you with is start early, scale fast, come to us. We are there to help you. Thank you for listening to the battery podcast. We're putting a lid on our first season of the show, but we'll be back soon with season two as we enter a new year. There will be more guests from around the world, new conversations from industry showcases and insights on the different battery markets around the world. Make sure to subscribe so you don't miss the next season's premiere or any new episodes if you're listening in the future. We'll be back soon, but you can always check out our website, Siemens.com/battery for more information. This podcast contains the personal views and opinions of the speakers. It is important to note that these opinions are solely those of the speakers and do not reflect the views or positions of Siemens, its management, its employees, or any entities the speakers may be affiliated with. Siemens as a company does not endorse any of these statements made in the podcast. Siemens bears no responsibility for ensuring the content's accuracy or verifying the statements made in this podcast.