>>Hey, we're here in the second manufacturing drill down session with Michael Gerber. He was the managing director for automotive and manufacturing solutions at Cloudera. And we're going to continue the discussion with a look at how to lower costs and drive quality in IOT analytics with better uptime and hook. When you do the math, it's really quite obvious when the system is down, productivity is lost and it hits revenue and the bottom line improve quality drives, better service levels and reduces lost opportunities. Michael. Great >>To see you take it away. >>All right, guys. Thank you so much. So I'd say we're going to talk a little bit about connected manufacturing, right? And how those IOT IOT around connected manufacturing can do as Dave talked about improved quality outcomes for manufacturing and flute and improve your plant uptime. So just a little bit quick, quick, little indulgent, quick history lesson. I promise to be quick. We've all heard about industry 4.0, right? That is the fourth industrial revolution. And that's really what we're here to talk about today. First industrial revolution, real simple, right? You had steam power, right? You would reduce backbreaking work. Second industrial revolution, mass assembly line. Right. So think about Henry Ford and motorized conveyor belts, mass automation, third industrial revolution, things got interesting, right? You started to see automation, but that automation was done essentially programmed your robot to do something and did the same thing over and over and over irrespective about of how your outside operations, your outside conditions change fourth industrial revolution, very different, right? >>Cause now we're connecting, um, equipment and processes and getting feedback from it. And through machine learning, we can make those, um, those processes adapted right through machine learning. That's really what we're talking about in the fourth industrial revolution. And it is intrinsically connected to data and a data life cycle. And by the way, it's important, not just for a little bit of a slight issue, there we'll issue that, but it's important. Not for technology's sake, right? It's important because it actually drives very important business outcomes. First of all, quality, right? If you look at the cost of quality, even despite decades of, of, of, uh, companies and manufacturers moving to improve while its quality prompts still accounts for 20% of sales, right? So every fifth of what you meant are manufactured from a revenue perspective, do back quality issues that are costing you a lot planned downtime, cost companies, $50 billion a year. >>So when we're talking about using data and these industry 4.0 types of use cases, connected data types of new spaces, we're not doing it just merely to implement technology. We're doing it to move these from members, improving quality, reducing downtime. So let's talk about how a connected manufacturing data life with what like, right, but this is actually the business. The cloud area is, is in. Let's talk a little bit about that. So we call this manufacturing edge to AI. This is analytics life cycle, and it starts with having your plants, right? Those plants are increasingly connected. As I say, sensor prices have come down two thirds over the last decade, right? And those sensors are connected over the internet. So suddenly we can collect all this data from your, um, manufacturing plants, and what do we want to be able to do? You know, we want to be able to collect it. >>We want to be able to analyze that data as it's coming across. Right? So, uh, in scream, right, we want to be able to analyze it and take intelligent time actions. Right? We might do some simple processing and filtering at the edge, but we really want to take real-time actions on that data. But, and this is the inference part of things are taking about time, but this, the ability to take these real-time actions or, um, is actually the result of a machine learning life cycle. I want to walk you through this, right? And it starts with, um, ingesting this data for the first time, putting it into an enterprise data lake, right in that data lake enterprise data lake can be either within your data center or it could be in the cloud. You're going to, you're going to ingest that data. You're going to store it. >>You're going to enrich it with enterprise data sources. So now you'll have say sensor data and you'll have maintenance repair orders from your maintenance management systems. Right now you could start to think about, you're getting really nice data sets. You can start to say, Hey, which sensor values correlate to the need for machine maintenance, right? You start to see the data sets. They're becoming very compatible with machine learning, but so you, you bring these data sets together. You process that you align your time series data from your sensors to your timestamp data from your, um, you know, from your enterprise systems that your maintenance management system, as I mentioned, you know, once you've done that, we can put a query layer on top. So now we can start to do advanced analytics query across all these different types of data sets. But as I mentioned to you, and what's really important here is the fact that once you've stored one history sets data, you can build out those machine learning models. >>I talked to you about earlier. So like I said, you can start to say, which sensor values drove the need of correlated to the need for equipment maintenance for my maintenance management systems, right? And you can build out those models and say, Hey, here are the sensor values of the conditions that predict the need for maintenance. Once you understand that you can actually then build out the smiles, you could deploy the models after the edge where they will then work in that inference mode, that photographer, I will continuously sniff that data as it's coming and say, Hey, which are the, are we experiencing those conditions that, that predicted the need for maintenance? If so, let's take real-time action, but schedule a work order and equipment maintenance work order in the past, let's in the future, let's order the parts ahead of time before that piece of equipment fails and allows us to be very, very proactive. >>So, >>You know, we have, this is a, one of the Mo the most popular use cases we're seeing in terms of connected, connected manufacturing. And we're working with many different manufacturers around the world. I want to just highlight. One of them is I thought it's really interesting. This company is for SIA for ECA is the, um, is the, was, is the, um, the, uh, a supplier associated with Pooja central line out of France. They are huge, right? This is a multinational automotive, um, parts and systems supplier. And as you can see, they operate in 300 sites in 35 countries. So very global, um, they connected 2000 machines, right. Um, and they once be able to take data from that. They started off with learning how to ingest the data. They started off very well with, um, you know, with, uh, manufacturing control towers, right? To be able to just monitor the data firms coming in, you know, monitor the process. >>That was the first step, right. Uh, and you know, 2000 machines, 300 different variables, things like, um, fibrations pressure temperature, right? So first let's do performance monitoring. Then they said, okay, let's start doing machine learning on some of these things to start to build out things like equipment, um, predictive maintenance models, or compute. What they really focused on is computer vision, wilding inspection. So let's take pictures of parts as they go through a process and then classify what that was this picture associated with the good or bad quality outcome. Then you teach the machine to make that decision on its own. So now, now the machine, the camera is doing the inspections beer. And so they both have those machine learning models. So they took that data. All this data was on-prem, but they pushed that data up to the cloud to do the machine learning models, develop those machine learning models. >>Then they push the machine learning models back into the plants where they, where they could take real-time actions through these computer vision, quality inspections. So great use case, a great example of how you can start with monitoring, move to machine learning, but at the end of the day, or improving quality and improving, um, uh, equipment uptime. And that is the goal of most manufacturers. So with that being said, um, I would like to say, if you want to learn some more, um, we've got a wealth of information on our website. You see the URL in front of you, please go there and you'll learn. There's a lot of information there in terms of the use cases that we're seeing in manufacturing and a lot more detail and a lot more talk about a lot more customers we'll work with. If you need that information, please do find it. Um, with that, I'm going to turn it over to Dave, to Steve. I think you had some questions you wanted to run by. >>I do, Michael, thank you very much for that. And before I get into the questions, I just wanted to sort of make some observations that was, you know, struck by what you're saying about the phases of industry. We talk about industry 4.0, and my observation is that, you know, traditionally, you know, machines have always replaced humans, but it's been around labor and, and the difference with 4.0, and what you talked about with connecting equipment is you're injecting machine intelligence. Now the camera inspection example, and then the machines are taking action, right? That's, that's different and, and is a really new kind of paradigm here. I think the, the second thing that struck me is, you know, the costs, you know, 20% of sales and plant downtime costing, you know, many tens of billions of dollars a year. Um, so that was huge. I mean, the business case for this is I'm going to reduce my expected loss quite dramatically. >>And then I think the third point, which we turn in the morning sessions and the main stage is really this, the world is hybrid. Everybody's trying to figure out hybrid, get hybrid, right. And it certainly applies here. Uh, this is, this is a hybrid world you've got to accommodate, you know, regardless of, of where the data is. You've gotta be able to get to it, blend it, enrich it, and then act on it. So anyway, those are my big, big takeaways. Um, so first question. So in thinking about implementing connected manufacturing initiatives, what are people going to run into? What are the big challenges that they're gonna, they're gonna hit? >>You know, there's, there's there, there's a few of the, but I think, you know, one of the, uh, one of the key ones is bridging what we'll call the it and OT data divide, right. And what we mean by the it, you know, your, it systems are the ones, your ERP systems, your MES systems, right? Those are your transactional systems that run on relational databases and your it departments are brilliant at running on that, right? The difficulty becomes an implementing these use cases that you also have to deal with operational technology, right? And those are, um, all of the, that's all the equipment in your manufacturing plant that runs on its proprietary network with proprietary pro protocols. That information can be very, very difficult to get to. Right. So, and it's unsafe, it's a much more unstructured than from your OT. So the key challenge is being able to bring these data sets together in a single place where you can start to do advanced analytics and leverage that diverse data to do machine learning. >>Right? So that is one of the, if I had to boil it down to the single hardest thing in this, uh, in this, in this type of environment, nectar manufacturing is that that operational technology has kind of run on its own in its own world for a long time, the silos, um, uh, you know, the silos, uh, bound, but at the end of the day, this is incredibly valuable data that now can be tapped, um, um, to, to, to, to move those, those metrics we talked about right around quality and uptime. So a huge opportunity. >>Well, and again, this is a hybrid theme and you've kind of got this world, that's going toward an equilibrium. You've got the OT side, you know, pretty hardcore engineers. And we know, we know it. Uh, a lot of that data historically has been analog data. Now it's getting, you know, instrumented and captured. Uh, so you've got that, that cultural challenge. And, you know, you got to blend those two worlds. That's critical. Okay. So Michael, let's talk about some of the use cases you touched on, on some, but let's peel the onion a bit when you're thinking about this world of connected manufacturing and analytics in that space. And when you talk to customers, you know, what are the most common use cases that you see? >>Yeah, that's a good, that's a great question. And you're right. I did allude to it earlier, but there really is. I want people to think about, there's a spectrum of use cases ranging from simple to complex, but you can get value even in the simple phases. And when I talk about the simple use cases, the simplest use cases really is really around monitoring, right? So in this, you monitor your equipment or monitor your processes, right? And you just make sure that you're staying within the bounds of your control plan, right? And this is much easier to do now. Right? Cause some of these sensors are a more sensors and those sensors are moving more and more towards internet types of technology. So, Hey, you've got the opportunity now to be able to do some monitoring. Okay. No machine learning, we're just talking about simple monitoring next level down. >>And we're seeing is something we would call quality event forensic announces. And now on this one, you say, imagine I've got warranty plans in the, in the field, right? So I'm starting to see warranty claims, kick kickoff. And what you simply want to be able to do is do the forensic analysis back to what was the root cause of within the manufacturing process that caused it. So this is about connecting the dots by about warranty issues. What were the manufacturing conditions of the day that caused it? Then you could also say which other tech, which other products were impacted by those same conditions. And we call those proactively rather than, and, and selectively rather than say, um, recalling an entire year's fleet of the car. So, and that, again, also not machine learning where simply connecting the dots from a warranty claims in the field to the manufacturing conditions of the day, so that you could take corrective actions, but then you get into a whole of machine learning use case, you know, and, and that ranges from things like quality or say yield optimization, where you start to collect sensor values and, um, manufacturing yield, uh, values from your ERP system. >>And you're certain start to say, which, um, you know, which map a sensor values or factors drove good or bad yield outcomes. And you can identify those factors that are the most important. So you, um, you, you measure those, you monitor those and you optimize those, right. That's how you optimize your, and then you go down to the more traditional machine learning use cases around predictive maintenance. So the key point here, Dave is, look, there's a huge, you know, depending on a customer's maturity around big data, you could start something with monitoring, get a lot of value, start, then bring together more diverse data sets to do things like connect the.analytics then and all the way then to, to, to the more advanced machine learning use cases there's value to be had throughout. I >>Remember when the, you know, the it industry really started to think about, or in the early days, you know, IOT and IOT. Um, it reminds me of when, you know, there was the, the old days of football field, we were grass and, and a new player would come in and he'd be perfectly white uniform and you had it. We had to get dirty as an industry, you know, it'll learn. And so, so my question relates to other technology partners that you might be working with that are maybe new in this space that, that to accelerate some of these solutions that we've been talking about. >>Yeah. That's a great question that it kind of, um, goes back to one of the things I alluded earlier, we've got some great partners, a partner, for example, litmus automation, whose whole world is the OT world. And what they've done is for example, they've built some adapters to be able to catch it practically every industrial protocol. And they've said, Hey, we can do that. And then give a single interface of that data to the Patera data platform. So now, you know, we're really good at ingesting it data and things like that. We can leverage say a company like litmus that can open the flood gates of that OT data, making it much easier to get that data into our platform. And suddenly you've got all the data you need to, to, to implement those types of industry 4.0, our analytics use cases. And it really boils down to, can I get to that? Can I break down that it OT, um, you know, uh, a barrier that we've always had and bring together those data sets that we can really move the needle in terms of improving manufacturing performance. >>Okay. Thank you for that last question. Speaking to moving the needle, I want to lead this discussion on the technology advances. I'd love to talk tech here, uh, are the key technology enablers, and advancers, if you will, that are going to move connected manufacturing and machine learning forward in this transportation space, sorry, manufacturing in >>A factory space. Yeah. I knew what you meant in know in the manufacturing space. There's a few things, first of all, I think the fact that obviously I know we touched upon this, the fact that sensor prices have come down and have become ubiquitous that number one, we can w we're finally being able to get to the OT data, right? That's that's number one, number, number two, I think, you know, um, we, we have the ability that now to be able to store that data a whole lot more efficiently, you know, we've got back way capabilities to be able to do that, to put it over into the cloud, to do the machine learning types of workloads. You've got things like if you're doing computer vision, while in analyst respect GPU's to make those machine learning models much more, uh, much more effective, if that 5g technology that starts to blur at least from a latency perspective where you do your computer, whether it be on the edge or in the cloud, you've, you've got more, you know, super business critical stuff. >>You probably don't want to rely on, uh, any type of network connection, but from a latency perspective, you're starting to see, uh, you know, the ability to do compute where it's the most effective now. And that's really important. And again, the machine learning capabilities, and they believed the book to build a GP, you know, GPU level machine learning, build out those models and then deployed by over the air updates to your equipment. All of those things are making this, um, there's, you know, there's the advanced analytics machine learning, uh, data life cycle just faster and better. And at the end of the day, to your point, Dave, that equipment and processor getting much smarter, very much more quickly. Yep. We got >>A lot of data and we have way lower cost, uh, processing platforms I'll throw in NP use as well. Watch that space neural processing units. Okay. Michael, we're going to leave it there. Thank you so much. Really appreciate your time, >>Dave. I really appreciate it. And thanks. Thanks for, for everybody who joined us. Thanks. Thanks for joining.