The 5 phases of a successful Predictive Maintenance deployment
Show notes
Welcome to the Trend Detection podcast, brought to you by Senseye Predictive Maintenance – the platform which enables predictive maintenance at scale across all of your assets, across all of your plants.
In this episode we share a tried and true methodology to implement successful and scalable PdM projects using our industry-leading Senseye Knowledge Platform. We will cover the five phases crucial to a successful deployment.
We cover:
Scope: Discovery, putting together a plan, defining success.
Design: Creating a detailed project design and machine selection.
Deploy: Set up and technical implementation.
Operate: Effective usage of Senseye, learning and tracking success.
Refine: Gathering and using feedback
Transcript
Welcome, everyone. I hope everyone is excited to learn how to get started with Sensai Predictive Maintenance. As Dia said, I'm Liz McGinn, the demand generation lead here at Sensai in the Americas. I'm joined by Doctor Natalie Kergen, the the delivery training lead for Sensai Predictive maintenance at Siemens. Today, we've pretty much been explaining Natalie's role with the help of companies, um, With helping companies get started with predictive maintenance. Before we get started, let's kind of go over our agenda. So we'll start, start with what is predictive maintenance, then we'll go into what is Sensii predictive maintenance, and then a little bit into our Senseii knowledge platform. Um, but before we get started, um, with Natalie's full job, uh, Natalie, do you want to explain a little bit more about Sunset Predictive maintenance? Sure. Um, before we even talk about what Sensai is, I think we should cover a little bit about what is predictive maintenance first, because sometimes there can be some different definitions floating around. So let's just make sure we're on the same page. Now, the precursor to predictive maintenance is condition monitoring. Uh, which is using sensor data from machines to support decision making in your maintenance journey. Condition-based maintenance builds on that and uses the data that you're collecting to replace planned maintenance. Um, predictive maintenance furthers that more by using online monitoring so you can monitor thousands of machines rather than having a single person, um, analyzing data. And the difference as well is that predictive maintenance is going to be using AI and machine learning, different programs to. See how data is trending rather than just telling you right when a machine is experiencing a degradation or a failure, something like Sensi would tell you when it's on its way towards experiencing a degradation. Um, so this way it can minimize disruption to operations and your maintenance by making sure you have everything you need to perform, uh, maintenance, and to even know if your machine will make it to the next planned maintenance event. So, let's get into what is Sensii. Now, even though predictive maintenance is in Sensai's name, it's important to not mix up. And conflate Sensei with being predictive maintenance. Predictive maintenance is a strategy. Senseii is a tool to help a customer achieve that strategy. Um, so what it's doing is it's using asset intelligence, um, to To help. Increase like efficiency, essentially, I like to say that one. it helps increase uptime of your machines. It reduces planned and reactive maintenance, which reduces costs, um, supports remote work, and increases sustainability. So let's just get into how it works a little bit before we get into how we onboard and start somebody with using a, a technology like Sensai. So this is sort of the roadmap of what Sensii is. I have this laser turned on. I'm gonna turn that off so I can click and you don't get distracted. So the way Sensii works is the first thing that happens is we have to take data from your machines. Now, this could include a sensor data that is coming directly from the machine itself while it's running. It can include information from the PLC like, you know, control signals, and it can also include information from maintenance. So as you're interacting with the machine, if there are hands on the machine, you're gonna want to tell that information to Sensi so that it knows to see changes in the data. All of this comes into uh our cloud-based application. And it's fed into our pattern engines. So the pattern engines are, I would say half of the, you know, the AI and the, the magic of Sensii. I kind of hesitate to use that word because there's a lot of logic that goes into it. Um, so the pattern engines take all of the data that you're feeding in. They establish what a normal baseline looks like for your data, and it takes minimum 2 days to do that for some of our more basic patterns. Uh, so they establish what a normal sort of fingerprint looks like. And then as your data and, as it moves forward in time, deviates from that fingerprint, The app will generate patterns to categorize those changes in behavior. So these pattern engines can run automatically with no additional input. Um, but you can also feed in, if you've seen failures in the past that are consistent and you have the right data for them, you can feed that in. Uh, to match to them in the future, and you can feed in expert knowledge as well, additional information, whatever else you know about this machine and how it fails and how it looks in the data, you can give that all to the system in various ways. So, coming out from the pattern engines are all these different types of potential patterns that can be detected. Um, I'll show a couple examples of the most, you know, default ones. So the top is anomaly. Now, anomaly is a short-term, short duration change. So there's a step change or a spike or a burst of spikes like what we're seeing here. Uh, the next one is a trend, which is a long-term baseline shift in the data. It can be going up or down. Um, those two are the ones that are gonna be run automatically out of the box. You feed your data into Sensa, it automatically starts looking for these with no configuration. Um, the caveat is you've got to be providing the app with condition indicator data, uh, which we work with our customers to help derive those and figure out how to do that. Um, what a condition indicator is, it's data that in normal operation of the machine, this is oversimplified, but in normal operation of the machine, the data is steady. And when the machine is in a state of degradation or failure, it's not steady. It's doing something different. So a change in the data represents a change in the, the condition of the machine. So we help our customers to drive those and make sure we're feeding that into the system in order to get these anomalies and trends. Uh, the other most basic type of pattern that we detect does require some expert knowledge, which would be a threshold. So if you see this gray line going across here, I'm not sure how easy that is to see. That indicates sort of like an alarm level. Um, if your data crosses a threshold, you will get a notification every time. Um, you only get one for as long as it's violating the threshold. So what we see here in green is basically the notification that you would be given for that duration of time. Um, we do have a bunch of other patterns that we can detect. Uh, another one related to threshold, I don't have another image, but it's called a forecast, where you can enable Um, an alert to be given to you if your data starts trending towards a threshold. And it will tell you approximately how much time you have until it gets to the threshold, you know, depending on whether or not you're interacting with the machine to get it to go away again. Uh, so those are 4. Um, basic types of patterns that are detected. Now, aside from thresholds, when an anomaly or a trend is detected, you don't necessarily automatically get a notification, which in Sens that we call a case. Um, instead, all of this pattern information is fed into our attention engine. So this information comes in here along with configurations such as asset criticality. So if you set an asset as being highly critical, you'll get a case a lot sooner than an asset that you said is low criticality. Um, there's additional configuration we can do as well. So all of that comes in, and if it's determined that the change in your asset is critical enough, it will raise a case. Over here. Um, this case opening is dependent on a value of attention index, which is just a value of how how severe. Um, the change in the asset is, how much it needs your attention. Each case will come along with evidence, uh, which is basically just a, a chart of the data that has a pattern on it. So you can review the evidence and provide feedback. In the form of case evidence and case closure, and, and even in just interacting with someone like me who can go in and tune the pattern engines and the attention engine to improve uh detections in the future. I forgot to click, sorry. Um, I hope, hopefully that all makes sense. That's the, a kind of oversimplification of what's going on here, but I think it's a pretty good roadmap. OK. There's no questions, right, Liz? Nope, no questions so far. Okey doke, thanks. In that case, uh, if you have any questions on how it works, drop them in the, the questions chat, and we can get to them. In a bit. Otherwise, let's just get into our process. Now, it's all like since I Um, We, we want to take a lot of care when we onboard a customer with Sensei because the, you know, how well we onboard can affect the amount of success that we have. So, we have something called the Senseii Knowledge Platform, and this was built by us using our best practices that we've developed over years of, of doing these deployments. It has 5 phases, scope, design, deploy, operate, and refine. Scope is pretty much everything that's done prior to a project charter. So that's before someone like me would get involved. Um, That's done to figure out, you know, what assets are involved, um, what sort of benefits might a customer have. And as soon as the project charter is signed. Someone like me in delivery will come in. And start designing the project. OK? So, design phase is exactly what it sounds like. We will do a number of asset discovery sessions to figure out what are your assets. Um, a lot of times we have to talk with customers to figure out what do you define as an asset and what do we define as an asset, and how can we sort of consolidate those two. We'll talk about what data you have available, how do they fail, you know, what are the pain points that you have with these assets, and how best can we configure them in the app to get you the results that you're looking for. As we're doing these asset discovery sessions, we're also having integration discussions to figure out how are you gonna get your data to us. Deploy phase is all about getting things in the app, um, get the customer account configured, start getting data in, configure the assets, and do training. Operate phase is just using the app, that is your regular daily use. Um, and we have different things to help with that, and we, we also help with case review sessions as we go. I think one of the really important things to note about Sensai is that we don't just hand you the app in the, the, you know, these steps here and say, all right, see you, good luck. Figure it out yourself. We work with our customers. For, forever and help them with their case reviews, help, you know, get feedback, what do you like, what do you not like? How is this working? How can we improve it? And that feeds into refined phase, which is when um the account relationship is in the hands of a customer success manager. And really, it continues the process. It's really a cyclical. So you might go back to redesign some assets if they're not performing very well. You might scale and add on more assets, um, And, and have those sorts of discussions. So this is sort of an example of what a project deployment could look like. Again, I'm not gonna read all this again because I kind of went over that. But I wanna draw your attention to the, the timeline here with all these little cyclical aspects, um, because it's not linear, and it's kind of really hard to represent that. And each of the steps that we have for design, deploy, operate, and refine, we just use as needed. So there's a step in deploy, for instance, for procuring sensors. Well, if a customer already has sensors, we're not going to do that step, right? So this deployment plan is really made custom per customer as we start working with them. Um, so you'll notice here. That right when the kickoff meeting happens, that is when delivery gets involved. Um, and we start setting up a timeline. What do we do next? Who are people involved? And then we start getting assets onboarded. Um, and even as we're onboarding assets, we might still be going through the design phase of other assets. Um, as we're working and operate with some, we may still be deploying some others, so it's very, very cyclical. Um, until we get into sort of a steady-state period, um, in which we would just be doing sort of periodic check-ins. Customers don't need us so much, but we are there when they do need us. All right, let's. Awesome. So that will pull us into our first poll. Um, so our poll question will, uh, ask it so everyone can answer. Natalie will go into a little bit more and then we'll come back to the answer. So if you only had 50 assets you could onboard into Sensai, what would you choose? The complex assets that cause the most issues in downtime, simpler assets of the same type, simpler assets of different types, or assets that are interesting to end users regardless of the type. And while you guys are answering that, Natalie will go into the design phase. Yes, ma'am. All right. So take your time with that. We will get to your answers in just a minute. As we pull up our description of what we consider design phase. So design phase consists of 4 steps, and again, they might not happen in this order. Um, I'm working with a new customer right now. We are doing machine selection, failure mode, and data selection as one piece, and data architecture is happening in parallel with another group of people. So just as an example of how we might custom deliver a project. So the whole point of design is to choose what assets we want to onboard, understand their failure modes. Um, understand how their data relates to those failure modes and figure out how we're going to configure it in the app. It might require filtering the data, it might require transformations, which we can do in-app. Um, and from there we, we decide how we want to integrate. It's really how the customer wants to integrate. And very importantly, we determine success criteria. We want to do this early on, um, so that we don't get to, you know, renewal a year, a year later, and, you know, they're, they're not sure if they've had success or not. So we start that early on to make sure that we know what we're looking out for. In design phase, typically the people involved would be the machine experts, because we need to talk to somebody who knows the machines and knows what the data might look like for them and um so we can really work together to figure out how to monitor them. And the Siemens delivery team. I will say in design, I don't want to confuse anyone and have them thinking that we are building custom models per asset. Um, the AI and machine learning builds the model. It does it on its own. All we are trying to figure out is what is the condition indicator. And for instance, if you're onboarding, let's say vibration data. OK, it's a very common condition indicator, but as it is, coming straight from the machine, it might not be a condition indicator because it might be including data when the machine is off. So, the first easy step is, OK, remove data when the machine is off, so we can filter that out and be providing a steady signal that's not dropping all over the place. Um, the other thing to consider with vibration is, is it dependent on any external factors? If you have a motor running on a VFD every time your speed change, changes, your vibration is going to change, but it doesn't necessarily indicate failure. So we can set up that correlation in the app between speed and vibration. That's what we're talking about, um, rather than actually going in and configuring, um, the actual algorithms that are looking for patterns, we just try to drive condition indicators early on. Liz, do you want to read the results of the poll? Or do you want to give it a minute? Nope, it's up. Um, so the audience has chosen the complex assets that cause the most issues in downtime as the, um, choice for the 50 assets. What do you think about that, Natalie? OK. I think that's a really common route that our customers usually want to go down. And we actually try to steer them away from that. Um, those most problematic assets, are not the best choice usually for two reasons. Number one, if they're all, if they're really problematic, they're probably being monitored really heavily already. So what value is Sensi going to show? It might just prove that it works, but we know it, it, it works. So, um, it might not be actually adding much value cause you're already monitoring it. The other reason we wouldn't want to start with that one is because it might be a very complex asset. And in the very beginning, we are coming from very separate spheres of knowledge. I know the app really, really well, but I don't know anything about your machines. You know your machines really, really well, but you don't know anything about the app. At some point, we have to kind of learn about each other. And once you kind of know what the app needs to look like, what it needs to see in the data to be successful, then we can really work together a lot better to derive those condition indicators. Um, so we would usually recommend, let's start with simpler assets that are more basic. And honestly, if you start with simpler assets of the same type, get them all on the app, get them generating value, while we then go back and revisit your more complex assets, so the app is working for you while we try to figure out those condition indicators. Does that make sense? That makes great sense. Could you explain a little bit on like what is an asset, because I know you spoke about it, um, that we have two different definitions usually from our customers. So maybe it's best to kind of explain what an asset is in our eyes. I'm really glad you reminded me to do that. Um, so an asset according to Sensai is the smallest replaceable unit of a machine that has data. It's the thing that you're maintaining. Um, you don't want to have like some big, let's say, let's say you have something that you call a fan, and you say, all right, I want to get this fan into the app. But actually, what you're calling a fan is a motor, a gearbox, and an impeller. To us, that's 3 assets. If you have data specifically on the motor, and specifically on the gearbox and the impeller, that is 3 assets, and we should monitor them separately. So you're getting cases that point you to exactly where the problem is. Also, when you split them up, driving condition indicators is can be a little bit simpler than for a wider, like a larger machine that might be more heavily influenced by process variables. Not always the case. Um, so that's what we would consider to be an asset. Um, I would also say from that poll. Um, While we do tend to start with simpler assets of the same type, Sensai works great in bulk. So if you want to onboard one motor, or 100 motors, to me it's the same amount of work. Um, so I would just say get 100 of those in. It's the same amount of configuration if they're basically the same thing. Get 100 of those in, get them generating value. Let's go back and look at the more complex ones once that's done. Um, but the other thing that's important not to overlook is to consider assets that are actually interesting to the end users, regardless of type. We want to balance. Whoever's using Sensii should care about what's in there. So, we try to balance the two. Um, so I would say that's one of the main roadblocks is just what assets to pick. If we go the wrong route, it can really influence the success of the project. Uh, and we've done this so many times that we kind of, we kind of know, let's start with the simple ones that you have many of, and a few that are interesting to the end users if they're not the same. And let's go that route. Uh, the other roadblock is that the integration design is always the blocker. …