Industry 4.0: AI and the Future of Manufacturing

TOM JONES: Hey, everybody. Welcome to AU. I hope everybody enjoyed the keynote this morning. And today, we're going to talk a little bit about the future of manufacturing. Let's see if this-- is this thing on? Maybe. There we go. Hit the button the right way. So my name's Tom Jones. I'm a solutions architect at Amazon Web Services. And my nickname is Elvis. I have two Las Vegas personalities in my repertoire here. Don't ask me to sing though.

So my focus is industrial software. So I help our partners in the industrial software space. I build solutions that run in the cloud. So what does that really mean when we say industrial software? Well, industrial software encompasses a lot of things. And everybody's probably familiar with most of these. But it's a pretty broad scope.

But let's start at the beginning. I want to give a little bit of foundation for folks in the room who may not be familiar with Amazon Web Services and may be wondering what Amazon's doing here. We're not here selling books. We do have a great website if you want to buy some books and have them delivered to your door in a smiling box. But let me tell you a little bit about AWS.

So today, Amazon Web Services offers a suite of reliable, scalable, inexpensive cloud computing services. And I know this is a little bit of an eye chart, but it's just to give you an idea of some of the scale. So we have over 90 different services, starting with some of our foundational services here on the bottom like compute as a service, networking as a service, storage as a service.

And then moving up to higher level services, like our database as a service offering, that's a relational database that you get to utilize but we manage for you. And then all the way up to these vertical services you see here on the top, things like enterprise application delivery, software development tools to develop and manage those services or whatever applications your building. Analytics and artificial intelligence, we'll talk more about that in a little bit.

So the AWS global scale. AWS has cloud infrastructure built around what we call regions and availability zones. A region, and we have 16 regions today, is made up of multiple availability zones, at least two availabilities zones. Many have three or more. An availability zone is one or more data centers. Each one of those data centers is a separate facility. They're on separate internet, separate power, separate floodplains.

And what that does is it allows you to build redundancy and high availability within a geographic region. All the green dots you see there are new regions that we have planned. So we have 16 today. We've got six more regions and 17 availability zones in the pipeline. At AWS, security is job number one. And we've built security from the ground up and have over a million active customers.

And one of the things that you get the benefits of when you're using AWS is you get to benefit from all the security controls and services that we've created for these million customers to benefit your application. We also have over 50 certifications and accreditations. And I love this quote at the bottom. And I can tell you all about our security, but I think it's a lot better for one of our customers to tell you what they think.

And here we see a quote from Capital One saying that they feel that they can be more secure in the cloud than they can in their own data center. And I know that security is a prime concern for people in the manufacturing space, because that data is your intellectual property. Here we're talking about money. I think that's pretty important, too.

So in summary, here's a list of reasons why AWS has been chosen by those million customers around the globe. And it's not just because of the things I'm telling you, it's because we've proven this over and over throughout the years. And we're really committed to that customer success. So we start with the customer and we work backwards when we're developing our services and how we're doing our customer service and approach. And that really shows.

And these are the reasons that companies, great companies like Autodesk, have built their products on top of AWS. So both A360 and the Forge sit on top of Amazon Web Services. And you get to take advantage of the global scale and the high availability that we offer through services like this. So you can rest assured that your project can also reap those same benefits.

So that's a little bit of a foundation. Let's talk about building a better mousetrap. Actually, what I want to talk about is better building a product. And when we build a product, we start with an idea, and we take that idea into production. And that process, if we look at it in a maybe more descriptive way, we start with product design, we go into production design, and then we go into production.

And you have a bunch of tools at your fingertips to help with this, so things like CAD and various products that can span some of these steps. And we build out this set of tools. And I think everybody is familiar with this or you wouldn't be at the Autodesk conference. But what I want to focus on for a minute is just this linear process.

If we take a look at this linear process today, traditionally, this process has been really a one way street. But what we're seeing now is the rise of data. So this is IoT data. This is data from the field, data from the factory floor. And that data is augmenting this process. And it's turning this process from a linear process into a cycle. And you're able to use that data to reinforce your design.

And what we're finding, or what I'm seeing across the industry, is that this trend is being fueled by data, by that data that's both in the design itself and part of the manufacturing process and then part of the lifecycle of that product. And so what we see is we see data at the center of this process, potentially using artificial intelligence to augment and pull more meaning out of that data. And then all of this is being utilized through tools by the humans.

So that's a big picture idea. Let's talk about where we are today. What are the trends? What are the challenges? What are the innovations that are shaping the future of manufacturing? So let's start with the challenges. What are some of the challenges? We have challenges around data. We have challenges around on-prem requirements. We have challenges around the amount of customization and the speed at which products need to be brought to market.

So let's look at each one of these in detail. Let's start with data. What are the challenges with data? One of the first challenges that we see is that data is in siloes. You've got proprietary data formats. You've got data in different physical locations. And that's being exacerbated as companies acquire other companies. And now you've got different systems and different areas of storage.

And just to highlight real quick, since we're talking about the Forge, the Forge, their model derivative API is one of the ways that you can try and address the first issue of file formats. If you're not familiar with it, it lets you translate file formats for over 60 different file types. But the other thing that we're seeing here is the volume of data.

So we've got these designs, and the designs are getting bigger, more complex, those models are large. How do you share those models with people you're collaborating with? How do you move this data? How do you ingest this data? So now we've got all these devices out in the field that are potentially communicating data back. How do you do that?

And I'll answer those questions, but I want to highlight one thing first and talk about, why is this important? Why is data important? Well, the first is that data has gravity. So data wants to be together. It's like gravity. The more that you can aggregate together, the more value you can get out of it. And data has value. And data value can, well, like we saw earlier, it can help drive the actual product innovation.

Collaboration. So putting all your data in one place facilitates collaboration. And then last but not least, having your data in one place allows new business models, like software as a service. So how do we build a place to store our data? So here's an architecture on AWS using various services. And we see here at the middle of this Amazon S3.

Amazon S3 is our object based storage service. And S3 today stores trillions of objects. You can have objects that are up to five terabytes in size for an individual object. So you can store pretty much anything you want there. And what we're seeing is we're seeing services here on this side that are sending data into S3. So you use basically create a data lake using S3.

S3 is really inexpensive. It's $0.03 per gigabyte per month to store content or objects there. So we have services like Amazon Kinesis, which is a streaming ingest service that will scale out so you can ingest whatever amount of data you want or have to ingest. We've got things like Amazon Direct Connect.

So maybe you have a factory and you don't want your data to go over the internet. You want it to go over your own private lease line. Well, you can do that with Amazon Direct Connect, and essentially take and extend your network into the cloud. And then once that data is in there, you can do things like transform it.

So we've got services like Amazon EMR. So EMR is Elastic MapReduce. You can do things like ETL using EMR, that's a lot of acronyms, and then put that data back in S3 for consumption through other services and potentially even display them with QuickSight, which is a way to create reports very easily.

But what I want to talk about next is this one here, Amazon Athena. Amazon Athena is essentially a SQL interface that sits on top of S3. So you create your data lake, and now you have a serverless way to execute SQL against your data. It's brilliant. Another way-- so I talked about EMR a second ago as a way to do ETL. Well, we have another service for that called Glue. It's also serverless. And you can use that for ETL as well.

Now, it's not going to meet every requirement, because there are still some limitations here. But over time, that will be built out. And I think this is really state of the art for building data lake on top of S3. So that's putting our data in one place. Let's talk for a second about our on-premise requirement.

So uptime is critical to a factory. Manufacturers want to ensure their operations are not going to be interrupted by a guy with a backhoe. And yeah, I know that's not a backhoe. I didn't have an icon for backhoe. So just use your imagination. Or maybe the factory has intermittent network connectivity. Or maybe it's got machines in it that were created before the machines had connectivity.

How do we address those? Well, one of the ways that we're addressing that is through a service we call Greengrass. And Greengrass is a way to do local processing but still have connectivity to the cloud. And Greengrass is made up of two components. We have Greengrass Core, and we have an SDK.

And Greengrass is software. It's not hardware. So you can bring your own hardware. Hardware requirements for green grass are pretty minimal. So if we take a look here at what the Greengrass Core does, so this is the Core piece, Greengrass Core is responsible for running your code, messaging, security, and interacting with the cloud.

And it has a minimum hardware requirement of a gigahertz processor and 128 megabytes of RAM. It'll run both on x86 hardware and ARM hardware. So pretty small requirements. Then the other component is the SDK. So this is where you can take our SDK and integrate it with your own devices. And this is currently an SDK for C++, but we'll have other languages coming soon.

And so those two components are one of the ways that you can do local processing. So in case you're not-- so our annual user conference is here in two weeks, a week after Thanksgiving. So we built this demo that uses a couple of industrial robots from KUKA that are being controlled by Greengrass device that's actually a Raspberry Pi.

So it has a Raspberry Pi running Greengrass controlling these robots instead of an industrial KUKA controller, which is quite expensive. Raspberry Pi is, what, $35? And the cool thing is that you can take that Greengrass device and disconnect it from the internet and these robots still operate. So it solves that a guy with a backhoe problem potentially.

Another way that companies are able to do this is through another device that we provide called the Snowball. The Snowball was initially created as a way to solve the data transport problem, where you've got lots of data on-premise and you want to move it to the cloud. So this particular version of the Snowball is called Snowball Edge.

It has 100 terabytes of storage onboard, but it also has on board compute. So you've got the equivalent of an m4.4xlarge compute instance on board the Snowball to run Greengrass. It's ruggedized, it's rack-mountable, and so on. And it's also clusterable. So we talked about a couple of challenges. Let's talk about another one, mass customization.

So I had a little bit of fun. I went and created my own version of Chuck Taylors on their website. And it's amazing to me the amount of customization that goes into that, not something really simple. They didn't show up yet, and I was hoping to be able to wear them here today. But you think about something like Caterpillar tractor or a John Deere tractor and the amount of customization that goes in.

So if you buy a tractor from John Deere today, you're buying a unique flower. But it's based off of configuration options, like a Chinese menu configuration options. Anyway, this push towards mass customization and what I like to call the commoditization of customization is a challenge. And one of the challenges with it is that demand fluctuates.

You never know at any one time how many people are going to come in order Chuck Taylors off your website. And then you have to go and build them. Well, how do you manage that? The other thing that we're seeing is simulation. So a lot of times when you're going to build something, and maybe it's something unique, you need to do simulation as well. How do you scale that?

Well, here's some tools, services, and best practices. And I'll just walk through these one by one. So we have this concept of autoscaling. So in the cloud, you can scale up and scale down to match your demand. And you can do that automatically based off of triggers. So you set a trigger that says, hey, my server, if it's at 80% utilization for X amount of time, spin up another server. You can do that.

You can take advantage of serverless technology. So now you're not even running a server. You just put in your code up there, and that code will scale, and we'll handle that on the back end for you to scale that to meet your demand. We have managed services. So I mentioned RDS earlier, relational database service. You can use that.

And all of these benefit from concepts of microservices and loose coupling, RESTful APIs. And this giant ball of hair that you see over here is actually the micro services that power amazon.com So if you go to Amazon and you load amazon.com, you're not hitting one application. You're hitting literally thousands of micro services.

Why is that important? It's important because that allows us to do over 50 million software deployments a year. That is agility. And that's the ability that you have to meet your customer demand and respond to your customers. So the other thing we talked about was this increasing complexity. And when you have more and more complexity, what do you need? You need more communication.

How do you facilitate communication between your teams, especially if those teams are distributed? Well, a couple ways that we have customers doing this today are through Amazon WorkSpaces, so this is a way to do desktops as a service, and through AppStream. So that's a way to do streaming of applications through a web browser.

And last but not least, let's talk about the last challenge, competitive pressure. This highlights a need for greater and greater efficiency. There's a few vectors here, like agility, which many of the previous topics addressed. But one here I want to highlight real quick is fiscal efficiency. So today, if you were building your own data center or buying servers, that's a capital expenditure. And it's a fixed asset.

Using the cloud, you pay as you go. You don't have any upfront commitment. So that gives you greater flexibility with your money. And if you decide tomorrow or next week that you need a different configuration, you can do that. That's agility. It's fiscal agility. And then it also means that the way business has been done in the past is not necessarily the way it will be done in the future.

So I mentioned software as a service based off of data. So you have that data in one place. You've got software as a service. And we see that today with the Forge itself. It's one of the things I'm so excited about the Forge. Autodesk has traditionally had a business around selling desktop software.

Now, they still have that, but they've taken their same intellectual property that they use to drive innovations into those desktop products and they're making it available through software as a service through the Forge. And they're allowing you to build products on top of that. That's a new business model. It's disruptive. And it's super interesting.

Well, and of course, we see things like additive manufacturing as disruptors in this place, too. Oh, that's my thing about the Forge. So what else do we see? We see automation as a way to address these efficiencies. So here's a picture of a couple of fascinating robots, Sawyer and Baxter. Is everybody familiar with Sawyer and Baxter? Anybody? No. A couple of people.

So Sawyer and Baxter are robots that were designed to work alongside humans. So most robots like the KUKA robot I showed you earlier, it's got to be in a cage. And you can't be close to it, because it could hurt you. It doesn't know you're there, whereas these do. And they know you're there.

So we're seeing that these types of robots are being utilized to do jobs that the humans don't want to do. So they're doing jobs that are dangerous or they're dirty or they're repetitive. And it's taking that human capital and freeing it up to do other things. And what we've seen today so far is we've seen we have a number of services that can be used for automation, Greengrass and Snowball Edge.

But the title of my talk is about artificial intelligence, and so let's talk about some of the artificial intelligence services. And I want to talk about how those artificial intelligence services can address the future of manufacturing. So let's go back to the very beginning of the talk where I talked about that linear process and those steps. We start with the designer.

And how can AI help this designer? Well, we see this through things like topology optimization, light weighting, materials optimization, and parameter tuning. And actually, I have a talk at our conference in a couple of weeks on parameter optimization through one of our partners that does hyper-parameter optimization for artificial intelligence.

But anyway, we start with design, then we go into production. We talked about automation just a second ago. So we're going to see how AI can influence that. It can do optimization. It can increase safety, and then extending that into cyber physical systems. So cyber physical systems are where you have AI and automation working together along with humans.

And then taking the fourth piece that we saw, the analytics, and we're tying AI into that, then we have things like operational tuning, so the ability to take a large set of operational parameters and tune them. So taking, for example, a turbine, a gas turbine, and getting the same efficiency out of it but reducing emissions. That's something that AI can do really well once it has the data. Because once again, data is key.

Doing logistics, so understanding where your supply chain is being restricted. You can do that, and AI can help there. Predictive maintenance is a big one for AI. But what we'll see is that using AI, you can move from predictive maintenance into prescriptive maintenance so you know exactly the state of an individual machine out in the field. And by using a digital twin, where it aggregates all of the data from all of the machines, you can make predictions about individual machines.

And then I think most interesting for me is this last one, which is descriptive design. So moving from this concept of pre-scriptive design where you have a designer who's saying I need a chair, it needs to be this big, it needs to be made of this material, you go into more of a descriptive design. And so let's explore that for a second. Well, I'll get to that. I've got a couple of slides about our AI services.

So today, Amazon offers three AI services. We've got Amazon Rekognition, which is a way to do image analysis off of photographs. And so if I took a picture of this room, it would be able to identify faces, people who were wearing glasses or not, sentiment, other objects in the room like chairs. Lex and Polly are voice-to-text and text-to-voice services. So these are services similar to what you see powering Alexa.

Next up we have platform services. We have a machine learning service, EMR, Elastic MapReduce, AWS Batch, which is batch computing service, and then a deep learning AMI. AMI is Amazon Machine Image. And so this AMI is something you can deploy onto our infrastructure, and it has various frameworks pre-configured to be able to run in the cloud.

And last but not least, we have our infrastructure offerings that support artificial intelligence. So primary one here being our EC2 P3 instances. The P3 family has Nvidia GPUs attached to them to accelerate machine learning and deep learning workloads. And then one of the things-- so here's an example of--

This was engineer Mikhail. This is an engineer at Autodesk. He built this demo. It's all based on serverless technology using the Alexa voice service and the AWS Lambda, which is serverless code execution, S3, which is storage, and DynamoDB, which is a NoSQL database. He's tied that altogether.

And you can see, it's a little bit hard to see, but there's an Amazon Echo Dot there attached to this CNC machine where he's showing, or guiding verbally and visually on this iPad, how to use this machine. So let me click on this, and we'll take a look. I've got a short little video. And I'll show you this in operation.

[VIDEO PLAYBACK]

- Let me show you how Alexa is going to guide me through the Haas machine operation process. Alexa, open Haas.

- Bonjour. I can assist you in starting your CNC program on your Haas. Let's begin. First, visually check the carousel to see if tools other than the probe are visible.

- OK.

- Look at the [INAUDIBLE] table and find the [INAUDIBLE] pocket.

- OK.

- Enter [INAUDIBLE] ATCFWD. Then check the [INAUDIBLE].

- OK.

- Now let's install tools into the machine. First, press MDIDNC. Then press [INAUDIBLE].

[END PLAYBACK]

TOM JONES: Well, I don't have to show you the whole thing. You can go to YouTube and watch it if you like. But I think that highlights how AI can be used to augment work that we already do today. That's a complex machine building this through a few relatively easy to use and accessible services. That's part of the future.

So backing up to that design process. So today, we're talking about prescriptive versus descriptive design, today, design is prescriptive. Design starts with the designer over here. And the designer narrows down the design, and then it goes into refinement, maybe through some simulation, and then it goes into fabrication. That's the process today.

What we're seeing by using that cycle that we saw earlier today is a refinement process. So where we start with requirements and we're more descriptive, we give the requirements of the design and say, hey, I need something that's this dimension or will support this weight, but explore all of these areas and surface back to me some of those results so that I can choose the ones that I think are most appropriate.

So it's both that artificial intelligence and a human working together in the design process. And of course, once again, we've got that IoT data that can feed back into the requirements and refine this process. So it's not that linear process. It's a cycle.

And one of the products I think that highlights this the best is Autodesk Dreamcatcher. So Dreamcatcher is a way to do generative design. And here we see some designs for a skateboard truck. And these are various design iterations based on the requirements for this particular object.

And what ended up happening was through a combination of computer modeling and simulation stress analysis and that human helping to guide the process, this is what they ended up with, a very organic shaped, 3D printed truck for a skateboard. That's something that can only be manufactured through additive manufacturing. Really fascinating.

So what we're seeing is manufacturing is undergoing a transformation. It's being driven by a number of factors, the growth of data, the ability to process that data through automation, connectivity. And all of this is going to end up benefiting businesses. But it's also going to make the workplace safer. And it has the potential to actually benefit the environment. Oh, that was supposed to-- there we go. Benefit the environment.

And all of it's centered around that data in the middle being driven through these other layers and controlled by humans. Once again, AWS as a service is to help you capitalize on these trends. And here's some resources that are available to help you get started.

We also have our conference, that I mentioned, that is going to be here in Las Vegas in two weeks. And I have a talk tomorrow talking about how you can use the Forge and AWS together. And we'll show some examples of development.

So where are we headed from here? Personally, I'm very bullish on the future of manufacturing. I see these trends pushing the democratization of manufacturing and the ability for anybody, or a small team of people, to build things. I see the enhancement, or the acceleration, of the commoditization of customization, where more and more products are going to be bespoke and custom created.

And they're going to be using that data, that IoT data, building digital twins to drive more efficiency into the manufacturing process. And like I said, that's going to be good for everybody. And I personally think it's a very exciting time to be in manufacturing. And I'm really looking forward to seeing where the future takes us. So thank you very much. I appreciate your time. I'll be here if you want to come up and ask any questions.

[APPLAUSE]