PODCAST: Razor Labs on The Northern Miner – Predictive Maintenance with AI Sensor Fusion

[00:00:00.12] – Adrian Pocobelli

Joining us today, I’m very pleased to welcome to the Northern Miner podcast for the very first time for this week’s Spotlight, Michael Zolotov, Co-Founder and CTO of Razor Labs. Michael, welcome to the show.

[00:00:11.12] – Michael Zolotov

Thank you, Adrian. It’s a pleasure to be here.

[00:00:13.18] – Michael Zolotov

It’s great to have you, Michael. One of the most interesting parts of the mining sector as a news narrative is technology. And it seems like technology could really change things in the coming few years. It’s already begun. So Razor Labs is a leading mining technology provider. For those that are unfamiliar with Razor Labs, could you give us a little bit of background about the company?

[00:00:38.03] – Michael Zolotov

Sure. Our focus is delivering predictive maintenance and condition monitoring via what we call AI sensor fusion. Both for fixed and mobile equipment. We work with some of the largest players like Glencore, Siemens, and others. And basically, we are like automated doctors for mining equipment.

[00:00:56.23] – Michael Zolotov

Very interesting. So automated doctors for mining equipment, do you repair equipment? Just to clarify, what exactly does that mean?

[00:01:04.05] – Michael Zolotov

So no, we don’t repair it. Obviously, the teams on site do it. But imagine just like if something is wrong with you and you go to a doctor and the doctor has MRI and CT and blood work and everything. We basically do the same for machines. We monitor them with a very diverse set of sensors, really from cameras to vibration to pressures and oil and everything. We use artificial intelligence to combine everything to give a very deep diagnosis of what’s wrong with the equipment. Again, whether it’s a haul truck or an excavator or a mill or a crusher, really weeks and sometimes even months before it fails, and therefore preventing equipment failures. That’s what we do.

[00:01:44.17] – Michael Zolotov

Okay, excellent. It sounds like a systematic approach. Is this then the AI sensor fusion technology? In a sense, why should mining companies use this technology?

[00:01:55.27] – Michael Zolotov

The value that we deliver is eventually, if you’re mining constrained, For example, in your processing plant, you care about cutting your maintenance costs, you care about reducing spare parts and labor and so on. And when you actually prevent a failure from happening, it would normally cost you between three to five times less than when the breakdown completely happens. And whether if you’re planned constraint, we’re looking at increasing the throughput and the productivity of the sites. So it’s literally pushing more tons by both reducing the unplanned downtime and also the planned one. Because you do not need to maintain the equipment if it’s healthy.

[00:02:33.29] – Michael Zolotov

So what is a concrete example, say, of how this might work on a mine site?

[00:02:40.14] – Michael Zolotov

Sure. So for example, right now the gold prices are skyrocketing and all gold sites that we know are really trying to push as much tons and recovery as possible. And the mills on these sites are always a bottleneck. They’re always working or they should be working 24/7, and there is no redundancy. And just A few months ago, we had a major failure that we prevented on one of the sites that traditional methods just missed. They had their own vibration team, but the vibration team was obviously looking only at vibration, and nothing could be seen there. Actually, our system that looks both at vibration, but also at the motor current and the tachometers, and even at the feed to the mill by looking at the cameras, were able to pick up a critical deterioration of one of the bearings of the motor that just went unnoticed. And they actually had a very long shot. They forgot to rotate the mill during that shot. As you know, you need to rotate it slowly to prevent exactly this failures. They just forgot to do it. Our system, when they went back online, indicated this very serious issue. And instead of having it breaking down in a matter of three or four weeks, they just reshuffled the test in the next shot that they had the week after.

[00:03:57.14] – Michael Zolotov

And in a completely planned manner, took a few hours, replaced the motor. They took apart the old motor, and they just saw this severe damage that would otherwise would have caused at least a few days of shutdown. So that incident alone was more than half a million dollars, and that’s one failure in one site. I mean, multiply it across multiple failures across multiple sites, and really the potential is enormous.

[00:04:21.17] – Michael Zolotov

Interesting. And ultimately, this is an AI technology. Ultimately, that is what solved this issue. Is that correct?

[00:04:28.22] – Michael Zolotov

Correct. The diagnosis that was performed by the AI by using the sensors that we deployed, all of this system together was able to deliver this value. It’s worth noting that we are using existing sensors wherever they exist on site. But if there is any gaps, any sensors that are missing, our teams just come to site and deploy the missing sensors.

[00:04:49.28] – Michael Zolotov

Okay, excellent. So a lot of people in the industry are using AI. It seems like there’s a lot of different applications that AI can be used for. So how is Razor Labs different in how it’s incorporating AI into a mindset?

[00:05:04.26] – Michael Zolotov

Sure. So it’s a very good question. And as you know, traditional AI basically learns by example. It means that you need to actually deliver it hundreds of thousands of examples of what you wanted to learn, in our case, failures. If I take this even mill example, I don’t think that in the history of humankind, there’s been hundreds of thousands of mill failures in all mine sites combined. For that reason, when you actually look at old-fashioned traditional AIs, the vendors actually ask the sites to train and to calibrate the system that they deploy for half a year or a year or even 18 months. It puts this crazy burden on these teams on site that are already overloaded. I guess that one of our key differentiations is that our AI comes pretrained with all the required diagnosis knowledge. It means that the teams on site don’t need to train it, they don’t need to teach it. We actually use the same training methods that ChatGPT uses. When you use ChatGPT, you don’t need to train it or teach it, you just use it. Obviously, it’s a huge advantage because you remove this entire requirement for long on-site training phases.

[00:06:12.08] – Michael Zolotov

It’s a very different training mechanism that we use for our AI to be able to deliver value, literally the way that we walk out on the site.

[00:06:20.15] – Michael Zolotov

Okay, excellent. You also use visual AI technology. What role does that play within your methodology?

[00:06:27.24] – Michael Zolotov

Yeah, that’s correct. When you think about it, cameras are maybe the only way to actually monitor the material itself. And many times the material is actually the one that is causing the failures. We’ve seen many clients that have either uncalibrated crusher gaps or worn liners that are causing oversize material that is then downstream, causing stoppages to the equipment. But the equipment is actually not to blame. And actually the material is the one that’s causing it. And when we put cameras really on strategic places like the feed to the mill, like the discharge of the crusher, it allows us to monitor the material itself, and we actually cross-reference it with the rest of the sensors, with the oil sensors and the vibration and so on. And it allows us to really make sure that we completely cover all the failures. We also use it to monitor the converses themselves, like misalignment and rips and so on. In gold and copper sites, where it’s highly beneficial, we’re also monitoring flotation cells. We’re looking at the bubbles, their velocity and density with AI. The AI actually analyzes these parameters, the sizes of the bubbles, their density and so on, to make sure that the gold recovery or copper or coal recovery are really optimal.

[00:07:41.26] – Michael Zolotov

That’s something that can only be done with a camera and cannot be done with any other way.

[00:07:45.29] – Michael Zolotov

Very interesting. And so what machines are involved in all of this? And how do you make this work, say, in harsh environments, say, if you’re up north or something?

[00:07:56.23] – Michael Zolotov

Sure. So we actually cover all fixed asset offsets on site from the crushing circuit to the grinding circuit and conveyors and critical pumping systems and compressors and everything. And we also cover haul tracks also now expanding to excavators. One of the key features of the sensor fusion is that we’re able to monitor the equipment under the same operational conditions and under the same load. And this is critical because obviously, if the machine is working harder, it will vibrate more strongly, all the data will be completely different. And regular tradition traditional systems that disregard it just give you these thousands of false alarms that you see in SCADA or other systems. Our system is able to monitor, for example, a haul truck always at the same gear, same load, same speed, same everything which completely prevents all these false alarms. It’s both the fixed assets on-site, but also the mobile fleet itself. When we look at how we implement, regarding your second question, we have really a very clear methodology. We come to site for a site survey where We really work with the client to identify what are the single lines of failure that if one machine stops, then the entire line stops that have the highest impact.

[00:09:08.17] – Michael Zolotov

We discuss the KPIs with the client and we help them build the business case, whether it’s reduction of maintenance costs, of spare parts and labor, as I said, or whether it’s increase of throughput depending on the site and its characteristics. We really map what sensors exist and where there are gaps. Then we send our local teams, and we have them really both in America, in Australia, in Africa, really every major territory, we have local teams that deploy exactly the sensors that are required. And we have a methodology of which sensor is required and where. And that’s basically it. I mean, once the sensors are deployed, the system is up and running. As I said, there is no teaching. The client does not need to teach or calibrate the system, and it basically runs from that point onward.

[00:09:51.19] – Michael Zolotov

And as we start to wrap up here, Michael, a lot of mind sites may already have monitoring tools of a certain kind. In a sense, why should How would they be interested in what Razor Labs is offering? In a sense, how are you different from what other people might have?

[00:10:06.27] – Michael Zolotov

Sure. So I think that there are many dashboards that provide you with the raw data, raw data that you need to analyze by yourself and that you need to continuously look at by yourself, which is obviously not practical. Our approach is different. We do not just throw terabytes of raw data at our clients, but we really provide, as I said in the beginning, a 24/7 automated doctor or a A team of doctors that are looking at your machines, and this is what the AI does. They’re giving you a very deep diagnosis, whether it’s gear measure that I found here or bearing deterioration there or performance issues. And it doesn’t tell you that something is wrong and provide It’s not providing you with these thousands of false alarms, but it’s really telling you what’s failing, what is the failure root cause, and what needs to be done. You can say it’s an Apple approach with giving you the bottom line in the recommendation. Obviously, you can see all the data, but the The agnosis part is really the key differentiator. In order to be able to deliver it, we also use sensors that others don’t, like cameras that we spoke about and like oil sensors that really allow us to cover failure modes that other systems cannot.

[00:11:13.13] – Michael Zolotov

By putting all of these sensors together, we can really make sure that there are no failures in the machines that we cover.

[00:11:19.23] – Michael Zolotov

Okay, excellent. Finally, who should consider this technology and where should they go in order to get more information on it?

[00:11:28.29] – Michael Zolotov

I think that the site leaders and maintenance managers and innovation managers that can see in their KPIs that unplanned downtime of equipment is causing them large losses. Again, whether it’s in maintenance costs or just lost productivity or There are sites that are over maintaining. We’ve seen sites that are stopping the site every week for maintenance. So they’re losing lots of throughput because of that planned maintenance. Any leader of a site that can see how they can benefit, are really welcome to contact us. Just go to our website, Razor-labs. Com. Send us a message through the site.

[00:12:07.13] – Michael Zolotov

Michael Zalatov, co founder and CTO of Razor Labs. Thank you for joining us and sharing everything you’re working on at the company.

[00:12:15.23] – Michael Zolotov

Thank you very much, Adrian. It was a pleasure to be here.