Hello everyone. Thanks for joining the session today. I would like to talk about predictive maintenance using iot and machine learning. Basically, it involves the sensors, the realtime data collection the analytics using machine learning, and it covers a bit of future respects as well. Okay, I have split the agenda into these categories. First, the types of maintenance and what is predictive maintenance? The sensors, the market, the challenges and the future direction, and also the machine learning part. Yeah. When you look at a, an industrial equipment that's installed in a factory or a manufacturing unit or the one that is used, transformers that are used for power supply. There are multiple types of maintenance that are available. Typically the one that is done is a breakdown maintenance, like where after we see a failure, we react to it and start the maintenance process. Like we call the company that maintains it and the spare parts that are needed and so on. So that is breakdown maintenance. The other one is preventive maintenance. It is like periodic maintenance that we do, like an oil change for the car. There's nothing that's broken down, but still you, you just want to ensure that it runs properly and you do a preventive maintenance. This is mostly done on a schedule, like quarterly or annually depending on the equipment. And what we find is one a PC study, they said they found that more than 80% of failures are random. So you can't cover them in preventive maintenance. And of course, if it breaks down, you have to do breakdown maintenance. Here comes the the power of predictive maintenance. Now we have IOT sensors, which simple like a microphone or a proximity switch or a. A simple vibration sensor, which can basically monitor or gather the the broad data like a sound or the vibrations, and which can be intelligently analyzed using ML algorithms. And you can predict even before the equipment goes into breakdown that it's going fail and without any downtime, you can do a prevent preventative maintenance. That's, that is called us predictive maintenance sensors have matured greatly. They use less and less of power. They have a more accuracy has increased and you can measure various parameters like sound. The vibration, as I mentioned, temperature, pressure, humidity, and this is real time, right? Like when the mission is running or the equipment is running, you collect the data and we are pushing it into cloud or some data lake where you can gather all the information. And now with the advancements of ai, we have sophisticated machine learning algorithms, which can process these detect anomalies and report as perform predictive maintenance. The key benefits of IOT based maintenance are, of course, like we, we predicted early before it happens. And even though there is a downtime, you can schedule it like in non-business hours or something that doesn't have much of an impact. And to be relatively less. And to add to that, in case the. There is a predictive maintenance part of it. The technician who comes in comes with the right part that is needed. So there is no back and forth to go for, go searching of the parts, finding out what's gone wrong. He comes and does a surgical maintenance, so that reduces the downtime. Of course, it saves a lot of cost about 25 to 30% and then since it's maintained even before a breakdown happens. The equipments lifespan increases greatly. And finally, what our feedback is gathered by the predictive maintenance can be fed back into the design table so that the next products that are iterated by the company can incorporate all the findings that they have in real time into their next design, which basically evolves their product. PTC is a company. We just got a, which is very good in microphone monitor, acoustic monitoring. Basically they can be deployed anywhere and it can be retrofit or it can be deployed on an existing missionary without disrupting, making a lot of changes around, around that. You can just go and place it on a relatively benign spot and you can start gathering the information. It's non-invasive and, it basically continuously sense data, which are compared by the ML algorithms. I will see about it later. Further broadly it's it's first data collection. Getting the operational equipment footprint, basically the creating the acoustic footprint of the equipment. It's like a thermal map or a, something like that to imagine. And the data is collected continuously, so it's real time. And, it's multimodal, right? Like it has got the sound profile, it's got the vibration profile, it's got the temperature profile, so it's multimodal which can produce a very complex pattern. Like you can combine all these to find out precisely what's going on. And what needs to be maintained. And the data can be compared like, say for example, a company which is manufacturing the equipment. They can compare it with previous failures or previous maintenance that they have done, and which can reduce the false positives and enhancing the credibility of the the preventative maintenance. Yeah. Now we have the sensors. How do we integrate the machine learning? When you see the diagram, it basically, there is a data acquisition part and there is a signal processing part. And third is the the signal is set to a machine learning model and. We also have something called as h computing integration to reduce the network latency and feedback for the model refinement. When you see I. Again, it cal collects all the data that is the first point. And signal processing is, as an example, it is converting an amplitude of a frequency into a time series data, basically converting the frequency, converting it into frequency domain so that the ML model can better analyze that. So it can look at the peaks where it is speaking. But for just with, without the amplitude they can look for the frequency time series also. Yeah. And we can train the machine, learn models with the different algorithms like random forest, CNN, to predict anomalies, and maintenance needs. Yeah, edge computing is very interesting because these sensors, they collect a lot of data. We can collect every second, and there are some which collect in milliseconds, and you can of course configure them, but you can do a localized signal processing instead of sending the raw data directly through the network and kind of overloading it. So you can reduce you can just process the signal here and eliminate what is not needed and transformations that are needed. And you can reduce the, by doing this, you can reduce the. The number of data the amount of data that we transfer over the network and basically to reduce the latency. So basically you can find out what's going on pretty fast, relatively, and model refinement. They keep on increasing, it's possible to keep on increasing the prediction accuracy through adaptive learning and supervised learning as well. So this is how it works. Basically, we collect the sensor, we signal process it, and affect the machine learning model. And if needed, we use the most of it will have an edge computing integration to reduce the network load. And finally, there is a supervised component to improve the model security because you start with something that is a false pass, it keeps on maturing and the data can be made available on a dashboard. And there can also be alerts that can be triggered on a mobile phone, like for a service engineer as an example. Yeah, it's got multiple application areas. I think maybe you already sensed it. It can be applied in manufacturing. If there is a say a bottling plant or if there is a emissions that are running. So in manufacturing basically, and energy, like where there are the turbines running and you can measure the transformers frequency, the data image, to find out if it's going to fail. Using the predictive analytics, because we have the sophisticated algorithms to even detect a very minor anomaly which is real, not of all positive, and you can act on it. And logistics per fleet management? Large trucks, 18 wheelers, they carry. There are so you can monitor it pretty much because by avoiding failures in that and maintaining it with predictions, ML based predictions, you can avoid failures and can ensure that the SUP supply chain is not impacted. It has got other applications also like automobile, like there are, there is a, in cars, they're used as a vascular health systems. And oil and gas industry as we discussed, it could be used in railways basically where, because that's a public transport or it could be a good stream which carries goods like, so those things before they, if they break, they cause downtime and supply chain issues. We can, it can be predictive, it can be maintained in using prediction and utilities. Of course, electricity, we already saw gas. Anything that that comes to us through internet can be monitored. Yeah. Compared to 2020, the market has grown or projected to grow 400%. So it was about $7 billion in 2020 and it's going to go more than $28 billion in 2026. 'cause there are a lot of advantages to this approach. Easy to sensors are easier, non-invasive can easily implement it. And integration with the maturing of machine learning, the prediction accuracy is pretty high. So thereby you can, you can avoid downtime. So there is a lot of investment going on and even new equipments that are, that already come with prebuilt sensors, which could be used for predictive monitoring. Some of the studies aggregated show that there is a 50% downtime reduction because of predictive mainten. And of course there's cost savings. It's more than 30%, but. Typically it's 30% around 30% and uptime, right? The mission is available for manufacturing, which basically caters to the, or makes the company efficient trustworthy to deliver on time. It increases by 60%. So what it means is if there were no predictive maintenance, then they, there would be a 60% reduction in efficiency. So because of implementation of the predictive maintenance using iot, industrial iot, the company profitable as well. Yeah. Some of the challenges to implement this is typically these factories or the whole systems have some mindset of, traditional reactive maintenance. So to, to change the mindset and come to a concept of predictive maintenance can be challenging because my, my system is working fine. Why should I certainly do it before it happens? So that is the mindset, but we have to educate them to come up to the speed. And there is a lot of skill gap. In, in terms of people who monitor it in case like the companies themselves decide to monitor it centrally or so people who understand iot technologies, data analytics, and even acoustics, like in case there is a relocation of a mission. So there is a skill gap in this, and the most important is the cybersecurity concerns. Like the IOTs devices are highly prone to cybersecurity attacks because the, there are no frequent updates, like unlike a desktop or a. System that's maintained. All these iot sensor are just a fit and forget kind of things. And so they always tend to get old or not up to speed to the current. There are no frequent updates to that. And they make them vulnerable. The certificates expire or they become legacy. Yeah. And the data that comes out of the system is massive. Because these are all sensors it is it's going to just bombard the system with so much of data. So it requires maybe a data like solution and sometimes you may not get the data as well. So it needs that kind of an object storage system, like an AWS S3, which can handle large volumes of data, creating a retrofitting old industrial equipment sensors. Could, because it's a old system, like we may not have the profile it has got its own implementation challenges. Challenges. So there's a lot of homework that needs to be done to successfully integrate a legacy system, which kind of very old to to this predictive maintenance system. The future looks bright, so you can have a. Can have a technician come up with the AR and virtual reality augmented reality so he can, when he looks at the machine, he can see, he can be guided by the machine or by a person who's an expert, different geography, location, to precisely go the things that he needs to do to fix the system. And you can have virtual models that's got a digital twins that's coming up and yeah the models currently they're supervised, right? So when the models eventually mature more and more, they'll start to learn itself like teach itself or to become better. And there are a lot of cloud platforms that basically companies which are generic and they offer predictive maintenance for multiple. So that's a evolving model that's coming up which will basically an outsourcing sort of the company themselves only it or the manufacturer of the equipment. It could be a third party cloud platform that could take care of it for different systems. So to conclude the. Predictive maintenance using iot delivers unprecedented insights and operational excellence. And it's going to grow in the future. And a lot of equipments, even some that are at home will eventually be having some kind of predictive maintenance in the future. And they deliver a actionable, intelligent intelligent plan. To call to action, to to resolve the issues which will basically ensure the overall reliability of most of the systems, thereby increasing the the quality of life, making the company's profitable and so on. Yeah, I have added some references which would go through. Hope you like it. Yeah. Thank you very much.