In episode 05 of the Let's Connect! Podcast, Nikunj Mehta, CEO and Founder of Falkonry, joins us to talk about his Intelligence-first approach to IoT.

Nikunj Mehta founded Falkonry after realizing that very valuable operational data produced in industrial infrastructure goes mostly unutilized in the energy, manufacturing & transportation sectors. Nikunj believes hard business problems can be solved by combining machine learning, user-oriented design & partnerships. Prior to Falkonry, Nikunj led software architecture & customer success for C3 IoT. Earlier, he led innovation teams at Oracle focused on database technology & led the creation of the IndexedDB standard for databases embedded inside all modern browsers. He has contributed to standards at both W3C & IETF, & is a member of the ACM.

Interested in connecting with Nikunj? Reach out on Linkedin!

Falkonry is the innovation leader in Operational AI. Falkonry enables predictive operational excellence for manufacturing and defense organizations by detecting and predicting events before they impact operations. By applying AI on real-time operational data from plants and field systems, Falkonry solutions deliver significant improvement in production uptime, quality and yield without requiring data scientists or data engineers. Falkonry products easily scale across the enterprise either on-premises, in the cloud or at the edge, and are optimized to run on major cloud platforms including Microsoft Azure and AWS. Falkonry’s Operational AI based products are built for intelligence-first operations teams to reduce planned and unplanned downtime, discover anomalous conditions, capture expert knowledge, provide root cause analysis, estimate time to critical events like time to failure, remaining useful life and show all this information through an executive dashboard making it easier for all stakeholders to understand and apply this analysis.  Follow Falkonry on Twitter!

Key Questions and Topics from this Episode:

0:00 Show Introduction

1:05 Nikunj Mehta Introduction

5:50 Industrial IoT and Downtime Avoidance 

7:00 What is IoT Intelligence-First Design? 

8:30 How does beginning without data work?

10:20 Anomaly Detection and System Faults

14:50 Digital twin as a modeling tool

19:45 How Intelligence-First design speeds adoption

24:00 Final Thoughts

Transcript:

- [Announcer] This is the IoT For All Media Network.

- [Ken] Hello, friends in IoT. Welcome to Let's Connect, the newest podcast in the IoT For All Media Network. I am Ken Briodagh, Editorial Director for IoT For All and your host. If you enjoy this episode please remember to like subscribe, rate, review and comment on all your favorite podcast and platforms. And to keep up with all the IoT insights you need, visit IoTforall.com. Before we get into our episode.

- [Announcer] The IoT market will surpass $1 trillion in the next few years. Is your business ready to capitalize on this new and growing trend? Use Leverage is powerful, IoT Solutions Development Platform to efficiently create turnkey IoT products that you can white label and resell under your own brand. Help your customers increase operational efficiency, improve customer experience or even unlock new revenue streams with IoT. To learn more, go to IoTchangeseverything.com. That's IoTchangeseverything.com. Now Let's Connect.

- [Ken] My guest today is Nikunj Mehta, the founder of Falkonry, not the sport, although awesome name, but the company. And we're gonna talk today about Intelligence-First approaches to IoT and sort of how that helps with spreading adoption and sort of encouraging the growth of this industry that we're all working in. And who knows where we'll go from there. So, let's find out. Nikunj, welcome to the show.

- [Nikunj] Thank you for having me, Ken. It's a pleasure to be here.

- [Ken] The pleasure is entirely mine as they're about to find out, everyone that's listening too. But before we jump in, why don't you tell us a little bit about yourself and about Falkonry. I demand to know why it's called that 'cause that's awesome .

- [Nikunj] Yeah, that's not too bad actually. I think that's a great segue into our conversation as well. So Falkonry, as we all know is the pairing of a beast and a man, right? A beast that is far better than any other living organism on the planet. In fact, for those of you or those of our audience who do not know this, falcons are the fastest. They are the most powerful when it comes to the force multiplier of their own weight and they have the sharpest vision. And so falcons have always captivated our attention because of these qualities since time immemorial. And Falkonry as a sport is also an ancient sport. But mostly because the challenge in it is how do you train or tame rather a wild beast to hunt for you? 'Cause if you can do it well, your force just multiplied a hundred times over.

- Sure

- And in many regards, the Falkoner's job is really to understand how to get the beast, to do its bidding. And at the end of the day, the beast is gonna fly away or perhaps a season in, right? You cannot hold onto it forever and neither should you, right? It's just not, it is not right thing to do. And so Falkonry is itself a place where all of this happens. So if you think about Falkonry the company it's the place where Falkoner meaning a subject matter expert, is steaming the wild beast which is algorithms and computers, to go do the hunting which is finding all the problems in their manufacturing and production on behalf of that, man. And the only difference is that once you get that Falkon in your Falkonry, you're not gonna let go of it.

- [Ken] I suppose that's fair. Although from what I am told occasionally it'll bite you back. So, I believe I've stretched that metaphor now, as far as the joke will allow me. Tell us a little bit about yourself. How did you end up founding Falkonry and getting into this sort of AIoT business?

- [Nikunj] Yes. Actually I've been in the AIoT business for 11 years now. I started in this space with C3 IoT, now C3 AI in 2010. And ever since then, I've been captivated by the opportunity as well as the possibilities of the space. Frankly, I must also admit that my father, my grandfather both hail from steel manufacturing. And so I've visited plenty of plants in my childhood and understand what it is like to be producing and dealing with all of the physics that arises in that environment. But when it came time to figuring out what do I do next? This was 12 years ago. I wanted to go into a field that I could feel happy about for a long, long time to come. And I think that is what manufacturing and production is like for so many of your audience members. They enter manufacturing when they just finished high school and they've never left it, probably never will. The beauty of the space is that software did not give it the sort of the fair treatment if you might. It's a hard field. And so we've done everything we could to make our lives more comfortable, our tasks more convenient but software left the manufacturing space by, for a while. It's great to see all this attention back. And that's what Falkonry we started with, is to create an even playing field for all of the manufacturing professionals to take advantage of the digital exhaust in their manufacturing context. And we believe that the continuous improvement practices like Lean manufacturing, Six Sigma, continuous improvement, operational excellence, these methods they have been perfected for a long time. And yet there's so much opportunity still left, even though it may only be a few percentage it's still going to translate into hundreds of billions of dollars.

- [Ken] Well right. And I think that sort of, the industrial space has been doing a good job over the last say, five years of starting to integrate a bunch of IoT technologies. Actually, in the realm of sort of downtime avoidance, I think that's fair to say. And focusing on that, obviously it's a huge issue, if unexpected downtime is, hundreds of millions of dollars a year or more, but I think the the rumblings are coming to want more than that, to actually get closer to this promise predictive stuff that people have been promising, but not executing on for so long and various other places where especially manufacturing and supply chain and that piece of the industrial IoT really is, in need of machine intelligence, I think. Where you're able to get to the outcomes that humans are bad at finding and that machines and algorithms can find. So is that what you mean by sort of intelligence first design is looking at, how analytics will get us to that point, to those predictive and other outcomes?

- [Nikunj] I will give you a very simple analogy. You go to a restaurant so you can order high quality meals and just enjoy it right away, right? You don't go to a realtor buy land and go to the university find some seeds and wait a few seasons before you will have your first meal. Unfortunately, in a lot of the IoT space data first approaches are very much like going to the realtor. We need to go to restaurants. And Falkon leaves approach is that intelligence first approach where, we've already curated the methods by which we find insights to improve uptime, to improve the quality of our production and to increase yields from maintaining tighter control over the production processes, and that these methods apply and work regardless of the asset or the process, and regardless of the industry. And that is what we mean by an intelligence first approach and approach in which, we can act upon the insights within a matter of days from when you got started. And that tells you a lot about what actually happens in your industry and in your manufacturing, and do something about it rather than wait years and years to have the opportunity to

- [Ken] So, pardon my skepticism, but it's sort of literally my job how exactly does that work? Because it's pretty, at least my understanding of data analytics is that it requires a certain mass of collected data. And the conventional wisdom seems to be that, that needs to be pretty specific to the conditions under which it's going to be operating. So how do you get that data without collecting it?

- [Nikunj] Okay, so the beauty of this is the premise manufacturing which is that, 90% of the time, 85% of the time and sometimes maybe 97% of the time you are doing productive work. Machines are doing the right thing and your production is not suffering constantly. Because of that, whatever it is that you're doing you're running many different recipes through the line in a single day or you're producing the same product, hour after hour after hour, either of those two scenarios. Your machines are telling a story in terms of their digital exhaust. And it is possible to find out what story they are telling if not in its completeness in the large parts, within a matter of days. And that is possible because, most of the behaviors that arise and that are captured in those sensors can very easily be brought, can easily be organized into what we call conditions and patterns. And these five things can be learned over just a few days, you don't need to have collected years work of data to understand what happens. And the mainstream of our customer base doesn't have years worth of historical data to work.

- [Ken] So doesn't that introduce the possibility of, thinking that anomaly is regular operation though? Well, you know, if you're collecting data for say, three, four days to start to find these patterns and whatnot, and anomaly appears. It sort of-

- [Nikunj] Beyond three to four days.

- [Ken] Yeah right. You haven't measured it. Or, to me even more weird is that and anomaly has occurred previous to your starting the recording process, and now it's treated as regular. And so, three months, when that anomaly becomes a broken machine you have very little idea of how that happened. Is that actually a possibility? Am I imagining something?

- [Nikunj] Yeah, so you brought up a few different points here. Let's try to break it down. So first of all, if it is really unusual even if it has occurred and included it in the recording it will be called out as unusual. So, there is no given that whatever we started with the period of time when we are recording, everything is acceptable. That's not an assumption. So we have to look a bit skepticism at the recording to know whether or not there is something unusual in it. Number two, if the unusual happened prior to your recording then the very first time that that unusual occurs it's gonna be caught and it is gonna be reported. And the reality is, in fact, this is one of the key things for us to think about Ken, if your production is so regular, and if anomalies are sorted, then naturally the question is, how are you gonna do anything about it? Because every anomaly is unique. In fact, we were working with semiconductor manufacturers for whom the same anomaly does not occur more than once in even two years. So then how are you gonna deal with those kinds of highly controlled manufacturing? On the other hand, if an anomaly has occurred prior to your starting the recording, you still have to deal with. So the challenge at the end of the day becomes, why would anybody do industrial IoT, if it is not for intelligence? And if they are gonna do it for intelligence how long do they have to collect data and give it to data scientists and engineers for them to get any value out of it? The intelligence first approach is premised on the fact that operators in their day-to-day manufacturing efforts should be able to see whether the intelligence approach is giving them any value. And they can only do that if they are able to constantly see what that intelligence is telling them that they did not know about. And the longer we wait for the two to be put together, the shakier ground we are.

- [Ken] That's fair and, are you using any sort of large existing data sets to inform the models and things too? I imagine that that's a resource out there that would help shorten this process as well.

- [Nikunj] Well, you know, fortunately or unfortunately that is not really an option.

- [Ken] Okay.

- [Nikunj] And I say, unfortunately, because every plant every machine, every industry is different. So you can just take it from one and apply to others. People are trying, they can do that maybe for a certain class of equipment, like pumps and motors, okay? But you're talking about so many possibly, hundreds of thousands of different types of industrial equipment how are you gonna have the library for all, right? So we can't. But on the other hand, Falkonry has been fortunate enough that customers see the promise of this approach and its performance to say, "Hey I want to apply it to my equipment as well." And in the process, they validate the approach that Falkonry has created that works for all the equipment. And so at the heart of this is the same approach that was used to study the human heart behavior. And then was later used to study the human brain behavior, right? And so it's signal processing at the end of it. And signal processing is a mathematical approach. So it is stemming from the foundations of math and electrical engineering they have been around for decades.

- [Ken] It's really, really cool. I love this idea of sort of being able to say, "Now we know how it's supposed to work" "anything outside of these parameters," "logically adjusted is abnormal." So let's figure out why and solve that problem. Instead of saying, "We've recorded every possible abnormality," "and now we know what they are," because, that does seem sort of impossible.

- [Nikunj] Exactly.

- [Ken] My listeners are gonna get so tired of me saying this but, it seems like everybody I talked to, I go, "Okay, this seems like a natural case" "for leveraging digital twin to experiment on" "and play with while the actual system is operating". And I'm gonna say it again, sorry guys because it seems like once you've created this model you can start running a modeling twin to allow for, you know, you let's try something, let's experiment, let's see what kind of anomalies maybe we can expect now in the physical system and avoid them in advance. And that sort of starts to lead us toward this predictive situation that doesn't seem to exist yet in the industrial world, but, everybody wants so desperately that they sometimes pretend it exists .

- [Nikunj] Right. So yes, we have actually looked at, for example NASA put together a simulator for the turbo fan that is used for example, on rockets and such.

- [Ken] Sure.

- [Nikunj] And the simulation has existed for awhile. You could inject some anomalies in it, like there are sensors going bad or falling occurring or spalling off its barrier. You could include some of those defects. And you could produce the data from that simulator and Falkonry for what it is good at will tell you that it has seen those behaviors very first time it'll tell you that this is an unusual behavior or what do you call it. And once you give it a name, it will tell you if it happens again with it, right? So it does that. And one of the key things for predictive that people have been asking for for awhile is tell me how much longer do I have to get this problem addressed before I have it failure on my hands. So not only that, unusual behavior and identifying what state we are in but telling you how long you have to go before this things will break down. So we can do all of that, right? And you can do that from simulation. However, you cannot simulate every possible behavior. It is still too distant a goal for many organizations they don't have the people to do that. They may not have the models because they never paid for those models. Many manufacturers are not gonna buy the design they're gonna buy the machine. So yes, we would like to have the ability to simulate the twin and maybe manufacturers can do it but we also, the OEMs can do it but the manufacturers still have to make progress even when the OEM is not investing. So we have to cater to all of those requirements.

- [Ken] That makes sense and I think that probably speaks to why some OEMs are trying to include that in their service as a service model where they're because I think that's good for all of the players within the ecosystem so, make sense that they're including those-

- [Nikunj] Yeah, I'm gonna tell you about one. One of our partners is IMA Life. They are a pharma equipment maker, and they make very sophisticated freeze drying solutions that are used for packaging, pharmaceutical products, basically capsules and tablets. And it is required for FDA compliance that those methods be applied for packaging. Now, it's very sophisticated computational fluid dynamics that is required to maintain a chamber of a very, very pure gas at a certain pressure level so that you get even less than that the product effectiveness is maintained. Now they have all the science and they are experts at this. They've been in this business for decades. But even for them to put together the full conditional logic across all of the physical quantities that they can measure would have been an impossible job. And even if they mastered it. Every two pieces of equipment as they are used, start evolving in different directions.

- [Ken] Of course.

- [Nikunj] Even within one plant you cannot get the same exact behavior to replicate. And what they found was that at some point they could not express the problem that they were looking to avoid, rules were just not doing it. That logic does not make sense. That's when they turned to patterns and they're like, "If you can tell me what pattern causes this problem" that's much better than us telling the computer what is the definition of rule.

- [Ken] Right, right. And I'm just sort of incorporating this, I love the idea of, because it's just so much more mathematically sound to look at the pattern versus to say, this is the rule that always works because, yeah you can make that prediction, but as any scientist or mathematician will tell you most predictions end up being wrong before they end up being rules .

- [Nikunj] That's right.

- [Ken] And I wanna move to the other part of this before we run out of time because, part of the advantage here to the approach that you're taking, as you see it is that it speeds adoption across the industry which obviously IoT For All, we're a big fan of. I think that there are segments of IoT that are slower to adopt than others. Healthcare is certainly one. I think the sort of consumer space is another. We just finished CES upon recording this although probably not when you're listening to it. But the consumer space has been way behind the industrial and enterprise space for IoT. And I'm very interested in methods to take these sort of slower developing spaces and help incorporate IoT and make the case for them. So if you could tell me a little bit about how this makes implementation easier, faster, lower bar for entry. I'd love to hear about that.

- [Nikunj] Yeah, that's a terrific question Ken and I'm glad you asked for it. That's been some of the story of Falkonry. We've constantly been trying to make AI approachable as well as beneficial to more and more players. So only a year ago, you would have to be a company that makes at least $5 billion in goods every year to be able to adopt our AI the cost of buying a Falkonry subscription and the type of people who you would train to use Falkonry, typically a production engineers could make it difficult for people less than that to be able to do it. But now we're working with companies that are sometimes as small as a couple of hundred million dollars in annual production. And the reason that that happens is because AI has become easier for people to apply to their operations. They don't have to learn about how to apply the AI What they have to learn about is how am I going to take the finding that AI produced and figured out why it happened and what am I gonna do about it? Those are your responsibilities regardless of the AI. Because when a problem or incident occurs you're gonna do forensic analysis is gonna try and find out why that. So, one of the biggest ways in which we know whether the AI is adoptable is, what is the cost? And the second is, who is gonna be driving the adoption? And only when those two factors are met are we gonna see this getting adopted in more places. For example, I would love to be able to understand my soil condition, I have a lot of fruit trees in my backyard. And I'd like to understand the drainage, the additions that are required, et cetera. I like to be able to do that just from putting soil sensors and monitoring it over the year. And knowing that the time like right now, what changes do I need to make so I get great fruit again next year? And if really be able to do that, or somebody do that for me, or is there an AI that just adapts to my needs and tells me when I need to think about something. And that to me is the end goal of IoT. It will require putting more things into devices driving down the cost, even one notch lower. For example, right now for Falkonry starts at $50K per year, most companies can afford it. At $5K per year, most prosumers will be able to afford it. At $500 a month, most households would be able to adopt it.

- [Ken] Interesting. You know, I hadn't even thought about the sort of household level 'cause obviously I'm thinking at enterprise level this is a no brainer because just the scale and everything, but you raise a good point. The idea of even if not households, certainly facilities' manager, buildings and that kind of thing can certainly find tons of reasons to look for, you know, more efficiency, more predictability in maintenance and all these other things it makes makes tons of sense. So as we get near the end of our time I wanna sort of give you the floor as, I suppose, in sound the floor is a different thing but anything we haven't talked about yet that you think that listeners should taking away from the idea of AI and IoT coming together in this sort of, much shorter than typically thought about process in terms of time and investment.

- [Nikunj] Yeah. I would say a couple of things. First of all, the viability of AIoT is very strong right now. What does that mean? The economic and social goals that we are trying to achieve as a humanity drive the need for AI and IoT. The value proposition of AIoT in the form of SAS, is also tenable. Meaning you are gonna get more benefit than you're gonna spend on it. And the third is that, we are moving from technology to insight, and that is a transition point where the mainstream should feel comfortable adopting. Now, in addition to this, I will say one thing because of the approaches people have taken in the bad and the bad rep that IoT has received for being too many POC and pilots and so on, the other thing is that, there is no such thing as POC. If you remember Yoda, right? There's no such thing as trying, there's only doing you've got to do it.

- [Ken] Yeah, yeah.

- [Nikunj] And only when you do it do you understand what are the actual on the ground difficulties for the manufacturing sort of personnel, like they are the ones at the end of the day, we're gonna create value. It's not people sitting in a lab with white suits. They are the ones who will perhaps be able to coach maybe people like us but the people who are going to play in the field they are the ones which will tell us whether this is right or wrong. And for that, you have to this, you cannot just do abuse.

- [Ken] That's a really, really good point. I sort of have this editorial mission that like I am so tired of the pilot program and the show me a solution, show me something in the field, show me a case study that's at scale. Those are the things I'm here for because I think as an industry we have slogged ourselves with, this is what we can do. Let's figure it out instead of saying, "This is what we should do and just doing it." So, I think we're turning the corner. I think companies have started seeing that and saying, "Yeah, yeah that's an interesting idea." "We'll talk about that later, right now we're doing this." And I find that to be very positive. And so we're ending on a high note, which I appreciate. Thank you so much Nikunj for being my guest and for telling us a little bit about what you've been working on at Falkonry. So thank you for joining me. Thank you folks out there for listening. Don't forget to like and subscribe, et cetera and come back for more on Let's Connect. I have been Ken Briodagh, editorial director for IoT For All, and your host. Thanks for listening and have a great day.

- [Nikunj] And thank you very much. We'll meet you on the predictive operational next and shortly

- [Ken] That sounds great. Thank you Nikunj.

- [Announcer] Thanks again to all of you listening out there. I hope you've enjoyed our discussion and if you have please make sure you like and subscribe so you don't miss out on any of our episodes. We posted every week and I hope you'll leave us a rating review and comment on your favorite podcast platform. If you'd like, yes, please there's a link in the description And we also have a great sister podcast network called the IoT For All podcasts. So make sure you check that out.

- [Ryan] Hey Ken, let me jump in real quick and introduce your audience to another awesome show on the IoT For All Media Network. The show that started all the IoT For All podcasts where I bring on experts from around the world to showcase successful digital transformation across industries. We talk about Applications in IoT solutions available on the market and provide an opportunity for those companies to share a device to help the world better understand and adopt IoT. So if you're out there listening and haven't checked it out be sure to go check out the IoT For All podcast available everywhere.

- [Ken] Thank you, Ryan. Now get back to your show. And thank you all for joining us on this episode of Let's Connect. I've been Ken Briodagh, editorial director of IoT For All, and your host. Our music is "Sneaking on September" by Otis McDonald and this has been a productive For All Media Network.

- [Announcer] You are listening to the IoT For All media network.