Welcome to the Trend Detection podcast, powered by Senseye, an industry leader in using AI to drive scalable and sustainable asset performance and reliability. This is a new publication designed to help you go away with ideas on how to achieve maintenance efficiencies.
For this 3-part series, I’m joined by Burak Polat, CEO at Skysens, a company that is focused on creating user centric IoT infrastructures for every enterprise.
In the second episode of this series, we discuss how to implement agile predictive monitoring for maintenance, how mature manufacturers are in terms of collection and quality and much more.
Key topics covered (click to jump to the section)
- How to implement agile predictive monitoring for maintenance
- Resistance to change within organizations
- Key Issues
- What data is collected?
- Examples of successful deployments of agile predictive monitoring
Niall Sullivan, Senseye:
Welcome to the Trend Detection Podcast, powered by Senseye, an industry leader in using AI to drive scalable and sustainable asset performance and reliability. For this three part series I'm joined by Birat Pilat, CO at Skysens, a company that is focused on creating user-centric IOT infrastructures for every enterprise. In the second episode of this series, we discuss how to implement agile predictive monitoring for maintenance, how mature manufacturers are in terms of data collection and quality, and much more. I hope you enjoy it.
Burak Polat, Skysens: I guess I'm just thinking from a maintenance perspective. I know you might have touched on that just before actually, but maybe if we saw the step by step approach for implementing it for maintenance, I guess, implementing our job, predictive monitoring for maintenance.
Niall Sullivan, Senseye: For maintenance, actually we do multiple things at the same time. We provide edge computing capable vibration devices, which can communicate wirelessly, from very long distances. We provide this wireless network capability as an embedded part of our platform. It's integral part of our platform. We provide support to deploy those vibration devices. We do also temperatures for different machinery. We do ambient temperature monitorings, and also the energy related sensors, like current sensors or monitoring the voltage of fluctuations
Burak Polat, Skysens: We also provide that application, but in addition to that, we also provide the pressure sensors and also the flow sensors in order to understand if the systems is working correctly or not. We create, we support different users to the visualization and to the integration. So some of the users can use the platform only for monitoring purposes. Some of the platforms can directly flow that data to some third party platform or do more analytics on it. So we provide this predictive maintenance application on top of our predictive monitoring platform with these capabilities.
Niall Sullivan, Senseye: And it's interesting there. So one of our questions we've got written down here is around sensors and it's been mentioned a few times, obviously quite key, but I guess it'd be useful for people to understand. Do you need sensors to enable all of this great stuff or can you leverage existing data, I guess, or is it a bit of both? I don't know. Is it identify... Yeah, I'll let you explain rather than me.
Burak Polat, Skysens: Actually we we don't have to use additional hardware as always, but we see a lot of benefits actually out of it, but we don't have to use that. I mean, one of our key expertise is data handling the integration industrial data into our system. So, I mean, we can handle the existing industrial data very easily from the industrial data sources. For example, we can communicate different OPC servers simultaneously with the Modbus TCP PA based PLCs or other protocols such as PropaNet.
We can communicate with them in two ways. We can deploy our system on edge or on cloud, or we can do hybrid deployments for these integration requirements directly to industrial systems. So in that sense, we don't have to use additional sensors at all, but to include the front lines, to include operators who are doing walkarounds all the time, who are testing the equipments, who are testing the assets, industrial assets, checking their performances, checking their status in terms of maintenance. We find it very helpful for that to provide this edge device capabilities with long range wireless network.
For example, for large facility, we can just double a couple of gateways, I mean low cost gateways, and we can create a coverage, the network coverage, and then we can just plug and play some sensors and integrate it with the other existing industrial data sources. Do the visualization, do small analytics and then integrate to existing predictive maintenance or IOT platform if they like.
Niall Sullivan, Senseye: And in terms of the... Oh, sorry. Sorry. I'll let you finish.
Burak Polat, Skysens: I mean, in essence, we don't have to work with the additional hardware. We don't have to bring this hardware to the table, but I mean, we find it very helpful for the user.
Niall Sullivan, Senseye: No, absolutely. What I was going to ask, actually, in terms of the type of companies you speak to, how would you... And it's a pretty broad question, so it might not be the easiest to answer, but in terms of maturity, in terms of the data they collect, the systems they have in place, the hardware they have in place, how mature in general do you think companies are and how ready are they for to implement this type of solution or these types of solutions?
Burak Polat, Skysens: It is a broad question, yes. I mean, it entirely depends on the maturity of the manufacturing enterprise in terms of digitalization. So in that sense, if they thought that data will be necessary for their main operation, which is a production, which is a manufacturing, then they already collect the data. But if they didn't think that, if they haven't think that before that, before today, they don't have the data. That's as easy as it is. But even if they collect the data, we always struggle to find the database, actually, because sometimes they deploy something, they are running something, but they don't know where the database is. I mean, they don't know how to normalize that data because they only deploy this network, this end to end data flow, only for a single application, mostly for MES or mostly for ERP reasons, ERP integration reasons.
So their data usually are not ready for this analytics or for IOT requirements. So these sometimes provide some engineering solutions to them as an integral part of our platform, but sometimes they don't even have that. They didn't communicate with their missionary, their data sources. So in these scenarios, we bring our additional hardware to them. So, I mean, usually, data is siloed. We can summarize like that. Data is siloed. And also data is being designed only for one or two main prominent application that they develop couple of years back, like maybe 10, 12 years back, for the support of the manufacturing operation. So for companies like us, if they say they already have a data, if they already have a connect, communicating with their missionary, we always need to check what is the current status of that data.
Niall Sullivan, Senseye: I was just thinking, actually, when you start working with your customers, or prospects, let's say, but do you find that there's sometimes resistance to change within the organization and you have to sort of... Again, it's part of that education piece we were talking about earlier, but do you find that sometimes there is resistance to change within organizations? And how do you overcome it, I guess, in those situations?
Burak Polat, Skysens: Yes. I mean, yeah, sometimes. Not always, but sometimes. If the company haven't already created some initiative to digitalization, I mean, we struggle with this resistance usually. I mean, industrial enterprises as a nature, I mean, they are conservative places. I mean, they are conservative companies, right? Because everybody is trying to be as efficient as possible, be as strict as possible, because it's an ongoing operation. If you don't be efficient or if you don't do your job right, then it means that there can be some catastrophic event happens, right. So they have to be conservative. I mean, their nature evolved around that.
So because of that, when you bring some solution which you say you are solving something, I mean, to their understanding, they always, I mean, not always, but sometimes they resist to that change. So in that sense, if it happens, we try to understand what is their current pain points and evolve our application or evolve our discussion according to their ongoing discussions. So another way is to do it with a small step. One of the key aspects of agile predictive monitoring platform is they can start small. They can start with only one or two devices or one or two data points, and then see the result and they see it using. I mean, I'm not talking about pilots or demo projects. I'm talking about real deployment with a small scale.
Niall Sullivan, Senseye: Yeah. So yeah, what'd be termed a joined up approach, across departments. I mean, it's the analogy of, there's silos within different systems, but it's also breaking down those silos between departments as well is just as important and showing the benefit, how it could benefit them and not just the maintenance team or other departments. Yeah.
Burak Polat, Skysens: Of course. Of course. I mean, not only the data is siloed, but sometimes people are siloed. Also the teams are siloed.
Niall Sullivan, Senseye: Yeah. I think that's natural, especially when you get towards the larger multinational companies as well. Think it's only natural these things happen. But what I wanted to touch on now actually is data and I guess the challenges around data quality that you see. So, yeah. So first of all, what challenges do you see around data quality? And maybe another broad point, but are you seeing improvements in terms of data being collected? Again, it's a broad question, but with the issue side of things, where issues are around the data quality, how do you overcome those? Or where do you start?
Burak Polat, Skysens: Actually, we start with the application, with the end result. So if you are talking about, I mean, that's where our partners come in too. So if they are talking about very detailed, very scaled application, we try first to make them start with a small step, right. But if their end result is to improve asset performance by, I don't know, 20% in every facility in everywhere in the world and in their company, we check their data first. I mean, we check what we already have, right. For example, sometimes they say that they have data, but as I mentioned earlier, they only have data for MES purposes, for the production purposes, usually. That's there because usually it's a starting point for the digitalization.
So when you check the data, you see that they only have for every operator change, for example, every six hours of something, it's kind of a meaningless, if you're talking about full scale deployment, right. So, I mean, if we see that, we go back and we try to design some kind of data capture strategy. Either we will use the same infrastructure, same network, which is already creating data to improve that quality, or we will just deploy another sensor and get the data from every five minutes readings or something like that instead of six hours.
So entire list, it starts with the end result. So some datas can be useful for some applications, but sometimes they're not useful for others. So it makes perfect sense for the users too, to start with the end result, but sometimes you also need to check the infrastructure. For example, when you see the data... I mean, we also have some data analysts in the team. When they check the data and they can understand maybe the physical source of the data is malfunctioning time to time, right. Or it is not sensitive enough or it has some hysteresis around the data point or something. So it is also a secondary strategy to go back to the data source and tell the customer, okay, this data source is not suitable for that.
Not because of the data interval, but because of the data quality, the creation of the data quality, and also because the company started with MES requirement at first, like 20, 30 years ago, they don't keep data that much. When you compare with the financial data or the ERP data, you can find historic data very easily. But if you are talking about physical data history, they don't usually have anything, if they already haven't start this journey. They only have like one month of data or something. So, I mean, back to the beginning, you have started with the application in order to decide whether data is good enough or not.
Niall Sullivan, Senseye: Absolutely. And in terms of the examples of the types of data that might be useful to collect, I guess depends on the use case, but if we talk specifically about say maintenance, what sort of data should they be collecting from their machines?
Burak Polat, Skysens: You mean as a customer or what we collect from the machines?
Niall Sullivan, Senseye: Yeah, both. Both, I guess. Yeah. So what would you recommend, I guess, for customers? What types of data should they collect from their machines to enable a program?
Burak Polat, Skysens: I mean, for the maintenance purposes, I would always recommend them to gather all the physical measurements from the machine which they can. It is irrelevant which product are they are using, because you are talking about the maintenance, you are trying to understand the asset, but it will be always helpful to collect the work order from the ERP too, from the MES too, I mean, because the machines can act differently according to different product type or something. So you need to get that data too. It is very hard to say before understanding the scenario or before understanding the operation which the machine is doing. So, I mean, in general, we can say, gather every data as much as possible, whatever you can from that particular equipment or that particular machinery, but you don't have to gather it every second or less than a second or something. I mean, it will probably be okay to get the data every 10 or 12 seconds or something like that. It's a good starting point for the basic algorithms.
Niall Sullivan, Senseye: Yeah. And actually something we talk about at Senseye, I think it's one of those misconceptions that you need lots and lots of data to get started. There's obviously the phrase big data, but we talk a lot about small data. Is that a term that resonates with you as well, from your experience?
Burak Polat, Skysens: Yeah, of course, of course. Yeah. We always say that. I mean, in order to create a quality end applications and quality algorithms, the data quality is much more important than the amount of the data. So I think it is because of the marketing that started like 10 years ago, about all the hype around big data thing. People still think that getting data without any quality check, without any strategy or something as much as possible makes sense, but it just creates other problems. I mean, it just creates data handling problems, it just creates database problems, et cetera. So, I mean, I think they should make sure about the quality and the existing of the data first, then having as much as possible kind of a strategy. So data quality is much more important than the amount of data, if you are trying to go somewhere which makes sense with the predictive maintenance or predictive quality or energy efficiency applications.
Niall Sullivan, Senseye: Yeah. And I guess another thing we talk about regarding data is also this context, as well. It's the context around it that's important. So bringing in different data sources to keep adding maintenance data and manufacturing and just keep adding it in, and the more data you have in that sense, good quality data, sorry, the bigger, the better picture, it builds of, let's say, the health of a machine in that particular use case.
Burak Polat, Skysens: Correct. Correct. I mean, there is like two ways of doing that. Either you can think it as asset wise. I mean, if you think asset's life cycle, asset does multiple things. It manufactures, it gets maintained, it consumes energy, it consumes some other, like spare parts or something, and it has a life cycle, right. You can think it as around the asset itself, but I think the more and better way, I guess, is thinking it as a process wise. There are lots of process are going. I'm not talking about only the manufacturing process, but also maintenance processes, quality check processes, logistic processes are ongoing industrial facilities. If you think about the process wise, it makes a lot more sense.
So I think digital teaming thing, digital teaming hype, as much as other technologies are the technology hype cycle. Digital teaming at the first point was kind of a meaning was in some part. But if you think that digital teaming as a process team, as a managing the processes, checking all the processes ongoing in the industrial enterprise facilities makes a lot of sense, that seeing things. Because the process mining applications, process mining platforms are very useful for finance world or other systems, and I think it makes a lot of sense for the manufacturing industrial world too.
Niall Sullivan, Senseye: No, absolutely. Absolutely. I just wanted to touch again on the hardware side of things. So we've talked a lot about the sensors, and I think there's been some mention other types of hardware, but yeah, so I guess I wanted, just for this data capture strategy we talk about, so what other hardware is required, or would be recommended, let's say, if it's not necessarily required, but what can help facilitate the data capture process?
Burak Polat, Skysens: Yeah, I mean, in our wireless network, our wireless technology, I mean, we develop it for industrial requirements, but it is LoRaWAN technology compatible system, compatible network. So we provide edge devices which can communicate with LoRaWAN network. But in addition to LoRaWAN, it again starts with the application requirement, the end result, the target result, actually. I mean, I recommend also to think about 5G gateways in order to get faster data, more data. 4G will be enough for the most cases, but if can think with the 5G too.
But I mean, if our users are also talking about real time control of the equipment itself, we always recommend them to go with the cable, with the wired solutions, such as fiber or ethernet fiber or PropaNet, for example, with the cabled PropaNet protocols, things like that. So it has a wide variety of devices of the hardware that they can use, but I mean, it all ends up with the end result and targeted result. But if you want to generalize, we work with gateway devices, we work with different RTUs, industrial gateways and also edge computers, which have like RTU capabilities, gateway capabilities also, so on and so forth.
Niall Sullivan, Senseye: Yeah. Excellent. I just wanted to touch again, sort of take a step back a minute, and just look at... Well, what I'm looking for actually is some examples of successful deployments of agile predictive monitoring, in particular for maintenance programs, I guess, if we focus on that. So have you got examples that you can share that stand out to you?
Burak Polat, Skysens: For the predictive maintenance?
Niall Sullivan, Senseye: Yes.Yeah
Burak Polat, Skysens: Yeah, sure. I mean, for example, we work with large automotive companies. Some of the companies, they already started agile monitoring teams, agile monitoring initiatives, inside their production facilities. And we work with large automotive brands who work with hundreds of thousands of different suppliers. So we help them to understand their supplier situation in terms of the asset, in terms of production capabilities, since our system can easily be deployed and their suppliers are usually lower technological maturity level compared to OEM itself. So it is very helpful for them just to offer something to their supplier in order to make more efficient production, more efficient manufacturing, and more sustainable manufacturing for the company itself. So we work with multiple automotive companies in that sense.
We work with airports, for example. Istanbul airport case is a good example. They have hundreds of thousands of different front liners working in the technical areas, different areas, open areas or closed spaces, for different infrastructure. For energy infrastructure, for water, for, I don't know, for other maintenance related operations that they're conducting. We help them to pinpoint the problem very easily and to alarm the related partner, related user. And then continuously checking it whether the problem is already solved or not with this system, very easily. We also had them to implement in more detailed, predictive maintenance applications on top of our main platform.
Niall Sullivan, Senseye: And for those successful, say, deployments, how important was stakeholder engagement to those? And which... Well, actually, I know we've covered that to a degree, so maybe we could focus on instead who are the key stakeholders in these type of projects that need to be onboard? Is it IT, is it... And what role do they play in the project?
Burak Polat, Skysens: I mean, we always try to engage with the C level first and then go and take the approval of the frontliner, because industrial enterprises, yes, the initiation comes from the C level and they hold the budgets, right, they hold the approval award for the project to be started, but it always need to be approved by the engineering manager or the frontliner itself. So we first get approval about the idea of the C level and then go down and tell the frontliner about what we are going to provide him or her to make their life easier and more efficient workplace for them. Because they do multiple things and they have other problems in their mind. So the engineering managers, the C level and the frontliner is the three main stakeholder that we usually bring to the table.
But also we always take the approval of the IT managers too, because nowadays industrial security, industrial network security, becomes a really important thing for the companies. And it is becoming common the OT networks are merging with the IT networks and the IT managers are getting more and more interested in this OT network, OT systems, so you need to get their approval too. But for us, it is much easier to get their approval because we can talk with operational people, frontline people, and talk about maintenance, talk about energy, et cetera.
But our system is very capable of integrating to other cloud platforms such as Microsoft or AWS, et cetera. So, I mean, it is easy for us to converge those two words and take the approval of the IT managers. So in vertical line, in vertical levels, it looks like that. But if you think about horizontal, vertical, horizontal organizations, we usually go for maintenance teams in order to create predictive maintenance applications or energy efficient applications, try to make it at the same time for the maintenance people. But we always talk with quality people, quality teams, and also the production monitoring teams at the same time, in order to make the project successful.
Niall Sullivan, Senseye: So that was the second part of our series diving into agile predictive monitoring. I hope you enjoyed it. An important takeaway for this episode is, before focusing on the quality of your data, you must first identify the result you would like to achieve. For example, improving asset performance by 25%. Please subscribe via your favorite podcast provider if you'd like to be notified about future episodes, and it would mean a lot if you can let us know your feedback by leaving us a review. You can find out more about how Senseye can reduce unplanned downtime and contribute towards improved sustainability within your manufacturing plants, by visiting senseye.io. Thanks a lot for listening.
Subscribe to our Trend Detection podcast
We post weekly episodes on a wide range of different subjects including predictive maintenance, industry 4.0, digital transformation and much more.
Here are all the places you can access/discover the latest episodes of the Trend Detection podcast: