PODCAST: Razor Labs on IoT For All – AI and IoT in the Mining Industry

[00:00:00.07] – Ryan

Welcome Michael, for IoT For All podcast. Thanks for being here this week.

[00:00:07.01] – Michael

Thanks for having me. Thank you, Ryan.

[00:00:08.23] – Ryan

Absolutely. It’s a pleasure to have you. Maybe kick this off by having you give a quick introduction about yourself and the company to our audience.

[00:00:13.23] – Michael

Sure. So I’m Michael. I’m the Chief Technology officer of Razor Labs. I’m a serial entrepreneur. So I founded. It’s already my fifth company that I founded with my friends. And the company focuses on predicting maintenance for heavy industry in the mining sector. We built a very unique, I would say IoT as a very diverse standard for monitoring the health of machines. And we’ve built some very cool artificial intelligence, takes all this input and just like an automated doctor diagnoses the machines, what’s wrong with them, what are their illnesses, so to speak, and basically prevent failures which really transfer to very dramatic financial results for our customers.

[00:00:57.07] – Ryan

That’s fantastic. Yeah, I want to dive into, like the use cases and applications of what’s. What you’re doing in certain industries in a few moments. But maybe we can start with a high level question, talk to our audience that may be unfamiliar. What exactly does it mean when we say predictive maintenance? And why is it so critical in different industries, particularly like the mining industry and these other, you know, large manufacturing type environments?

[00:01:19.24] – Michael

Traditionally, what was done is that you would run a machine up until the machine basically reach its end of life, it exploded or just stopped working and so on. And the problem with this approach is just like with your car, if you do it, it means that you need to replace the entire machine or large components of it, and it boils down to very, very high costs. And even when it happens, it takes usually a few days to make these replacements. And the downtime itself, measured by what is the cost per every hour, is very high. For example, we work with gold sites or copper sites or power plants and so on. And you’re speaking of at least a few hundreds of thousands of dollars for each hour of downtime. So it kind of gives you the scale, and that’s for one hour. Imagine that when you have downtime, it’s probably many hours, even a few days. The goal of predictive maintenance is to know what is the right optimal time to maintain my equipment, to prevent these failures from happening, to prevent these breakdowns, and to have the maintenance be much, much shorter because you’re just doing something small instead of replacing everything.

[00:02:27.02] – Ryan

That makes total sense. And how are you kind of as a company applying AI, specifically AI sensor fusion, kind of to transform the way things are done? In the field from an operations perspective in mining to help kind of achieve what we’re just talking about.

[00:02:40.18] – Michael

It’s a very good question and I think that the best analogy is doctors. Like how do doctors diagnose humans? And first of all, it’s being able to look at a very diverse set of sensors. Imagine you go to a doctor, the doctor has ultrasound, MRI, CT scans, blood work. And the reason is you cannot diagnose all illnesses with just one data source. And all of them together kind of COVID everything that can go wrong with the equipment. In our kids we’re using cameras, oil sensors, vibration sensors, depressions, flows, modular current. So many different sensors with many different kind of, they look completely different. Drive cameras give us pixels, oil gives us parameters of the oil. Temperatures are just kind of time series. And then it goes into AI that looks at everything and basically diagnoses what’s wrong. So what are the symptoms but what is the root cause and what needs to be done to prevent the failure from happening?

[00:03:38.19] – Ryan

That’s a great analogy to it. I think one of the things that people don’t understand necessarily or don’t, I guess consider as often is, is how much data is required for the AI component of a solution to actually do do its job. So the fact that you are collecting lots of different data points from different types of sensors, as you just mentioned, what you’re able to do with that data becomes more and more powerful. The more data you collect, the more the models are trained, the more types of use cases or things that the customer wants you to be able to identify or alert them to. And that’s really the power behind what AI can do when you do it at scale.

[00:04:11.27] – Michael

I can tell you that one of of our largest challenges were that in traditional AI, you actually needed to ask the customer to help fine tune and pitch the system and train the system. And in the heavy industry sector, you know, whether it’s mine sites or processing plan power plants, nobody really has the time to do it. And we actually needed to go to methods that are much more similar to how ChatGPT is trained to give the customer a model or a system that already works. So we put the sensors, whatever is missing, and we turn up the system and it just works. We don’t require from the customer to do any training or teaching of the AI, which is very, very innovative.

[00:04:51.16] – Ryan

Yeah, I think that’s something that’s a really good point to mention is the fact that when you’re going into these industries, especially in like these heavy industries, you’re talking about the technology that they’re looking to adopt is not their primary focus. Right? They’re focused on doing their job, growing the business and performing well. They wants to be able to adopt something and know that you or you and your partners have it all under control, right? They don’t really care about the technology as much as they care about the solution and the output. I think that’s something that’s over time throughout the IoT industry has changed. Initially people were so focused on the technology, like, oh, explain to me how Bluetooth works, explain to me how cellular IoT works. Now people are like, look, that’s all table stakes. I just want it to work. I don’t care what the sensor is, I don’t care. I just can you produce the ROI and the output that I need? And for you being able to just kind of go in with a business case, solution is probably the best approach as opposed to trying to convince them around the tech.

[00:05:42.23] – Michael

Right? They don’t care about how the AI works, they don’t care under the hood. They care about what’s the value for me, how do I measure it in a business case, what are the KPIs and when can I get it?

[00:05:53.08] – Ryan

So can you take us through some maybe real world examples of, of mining sites where this technology has made a measurable impact and just kind of take us. What does that journey been like when, when engaging with, with these organizations that are doing mining?

[00:06:03.26] – Michael

I can take an example, right? We come to a mine site and first of all, exactly as I said, we build a business case, right? So what are the critical machines that when they fail they cause the maximum financial impact. It can be side, that is a site constraint, meaning that site is the constraint and they’re looking to increase their productivity. Or there, there, there can be another side that’s not a constraint and then there will be looking to reduce their maintenance costs. So each side basically has the wrong KPIs. We choose the machines and then our differentiation is the ability to find failures that no other system can because that continuous sensor fusion that’s happening. And I can tell you that literally just a few months ago we had a major breakdown that we prevented where we found a failure that their traditional systems did not find. Other traditional systems were just made on vibration. The problem, the mining sector is that the machines are very noisy. I mean the machines, I don’t know. The way that processing of gold or copper or minerals work is that you crush the Oregon and you grind it and that’s obviously a very noisy process.

[00:07:14.20] – Michael

So you have high vibrations all the time. And these other systems have tons and tons of false alarms. But because our system is able to look at many sensors together, it’s able to always compare the machine to its same operation mode and to reduce this noise. And we’re actually able to find the failure in a machine called mill. It’s a major machine. Every downtime from mill, you know, every hour is more than 3, $400,000 per hour. So really crazy numbers. And when the team site actually send their own team to kind of validate what we’ve seen, they didn’t see anything because of the noise, They’ve just seen noise and they took apart. They had an upcoming issue with their motor, they took it apart before it failed. They removed the bearing in that case and they just saw that severe damage that would really in a few weeks time would have caused a major downtime. And imagine that for any machine, whether it’s pumps or fans or mills or crushers, conveyors or whatever the site has.

[00:08:15.08] – Ryan

What’S the reaction when you are working with a customer and you’re able to actually identify an issue that they were unable to identify with their existing system? I imagine they must be just super ecstatic about this kind of thing for sure.

[00:08:27.01] – Michael

But I would say that building trust is very important. One of the key features that we implemented, we call it evidence. So the AI actually shows you why it made this diagnosis just like a doctor. So it’s not a black box that you just need to believe, but you can really see what it looked at and really helps in building trust. And the human component is also very important. I mean we have experts that speak with the customer regarding answering any questions. What we found, why did we recommend this one and not that one? And the change management is a very critical part of the process. Because you mentioned these are pivot, not some send and forget AI.

[00:09:03.05] – Ryan

Tell me a little bit about you mentioned cameras. I don’t want to dive into that for a second. So obviously anytime you bring a visual component into it or a visual sensor into any kind of environment, you really unlock a lot of different things, right? That basically something that somebody who had to use, you had to use humans to look at to find defects in, to find issues, whatever it is to monitor. Now you can put a camera in place. Where have you seen the biggest benefit of bringing cameras into the these heavy industries, into the mining industry that have played kind of the biggest role.

[00:09:33.25] – Michael

So it’s a fantastic question Ryan. And the answer for me is looking at the material. When you Think about it. All these sensors allow you to look at the machine. It allows you to see why it suffers. But many times the machine suffers, but it is not the one to blame. And actually the material itself, whether it’s too large, whether it’s too, I mean too much volume and so on, there is no really way to look at it. And the cameras allow us to look at material which is, or in most cases in mine size and allow us to, to measure it, to quantify it. And it is actually many times the reason for the failures. Right. If your material has anomalies inside, it will actually screw up the machines and cause failures. And replacing the equipment will not do anything because it is not the root cause. So we’re able to really find many more failures that way and to find the root cause which doesn’t only prevent the failure from happening, but prevents all the further failures that would have happened if you didn’t take care of this one. We also do process optimization.

[00:10:33.27] – Michael

So you can see visual parameters. For example, in gold there are kind of bubbles in what’s called flotation cells. It’s a key recovery component of extracting gold from the entire ore. And in one ton, let’s say you have between 2 and 4 grams of gold and it can be the reason of increase of dozens of percent in the efficiency of the process. So whether you extract three or four grams from the same tongue, it’s, it’s really the arm of a difference.

[00:11:03.19] – Ryan

Imagine you already mentioned one talking about vibrations here a second ago. Building and deploying a solution in these harsher environments, right outside the very different environment, than just doing something in a contained inside environment. What are some of the biggest challenges that you all have come across from kind of the early starting into this space and, and to, and then you figured out, been able to solve to now that have made the biggest impact that maybe other companies haven’t been able to solve, given all those different variables that come in, come in. When it comes to deploying in these different harsher environments, there are I would.

[00:11:33.18] – Michael

Say a billion vibration sensors out there. And these vibration sensors are really meant for very simple equipment. If you have a small pump, you have a small fan, these would work. But when you look at that crazy mining equipment, which is large, a motor can be the size of this room, you’re speaking about dramatically wider band of frequencies that are required that most of these sensors don’t support. And, and on top of that is the ability to look at the operational variations themselves. Because take, let’s say you take Your car on the mountain, the engine will be vibrating more strongly because just working harder, it doesn’t mean that there’s something wrong with the car. It’s the same in mining. They have tons of false alarms because vibration is higher. But the machine is just working harder, harder. And what we’ve done is we’ve built some very specific algorithms for heavy industrial applications like oil refineries, mine sites, smelters, and so on that are able to take these ones into account and actually prevent these false alarms from happening. And I can tell you that a year ago, we’ve signed a major deal with, with Glencore across 14 sites globally.

[00:12:42.27] – Michael

And we’ve seen today really everything from underground mines to smelters, to processing plants to oil refineries, and really made sure that what we search for in the signal is actually completely generic for all of this equipment. So we’re supported right from there, right from the start.

[00:12:59.29] – Ryan

When you’re deploying in these environments and now you’re working with these organizations at scale, what are the KPIs or success metrics that these companies are looking to attract when adopting your solution?

[00:13:09.14] – Michael

Number one is how much downtime did you save in terms of hours? They can convert it to dollars. How much was my maintenance cost by how much were they reduced? And then they’re looking into spare parts, labor, and so on. When we look at mobile fleets, we also do trucks and excavators and so on. It’s even the ability to how can I shrink my fleet size and still achieve the same productivity? Because every hour of my trucks, for example, obviously has a very high cost. Tires alone are high, are many times up to half of the entire maintenance cost of the entire fleet. So if I can either reduce my fleet and still achieve the same goal, or I can achieve more productivity with my fleet, that’s really major.

[00:13:55.10] – Ryan

One of the things I wanted to ask you is about just kind of the overall, I guess, sentiment in the mining industry to adopt digital technologies and solutions. What were the biggest challenges in getting these organizations to be open to bringing this kind of technology into their environment? And then where do you kind of see things going into the future over the next few years? And like, with AI growing, how’s that going to continue to play a role?

[00:14:18.22] – Michael

I see today in mine sites a lot of manual labor. It’s not uncommon today, even in the US and Canada, to find sites where you have personnel that manually go to machine by machine, take, you know, manual measurements of vibration or oil and so on. This is obviously also very unsafe because Imagine you’re near a machine, the machine explodes and you’re there. So I think that going to higher and higher automation where things basically get analyzed continuously and with full coverage and then the team on site basically gets directed to guys this and that one need to be repaired or these are the actions and so on. And you’re basically proactively doing that before you get to these catastrophes. That, that’s the future that I see.

[00:15:07.27] – Ryan

One of the areas I’ve talked to some companies about is, is on. And you, you mentioned ChatGPT earlier on the gen AI side. Are you seeing demand for being able to interact with this data in more of a natural way for the, for the, for the people within these organizations that are on site, maybe management being able to just have better access to the data in a more immediate, more natural language type environment like in their, you know, Slack in their team. Just, just kind of in that experience that we’re all now growing used to using ChatGPT, Gemini and so forth, how, how is that kind of engage or how those kind of conversations been?

[00:15:42.04] – Michael

My intuition is that as long as the, the user experience is great, they don’t care about whether it’s a dashboard or, or a text. What they do care about is they understood up until ChatGPT or once ChatGPT came out, they, they understood that they were basically drowning in data, but they weren’t really extracting any major insights or it was very hard, let’s say, to extract major insights because they had sensors on sites. Right. It’s not something new, but really the ability to convert it to, to an action like what ChatGPT does opened many eyes. And I think I can tell for ourselves that in the last two years since ChatGPT came out, the, the ability to embrace AI and to implement AI on sites became much, much simpler. People were much more open to innovation.

[00:16:31.06] – Ryan

So when you look at where you all sit kind of in the ecosystem, how would you, I guess, compare yourself to others in the market? Like what is your biggest differentiator from other conditioning monitoring tools and just kind of how you, how you see yourself playing a role?

[00:16:45.29] – Michael

I would say we have several key differentiation. One differentiation is that we’re literally built for heavy industrial applications and it means everything. It means that the diagnosis that we’re performing is built for these machines and has very deep familiarity with these machines. It means that the hardware that we deploy is built for these machines while other sensors are just unable in the first place to even get the right frequencies. They would look at much lower Frequencies that are just irrelevant for these machines. And obviously it all goes to the diagnostic engine. That invention is able to give them the bottom line. With this perfusion, it’s very easy to buy a sensor, but the ability to actually convert it into what’s the diagnostic, what do I physically need to do tomorrow? That the hard part?

[00:17:33.12] – Ryan

We’ve kind of talked about a lot of good topics that I wanted to cover. The last thing I wanted to ask you before I let you go here is if you were to kind of just talk, talking to other companies, not, not, not competitors per se, but just other companies looking to deploy solutions in harsher environments more, you know, outdoors, maybe even in other heavy, heavy industries, what advice would you have for them to really think through before they get to a point of entering a market? Like, what are the biggest learnings that you’ve kind of been able to make, you know, over the course of, of, of growing your presence in this space?

[00:18:05.21] – Michael

I would say my largest advice would be to go and meet as many. In our case, it’s mine sites or heavy industrial sites, as many as you can. Because speaking with these people, hearing their challenges firsthand, there’s nothing that can replace this one. And they will tell you what are their challenges, and from that you can really go backwards and ask, okay, what hardware do they require? What AI do I need? And so on. I mean, we’ve visited dozens and dozens and dozens of sites. I think it really allowed us to make sure that our product has the best fit for them.

[00:18:41.29] – Ryan

Michael, thank you so much for taking the time. For our audience who wants to learn more about what you all are doing, reach out, follow up, any questions like that. What’s the best way they can do that?

[00:18:50.08] – Michael

Sure, just go to our website. It’s razor r a z o r-labs labs.com. you can contact us down there.

[00:18:58.29] – Ryan

Really appreciate it. It was great to have you. I’d love to maybe find some time later this year, kind of as we head into next year, and just learn about kind of how things have been growing in the space, how AI has been growing in the space, what you’re seeing kind of evolve as a leader in this industry. It was great to have you today and hopefully we’ll have an opportunity to talk again soon.

[00:19:15.25] – Michael

Thank you, Ryan.