From Imagination to Implementation: Generative AI's Revolutionary Role in Manufacturing

All right, welcome everybody. I'm Lucian Fogoros. I'm your host for today, and welcome to day 2 of IoT World's Manufacturing Day. Days 2024. We're excited to kick off the session with this AI, but before we do, those of you that attended yesterday, I'd like to ask you if you could put in a chat what exactly— what are some of your takeaways? What do you like the most? And if you didn't attend yesterday, all the sessions will be available on demand, so you can go back and watch those, or feel free to let us know what you're most excited about today's session. I'd like to welcome to the stage Mike Ungar, the CEO of Industry 4.0 Club, who will moderate this session. Welcome, Mike.

Thanks, Lucian, and welcome everybody. Really glad to be here with you today. As Lucian mentioned, we'll be taking questions at the end of our panel discussion. This track's going to run about 59 minutes, and so we plan to have all of you out of here on time. I want to just mention that it wouldn't be possible to run this virtual conference without our sponsors. And thank you to the sponsor today of this session, NanoPrecise SciCorp. You'll notice in the tool there's a place for Q&A, and that's where we'll be taking our questions. So please put your questions there rather than in the chat. So I'd like to introduce our topic and panelists for today's session, From Imagination to Implementation: Generative AI's Revolutionary Role in Manufacturing. So our speakers for today, uh, we're really— it's really gonna be a great panel, some great expertise. Eric Thompson is the Director of IoT and AI Solutions with Iris. With a robust foundation in contact center as a service, IoT, generative AI, and cloud technologies, Eric is a globally recognized trusted advisor. His 10-year tenure encapsulates extensive leadership in business strategy, innovation, and transformative initiatives on a grand scale with multifaceted environments. Eric's expertise is marked by a consistent track record across the entire spectrum of customer success, encompassing pre-sales, onboarding, adoption, and growth phases. Welcome, Eric.

Thank you.

Kathy Cahalan is founder and CEO of Bennett AI. Kathy brings a passion for improving the lives of manufacturing workers, and those that support manufacturing environments. Following years of working in advanced manufacturing with Goodyear, she earned her master's degree in architecture with a focus on human-centered design. She returned to industry in 2017 to take the helm at Bennett AI and shape its mission to connect people to solve manufacturing problems. She sits on the executive committee of the board of directors at SESAME, the United States National Institute on Smart Manufacturing. Welcome, Kathy.

Nice to be here.

Scott Wyant is a senior systems engineer at Swagelok, a leading provider of fluid system solutions. He provides deep technical expertise in many areas, including operational technology, system architecture, integration, software development, DevOps, and SCADA and MES. Prior to his focus on digital manufacturing systems, Scott held roles in various functional areas including applications, quality, and most recently manufacturing engineering, where he focused on CNC machining of stainless steel valve bodies. Welcome, Scott.

Thank you.

And our last panelist, but certainly not least, is Manpreet Singh. He is currently the CTO of NanoPrecise SciCorp, our sponsor. NanoPrecise is a leading is a leader in the industrial IoT domain with a focus on energy-centered maintenance and condition monitoring. He heads the hardware, software, and data science teams at NanoPrecise. He's led product and technology roles in the past and has been an entrepreneur in the technology space. He is passionate about technology and AI and brings a depth of understanding and implementation of AI algorithms to create business value. Welcome, Manpreet. Let's get started. Merrick, maybe you can get us started just to help us set the baseline. How does generative AI differ from analytical AI in manufacturing?

It's a great first question to differentiate generative AI and analytical AI. GenAI, so we'll probably use that for short, is really the implementation needs and use cases from those of analytical AI. So if you think about analytical AI, it's primarily concerned about just analyzing existing data for insights, it detects patterns, it makes predictions, and that includes techniques such as like regression analysis, clustering, classification of data, and then just really leveraging those models from a decision tree, neural networks, and support vector machines. Generative AI, meanwhile, really opens up a new frontier for problem-solving, illustrated by recent innovations in content creation, insight generation. I think we've all heard about GenAI Copilot and human-like interactions. A couple of examples of applications in manufacturing for analytical AI. Obviously the big one, predictive maintenance, analyzing that sensor data and operational data to really predict equipment failures. The next one would be around quality control, using analytical AI to really identify defects, ensure product quality, and really analyzing production data. The last one around supply chain optimization, using analytical AI to really analyze the chain data to really optimize logistics and inventory management. And so when we think about applications in manufacturing from a generative AI perspective, the first thing that really— where GenAI has really kind of helped is around design and innovation, right? It helps generate new product designs. It optimizes existing ones. Secondly, process simulation, right? So GenAI really helps create detailed simulations of manufacturing processing for testing and optimization. The third one is really around data generation. It's gathering this data, it's augmenting, it could do training datasets, improving model robustness. Lastly, the big one is really around insight generation. It's really analyzing the data to really uncover hidden patterns, insights, aiding strategic decision making. And then probably talking about really roles, you know, in manufacturing. So from an analytical AI, you know, it helps with data-driven decisions. And the difference with generative AI, it really helps with creative decision-making, proposing new ideas and scenarios. So just high level of really the difference.

Thank you, Eric. I appreciate those, some of those examples to get us started. Kathy, do you have some additional examples, or would you like to maybe enhance some of the comments Eric made about examples in manufacturing?

Sure. I think Eric hit on most of the main ones. I might just go in a little bit extra layers for those that are still wondering what entails in those each sort of categories. So for specific areas that I've— that come to mind for me, we already just mentioned about the written content augmentation and creation. So this can include generation, review, and translations of documents, methods, and procedures. It's training content, it's maintenance manuals to help with consistent execution for like the frontline worker, which in my intro I didn't put. So I spent 5 years on a factory floor and this was over 30 years ago. So when I think what's available now as technology and aids and tools to be used on the floor, that one stands right out.

Sure.

And boy, I could have used it back then. The data augmentation, bolstering predictive models to identify different patterns and relationships and anomalies and defects. The frontline worker I mentioned, the training and the guidance and how you can expedite that process and you can get your frontline workers up and working faster and keeping them more up to date. That to me, like I said, it's pretty personal there. Then you can go to like software coding. You know, there's the classification generating software codes to analyze and classify downtime, identify root causes. Code assistance can be used for generative AI, as we've heard. So that's huge for developers.

Mm-hmm.

And then I think the big thing is that there's going to continue to be additional use cases that are going to emerge as companies identify opportunities that exist within their organizations. Again, augmenting downtime, using series data, unstructured content. We're hearing about synthetic data. I'm not as familiar with that, but it's something I'm eager to learn more about. That's what I would add. Probably one other thing, I did pull up a little statistics because I like stats. I did take a look and Rockwell had put out their annual report. And regarding generative AI, they had stated that obviously the technology has rocketed over the last year and it's creating a demand for industrial applications with 83% anticipating using generative AI in operations in 2024, as well as generative AI ranked number 2 for return on investment. I thought that was an interesting part and I'm eager to hear what my fellow panelists think on that. That generative AI is the new number 1 area for technology investment over the next year.

That's some great statistics. Thanks for sharing that information. Scott, anything to add to Kathy's comments or Eric's comments right now?

Yeah, I think they covered it pretty well. My experience would weigh heavily on the developer, So Kathy mentioned, you know, like code generation. I've used that, and like my team has used that extensively to kind of help cross-train and just understand concepts that, you know, we're running into on the shop floor. That's definitely one that hits home and something that I've used extensively.

Excellent. Manpreet, any other thoughts on the applications of generative AI in manufacturing?

I think my friends here have done an excellent job of explaining what the use cases are. I would like to talk about a couple of them which are derived from what Kathy or what Eric mentioned earlier. So some of these use cases are pertinent to the operations in manufacturing, which is how smoothly your machines are running over time, how much efficiency you're seeing on these machines, are there any issues that are causing you to, you know, consume more energy? And how do you run your processes more efficiently? How do you use GenAI to simulate some of the process scenarios? Because traditionally simulations have been very hard to do, especially if you have to vary a lot of process parameters. So GenAI gives you that leverage to do things differently now. And I do see a lot of scope in terms of optimizing what we do on the shop floor, optimizing how people work on the shop floor, how machines work on the shop floor, how they communicate, machine-to-machine communications, machine-to-human communications. I think there's a wealth of tools that will be available very soon. So very, very optimistic about GenAI now.

Yeah, thank you very much for those insights. Scott, you mentioned you do some programming with GenAI today. What are some of the pitfalls and issues that you see in the use of generative AI?

Yeah, so maybe I'll start by just elaborating on my previous comments. So I think, you know, taking it slightly different direction than focused on the shop floor, but focusing on some of the support areas that are around the shop floor. So things like engineering, even maintenance team or facility team type roles. You know, the level of technology on the shop floor is only increasing, and of course that's commonly seen to simplify things for the frontline workers, which is great. The reality is that the industry needs to figure out how to support all that technology on the shop floor, and that's where like my role and my peers come into play.

Yeah.

So, you know, one of the things that we're facing is, you know, the same problem whether you're a manufacturer or a systems integrator, or a machine builder is just how do we troubleshoot, maintain, or even extend the technology that we have on our shop floor, right? So there's just— there's simply too much for one person to know. And that's where I think GenAI can really help. And I think that's what I was alluding to when I referenced, like, code generation. So, you know, I think I've seen that help at even just an individual level. So I'd like to share Just kind of a short example here. When approaching a new project or problem space, I always like to draw a comparison to my past experience, and I help define that and define the unknowns, often now by having a conversation, if you will, with one of the GenAI tools. Later this year, I'm going to be Participating and supporting some work on a project focused around electrochemical machining, which is very common. It's a very common type of equipment that's used in my industry.

Yes.

I personally don't have much experience with it, but after 20 or 30 minutes of— and I'll even share, I looked back and looked at my prompt. I asked AI, What is critical to the success of electrochemical machining? And, you know, learned very quickly about surface finish, corrosion, material removal, tool wear, like all this technical stuff that just getting, you know, kind of back into the headspace for like that type of process. So all things that are not revolutionary, right, things that you could learn out of a textbook, but I was able to do that in 15 or 20 minutes, not, you know, Paging through a textbook or coordinating, scheduling to talk with peers.

Right.

That's all like the happy path. I will say the issues she asked about are that the hallucinations are a problem. Just be blunt about it. I think that's probably something that the audience has already heard about quite frequently, but I feel compelled to carry that message. I think just to share one more quick thing here, I recall a time where I think of this anecdote quite often. I was working in the lab with one of my peers. We needed to just see binary representation of a number. something computers are good at, right? And we had an AI tab open in the browser, threw it in there. 20 minutes later, we found, you know, things still aren't adding up. We went back and found that the AI gave us the wrong answer. So, you know, that worked itself out pretty quickly. I'll admit that some, you know, basic critical thinking would have caught that error right away.

Right.

And I just wanted to share that story of like, you know, the bigger point to me is The ease of access of just opening up a browser and typing in a question is sometimes mismatched with the correctness of the answer you get back. So, you know, always encouraging people to just, you know, apply a degree of skepticism. But give yourself time to learn how to, you know, write prompts and interpret the answers, and you'll be fine. It's a fantastic tool. But, you know, definitely in the Trust but verify stage.

Yeah, I think that's great advice. I think we hear about it today a lot about this idea of how it can help us, but the fear that it's going to replace people, at least as we can see it today, as you described, we still have to be vigilant and know that the answers that we're receiving make sense. I appreciate the insights there. Manpreet, can you Talk for a minute about how Genitiv AI can drive cost reduction and improve energy management in factory operations.

Yes, of course. So if you look at the factory operations, most of the cost, the lifecycle cost of running an operation is not the capital expenses that are incurred on setting up these machines, but a lot of it has to be related to the energy that the machinery consumes. A lot of it is in servicing those machines. So what we are trying to sort of work towards is making sure we tackle both of these problems at the same time. So just to give you some stats, the electrical machinery which is deployed, which is run by motors, is responsible for consuming 50% of global electricity consumption, which is a huge deal. And The other fact is when machines are running inefficiently or have a fault, we've seen in various case studies that they consume at least 10% to 30% higher energy. So these are costs which go unnoticed in traditional manufacturing setups because a lot of times people are not monitoring their energy use. A lot of times they are not able to segregate what is happening because of the process that they're running and what is happening because they are not maintaining the machines properly. So what we do is we have a 6-in-1 sensor which can hear the signals coming from the machine. So I, in order to link it to the, to the new gen AI universe, what we capture is vibration signals, acoustic signals, magnetic signals. All of these are like sounds of the machine. So what we are really hearing is the language that machines speak and then trying to analyze what they are trying to tell us. So we have already started using some of the GenAI tools that are available to detect some of these failures or fault modes, and we give early insights to our users to help them maintain machines quicker. And we also have a dedicated energy, uh, linked, uh, dashboard which shows how energy consumption is trending up or down over time so that our insights, when we tell them that maybe this is because of a bearing issue, you can just go and lubricate that bearing. And we've seen tremendous improvements in energy consumption and we have already saved more than 20 million kilowatt hours of energy last year. So that's the opportunity and it's just, it's less than 1% of what we can do with just our sensors. Imagine what the world will be able to achieve if this was scaled up to all the machines. And the other element of costs is about really preventing downtime on the machines because downtime is not just a cost in terms of repair or maintenance of the machine, but it is also a cost in terms of lost production, lost opportunity. So that is what traditionally predictive maintenance has been about, but we are moving away from traditional maintenance approaches because we feel that by delaying your maintenance, you are probably saving a little bit on maintenance costs that you would incur, but you spend a lot more on energy. And that's what is harming our planet today. So we believe very strongly that GenAI use cases in both these scenarios are worthy of implementing, and we've already seen great success there. I hope more and more people and people in the audience look at some of these options available and they start thinking critically about how to use GenAI in those scenarios.

Thank you for that. While we're— it's been a great discussion thus far about generative AI. I would like to just mention to the audience, please, if you've got questions for the panel, we'll have time at the end. to take some of your questions. So if you please put those in the Q&A section, we'll be glad to try to get to as many of those as we can before we end this morning. So Kathy, could you talk for a minute about a little bit more about the benefits and really more about why more folks aren't implementing generative AI? What are some of the barriers and concerns that you see that manufacturers have and how can they overcome them?

Sure. Obviously, we've already talked about some of the quick wins that we see with generative AI, which help reduce time and simplify different methods or procedures. Things definitely become more complicated if you're looking to curate and train large language models to assist in programming automation. This will require significant more time and investment, and these investments can be hindered by excessive backlogs of systems upgrades, integration requirements, skill development that raises concerns about generative AI scalability as a whole. So some of the specific areas that I think prove challenging, and I'm really interested to hear what the rest of the panelists say because they're actually in facilities, is the change management that's needed. That includes, you know, a consideration for the risks and the challenges that threaten to impede it across several areas. And some of those areas are like the data information access, how you manage and the security of it, designing new processes while you're refactoring old ones, and ensuring the trust of the workforce to adopt the new digital experiences that they have. To overcome some of these barriers, I really believe that the leaders need to assess generative AI's advantages, the requirements, the limitations for their organizations, and they really need to engage with technology providers for guidance. I'm always a big advocate for collaboration. This is a new field, a new area people are moving rapidly towards it, but you don't have to do it alone. So reach out to those experts that have the skills and the advice because you need someone to help maneuver you through this process. You can look for providers that offer like a managed generative AI service with a variety of foundation models that can be augmented with your own data, as an example. Just be careful. Everyone's out there offering solutions. Just do your due diligence. As Scott said, there's a lot of considerations that go into it and you have to be aware of the pitfalls.

Thank you. Eric, do you have some thoughts on how companies can overcome some of the challenges with GenAI?

Yeah. The first one, and we've implemented here at Iris because here we have a software team, hardware team, and then we actually have manufacturing of infrared windows. Create an internal generative AI committee or center of excellence, right? Just really have a team that is really focused on doing due diligence, understanding what use cases, you know, where is the data, do we have enough data? And so I think that would be a great place to start. And then on top of that, really understanding data privacy, right? So once you understand your data and what you're gonna, you know, how those— that data is gonna be trained, just making sure you understand you know, just from a privacy standpoint, you know, do you have proprietary information? Are you going to use the open, uh, you know, website or network? Um, so just really understanding those two. But again, you know, if you start with just a, a center of excellence or an AI committee, you're really going to start having some, you know, more structured way to actually implement generative AI into whatever process or business model that That's a really great suggestion.

Manpreet, Scott, any other thoughts on overcoming some of the barriers and concerns? Why aren't more people using GenAI today? I mean, Kathy mentioned earlier the statistic that most folks are planning to, but how do we really accelerate this opportunity for manufacturers?

Yeah, I think I would like to add in there. I think, you know, Challenges are both cultural and technical, and that, you know, does combine to create a pretty large barrier. You know, I can expand on some of my earlier comments from, you know, like learning about a new manufacturing process. Well, I could, you know, hope that the day is coming where I can learn not just the basic principles of the process, but I can then prompt AI to say, okay, well, As a manufacturer, what part numbers go through that process? Have we had any? Have we had any quality issues that are you know popping up again in a trend, or you know tell me about the specifics of how that process is operating within my plant. We're not there yet, I don't think, but we're we're definitely heading that direction. And you know doing things like we just recently, you know I saw. So I'll publish— we have an AI policy that's been circulated around the engineering teams and just establishing some of those guidelines and understandings of how you use the tooling is— I think that's got to come first. The technical stuff can come second, shortly behind, but just understanding the technology and how you're willing to use it is really important.

Yeah, I think that's a really good point. One of the things that I think all companies should be doing today is really publishing a GenAI policy. And you said it, but the reason for that in particular is if you've got confidential information, you don't want it in the public space. And even though we all want to— a lot of people want to just play, but playing in the wrong way can be quite dangerous to your company. So I think that this idea of a policy is really important.

And it's something that I recently learned about or really thought of for the first time was it's not just the public space, right? There's like internal walls as well. You don't necessarily want everybody to know what everybody else is using the tooling for. You know, the tooling can respond with data that is potentially sensitive in nature that, you know, would be appropriate for some people but inappropriate for others.

Yeah.

My point would be—

go ahead, Kathy.

I was just going to add that we're talking about organizations as a whole, and technology is advancing so quickly. Generative AI, when you think about ChatGPT landing on all of us just over a year ago. And for the audience, you got those little buttons at the bottom. Give a click if some of those early adopted tools you're using. It'd be great to just see how many are actually using them currently in their work environment. I think some of the considerations that are really hard for organizations right now are around having to create new standards and governance models in the production systems that align with— there goes the hearts, thanks— that align with what they want to do with generative AI. You're protecting your IP, and what does that mean to the organization? So just as another little shout out, there was a report put out recently, and I'm mentioning it because Bennett was mentioned in it, and it was by Gartner, Simon Jacobson, and it's called What Generative AI Means to Manufacturing Strategy Leaders. So it's something to go take a look at because he actually touches on some of these points and just extra consideration. I think, you know, we all have to do our due diligence and be aware, and this is just one way to do it.

Thank you for that. Manpreet, any other thoughts on this issue of overcoming barriers and concerns?

I think the way I think about using GenAI solutions for any industry per se is, one, we need to understand what problems are we trying to solve using GenAI, because you would want to prioritize your biggest problems first. And the other thing that you need to understand is the limitations of GenAI, like Scott mentioned, that there is a problem with hallucination. So how are you going to build checks and balances to make sure some of these things are not impacting some of your critical decisions? I think those 2 things, if you sort out internally, then you need not build the entire know-how or IP in-house. There are enough and many partners available who you can rely upon to speed up your digitization efforts, because I think everyone needs to understand they're good at doing something. So if they're running a big industry or a big factory, they are good at doing those kind of tasks. The moment you start thinking about doing it internally, then you get into all the issues of building a team, a solid team who can bring those solutions quickly and invariably getting the right teams in place and building it in-house takes a lot of time. So I would suggest think of it in terms of what's your biggest problem today? Are GenAI solutions good to solve that problem? And then just find the right partner to help you with it and speed up your digitization efforts because the cost of not doing it is much higher than building it in-house or build versus buy decisions.

Yeah, thank you. And you mentioned speeding up and it's a It's a wonderful thought. Kathy mentioned earlier the issue with people and change management, and quite often that speed is determined by how fast your folks in the factory are willing to accept and adapt themselves to the new technology. What are some ways that we can do that in a way that's appropriate? I mean, we hear in the news today, for example, There are enough articles out there about GenAI is going to take my job and GenAI is Big Brother and that thing. I'm interested in the panelists' thoughts on how do we handle the change management, the people side of this? Because quite often the speed is determined by how fast the folks are willing to change. Maybe Kathy, you start us off and then we'll go around the room.

Sure. Coming from my background, my experience on the factory floor, you need to engage with your work. So the people that are doing it, let them have a seat at the table, get that buy-in with them. When they realize how these are tools and these tools are going to help them augment their roles and help them do better quicker— I can tell you, as a person on the floor that was running in a department where keeping machines running meant I wasn't offloading extra tires to have to work them back into the system— you give me a tool that will help me manage and see what's coming so I can make good choices and decisions ahead of time, you've got my buy-in.

Thank you for that. Other thoughts from the panelists? Anybody who'd like to share a thought next?

Yeah, sure. I think I've had to remind people at times that we live in a world where scary headlines get clicks, and that's just a fact of life at this point. The sense that AI is coming for your jobs, That certainly gets people's attention. I think I'm quick to point out the reality of, like Kathy's saying, look, in real life, like, here's how we're using AI in our organization and here's how it's making people's jobs easier and better, not taking over for anybody.

Yeah. I'd like to add. I always say, or I've heard, AI is not going to take Yeah. So exactly what Kathy said, right? Get them to the table and have them understand that this is a tool that is going to help you, right? If you think about, you know, conditioning, monitoring, or maintenance, right, you might have someone who is fairly new to the industry, but being able to use some type of generative AI assistant to walk them through through to how to fix a machine or something. So it's only going to help and improve their skill sets, just really getting that buy-in. But typically there is always that affinity to change, but I think it's getting easier and easier, especially with, you know, ChatGPT being open to much, being able to have it on your phone as well.

So, so, uh, with every new technology trend, we do hear this, uh, this, uh, phrase, right? It's going to take your jobs. But so far what we have seen historically is that jobs have evolved, they are not necessarily taken away. So I, I feel the same way about GenAI too. Jobs will evolve. You will need to rely a lot more on some of those decision-making tools that are going to be available and be smarter about doing your jobs. It's going to change the productivity levels drastically. We are already seeing that in tech industry where a lot of software development has sped up just because people have started using some of those GenAI tools to do the boilerplate for them quickly or do some solutioning quickly. But at the end of the day, if it's a very specific problem that you're working on, a very broad tool like GenAI may not be best suited because it doesn't have those critical thinking functionalities or things where it can bring out new solutions. It's, it's trying to reiterate what's already there. And the other way which we think is important in terms of change management, because that's where we started off some of this discussion, is also to understand what the needs and wants of our end users are and make sure that the products that we are building are tailored to ease their job. So, uh, you will always have the problems like hallucinations and problems with AI tools of inaccuracy to a large extent. Some of it can be mitigated because a lot of processes would have some underlying physics involved or underlying, uh, logical steps involved to solve, uh, which can still be automated. But with GenAI, you would still need some human oversight on the decisions.

Yes.

that are being generated. From that perspective, it's important to keep your end users in mind and make sure that you ease their workloads, make sure they are able to do their jobs more efficiently. And that's, that's what my biggest hope is from all the new AI and GenAI tools that are coming up.

Those are some really good thoughts. Maybe I'll just add, when I started my manufacturing career, Nearly 40 years ago, when I entered the plant for the first time, we had one computer in the plant, and everybody was afraid that computers were going to replace people. And what's real interesting about that is 4 decades later, there's even more employees in this plant. It's expanded, and it's more productive than it ever was, but the number of jobs has actually increased because the plant's productive. And so I think we We tend to think about what sometimes the worst case, but when in fact, if we do this correctly and focus on what we're trying to accomplish, it can be very positive for the future. I just want to add that story from my own experience. Maybe we'll pause here and begin to take a couple of questions from the audience, and then we will give some final thoughts in a few minutes. We have a question from the audience, and that question is, are there known case studies of GenAI being used for simulation, especially for discrete event simulation within manufacturing operations? Who's got some experience with simulation and might take that question?

I can talk a little bit about it because we are still in the early stages of simulating some of the signals that we receive from the machines. We have simulated a bunch of signals which correspond to different types of faults that machines can produce. So we have a dataset of more than 5 years now on thousands of machines, and we have seen a lot of events in the past. So that gives us a very wide distribution of data in terms of what can happen and what can't happen. What we are trying to simulate, or what we have simulated, is How these events lead to energy consumption. Does it increase energy consumption? Does it decrease? What kind of maintenance practices lead to better outcomes? So some of these things we are using to generate simulated data. So those are some of the use cases which are still internal. We have not published any of these yet, but I'm pretty sure there are other companies who have who have published some of the case studies. I don't have specific examples in mind, but it's one of the leading use cases that we have for GenAI.

Thank you, Manpreet. Are there anybody else, other examples of where they've seen simulation, GenAI use simulation? Eric, go ahead.

Yeah, so definitely from around maintenance. So if you have a lot of electrical assets or just assets that need Weekly, annually, biweekly, you know, maintenance, utilizing, having those routes and then using generative AI to enhance those routes to make sure for efficiency, to be honest. But I've also seen it in actual data center, you know, building out data centers, right? Simulating, you know, hey, this is the best kind of routes or where you want to set up from a data center perspective of where you put your electrical assets and so forth. So a lot of use cases around that.

Thank you, Eric. We have another question, and this question says, what are some of the ways generative AI is used specifically in IIoT?

Sure, I can take that.

Start that?

Yeah. So essentially in IIoT, what you're doing is putting some sensors on the machines or on your process those lines and trying to get all the data points over a time, uh, over a, you know, temporal view. And the way to think about generative AI in this context is what you're really capturing is the language of the machines. Uh, what you need to now start doing is, if you, if you know the traditional or the GPT kind of models, they look at a text and break it into tokens. Each of these tokens then has certain embeddings Similarly, machine signals can be broken into features which are giving us credible information. Then because you're taking those signals over time, there's enough context available on each of these signals. You can completely use a language model transformation tool which will take in these inputs, which is like what machines are telling you, and then try to devise or try to generate those insights that, okay, The machine is telling that there is a bearing fault developing or a shaft fault is already there. Your energy consumption has gone up. All of those can be thrown out in a very human understandable manner. So traditionally, if you look at IIoT, it's a bunch of graphs, numbers which are displayed on various dashboards. Somebody needs to interpret those. I think with new tools, a lot of that interpretation problem will be offloaded to some of the GenAI tools that are going to be available. And for this kind of work, you need not implement huge models like GPT, which needs billions of parameters, because this is a very specific use case you're talking about. The number of embeddings, the number of context kind of things that are required may be very limited to the use case that you're trying to solve. So from that perspective, I think we'll see great improvements in how GenAI is specifically improving the industrial IoT use case. The other thing is the knowledge gaps, right? So we have seen in a lot of our customers that they don't build, they don't have the, you know, wherewithal to build great reliability teams in-house. So if they're working with maintenance floor workers, how do we make sure that these workers get the knowledge quickly, they are able to extract the relevant data from the machine, manuals or knowledge that is available in-house or knowledge available on the internet and get to a very specific action item. So if XYZ is the insight, ABC is what I need to do. That kind of simplicity will definitely solve a lot of use cases in industrial IoT. I hope that gives you some clarity.

I'd like to add, so if you think about industrial IoT, you start with condition monitoring, then you get Predictive maintenance. And then you add on GenAI. So exactly to Markit's point is, you know, you have this alert, you have this, you know, this piece of machinery might be coming down or it needs a part replaced. And so now you have this assistant that is gonna give you detailed information of how to fix that part. It can order parts for you. It can create a work order for you. So it's just taking it to another level rather than saying, hey, we know in this time that this machine or this part will go down. You're just able to actually in real time, you know, have an assistant order parts, you know, create work orders and so forth.

Excellent. So we have a couple of folks, Kathy, that have asked you to please repeat or share the The article, the link to the article that you quoted earlier on, could you repeat where that came from again?

So there's 2 reports that I had stated, and I'll just have to— I had them on a little side note, so just give me one second. So one was by Gartner, and it was— it's titled Innovation Insights: What Generative AI Means to Manufacturing Strategy Leaders. And it's by Simon Jacobson. The other reference was to the annual report by Rockwell Automation. So Blake Moran, the CEO, gave his yearly update, and there was a lot of generative AI, and that was the statistical information at the beginning of the session.

Thank you so much for that. We have another question. It says, if you're using AI to manage industrial manufacturing processes, How do you access real-time data like temperature, pressure, and other things like that? Scott, do you have a comment on that maybe?

Yeah, I— so my first thought there is typically you would be collecting those low-level data points and then brokering that data to the AI engine. So you would have access to both.

I know, Manpreet, you also have some thoughts on that.

So we do collect data from the machines, especially pressure, vibration, acoustic, magnetic kind of signatures continuously. It's not continuous in the sense that every second we are taking those signals. It depends on your use cases. If you're working with a machine which has a lot of transients and you're working with a machine that needs automation or PLC controls to switch it off if anything happens. Those kind of use cases require you to continuously monitor those machines and build the automations to take actions automatically. But a lot of times the use cases don't need a continuous evaluation of the data. So that's where you would want to optimize between your sensor's battery life, the amount of data that you push to the cloud, the amount of insights that you're generating. Because all of this is an expensive operation at the end of the day. You have to really think critically about what's continuous for your use case. Do you need data every second, every 5 minutes, every 10 minutes, every half an hour, or every day, or every month? Because some of the things will not change over seconds. For example, if you're running a consistent process, the machine is going to run with the same data points for hours and hours together. So even if you're generating one insight per day, it might be sufficient. But there might be other cases, for example, some of the turbo machinery that might be there in your plant, a small unbalance can cause massive catastrophic failures.

Yeah.

There you would want to monitor those data points consistently and make sure you have all the tooling in place to stop the machine immediately the moment you notice some deviation. So it really depends on your use case. In terms of tools, yes, they are capable of continuously monitoring, capable of monitoring intermittently, capable of monitoring only when required. So it totally depends on the use cases that you want to solve.

Yeah, excellent. Thank you so much for that. Um, you know, these have been some really great questions. I want to make sure we have a couple minutes now for our panelists to be able to share some some final recommendations for those wanting to implement GenAI. Panelists, we have plenty of time, please feel free to share your final recommendations to our audience today as we reach the end of our time here together. Maybe Scott, you'll start us off.

Sure. I would recommend wholeheartedly anybody that has not interacted with generative AI engine yet. So just go ahead and make an account if you need to. Set that as one of your tabs that open automatically when you open your browser and just start to use it. Start to learn it, learn how to prompt it. Use it for anything and everything just to kind of learn how it works. And I, I'm saying that because I myself was pretty skeptical about, you know, how, how conversational it could be and like how you can just use it day to day. And I have made the plunge. I am recommending what I do personally. I have that thing open all day, every day, using it for all sorts of odds and ends. Like I said, paste something in just to get— if you're in an engineering role like I am, paste something in and get an understanding of syntax or whatever. I asked it all sorts of oddball questions, not necessarily engineering-related, and it's fantastic.

Can you say the name of it again? That part went really quick, and I don't know if the audience caught it.

So I, you know, because I work with code a lot, I tend to prefer ChatGPT. That, of course, we've all heard of. There are other AI engines. I think they just recently renamed Bard, but Google has, has theirs. And yeah, there's others out there, of course. But yeah, I'm Most familiar with ChatGPT.

Thanks, Ken.

Kathy, any final thoughts or recommendations for our audience?

So for me, as I mentioned earlier, I think it's collaboration. You don't have to go this route alone. At Bennett, you know, we are matchmaking, I like to call us. We align subject matter experts, solution providers with people that have, that have needs. So we have solvers and we have seekers. In our solvers, we have a company, it's called TXI, and they're working with an organization around the RAG system. I don't know how many of our audience members would know what it is. It was pretty new to me, and it's the Retrieval Augmented Generation. It's a set of approaches for helping the language models perform more reliably. Excuse me. It's to help deal with what Manpreet has said many times about the hallucinations. It doesn't know what it is and making up. There's so many different advancements that are still happening. It's pretty difficult to keep up yourself or your organization. As Eric said, put a team together and get a whole pile of people just working together around this, and you'll be good.

Thank you very much. Eric, your thoughts, your final recommendation to our audience today?

Yeah, so, you know, if you've started your digital transformation or your smart factory journey, you know, just seeing where, you know, generative AI plays or enhances some of those solutions, um, I think is a good start. And again, back to Kathy, just center of excellence, getting an AI committee, uh, kind of formed. But it doesn't have to, you know, exactly what a lot of people are saying, you know, if you're not, if you haven't done anything with generative AI or AI, you know, start creating employee training videos, you know, start utilizing with writing standard operation procedures or policies. You can start on a very basic level where you will see the value of utilizing generative AI. You get familiar with it and then you just start building on top of that.

Thank you very much. Manpreet.

I would suggest that figure out what are the top problems that you want to solve because there are many solutions already being built. So you need not reinvent those wheels. Figure out what's your top 2 or 3 places where you're facing maximum issues and prioritize those for your digital journeys. And then I think a better reward for your efforts would come from finding the right partner to help you with and making sure that you understand the limitations of what GenAI can do or can't do for your problem. I think that bit of thinking and assigning some resources to understand what GenAI can really do for you is important. And then my recommendation to anyone who I speak with is not to build it in-house because that would just delay and increase your costs drastically. If somebody has already built a solution, use that and flow with it.

Yeah, thank you for that. I've heard lots of great recommendations. Just to summarize some of them, and I appreciate this, play. In other words, experiment a little bit to understand the potential. Be vigilant. We know that it's not foolproof. It doesn't always— the hallucinations don't always tell us the answers we want to know. Also be vigilant in that recognize that If you use a public GenAI, you're giving your information to the whole world. Recognize confidentiality. Something we mentioned earlier, have a good AI policy so you know that you're ensuring there are some— what the boundaries need to be. I appreciate the conversation around involving the shop floor. It's so important from a change management perspective to make sure that we start Involve the folks who are going to use it right from the beginning so that they can become comfortable. And then the other piece that I guess I heard a couple of you talk about is start with what the problem is that you were trying to solve as well. You know, sometimes we go and play with technology for technology's sake when in fact we need to use technology to solve our problems. And if we have that in mind, that's quite often how we can succeed. So, and then finally, I think I heard also You don't have to be an expert yourself. There are plenty of great partners out there with tremendous experience. And in fact, the technology is moving so quickly that it's much more cost-effective to rely on someone who's monitoring that because there's a lot of technologies that, in addition to GenAI, that you've got to be engaged with in today's world. So I just want to thank our panelists for their amazing insights and knowledge. It's been really a great hour that we've spent together. I want to thank again our sponsor for this webinar, NanoPrecise SciCorp. Your support ensures content to share. So again, thank you, Manpreet, to you and your company. And thanks to all for attending and for your great questions. And we didn't get to all the questions, but, you know, that's— hopefully we got some of the more important ones or some of the more interesting ones, and they helped The rest of you we couldn't get to. So we're getting ready to move on here at the top of the hour to our next session entitled Smart Networks, Smarter Factories: Building the Foundation for Next-Gen Manufacturing. So we look forward to seeing you in the next session. I want to wish everybody a great day.