Knowledge Builder and Generative AI - with Dan Reid
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 are joined by Dan Reid to discuss the features and benefits of Knowledge Builer in Senseye Predictive Maintenance.
You can find out more about how Senseye Predictive Maintenance can reduce unplanned downtime and contribute towards improved sustainability within your manufacturing plants, by visiting: www.siemens.com/senseye-predictive-maintenance
Transcript
Welcome to the trend detection podcast powered by Siemens, and now your host Nel Sullivan. Hello and welcome to the latest episode of the trend detection podcast. I'm really pleased to be joined by my colleague Dan Reed. Hi, thank you for having me. Um, once again, actually, so you're one of our regular guests now, so it's really 3rd time, 3rd time too many. 3rd time lucky, I hope. But today we're talking about, um, another enhancement to our maintenance co-pilot, um, within the Sensi Predict maintenance platform, of course. Um, so we're gonna dive into that. That's Knowledge Builder, um. And yeah, we're gonna go into that today, but um I guess first of all, Dan, um, let's sort of zoom out I guess because we wanted to look at sort of the, you know, the context to to all this, what, you know, why we've developed it and we'll talk about the actual functionality, but if we just start with the, the big issue in industry today, I talked to a lot of people um around knowledge loss, um, in industrial maintenance, um, and what impact that's had on operational efficiency. Yeah, so it's certainly, you know, we, we talk to customers a lot and obviously a lot of what we talk around is predictive maintenance and downtime and, and related topics. But the other major topic which gets brought up is, is knowledge and capturing it and surfacing it. Um, and uh there's some quite interesting reports on this topic. So, um, Deloitte published a report and it said that 62% of manufacturers. Uh, their biggest issue is, um, talent recruitment and retention, and that's their biggest issue. It's not one of their biggest, it's, it's the biggest. So this is a problem that's very much front of mind for manufacturing and our customers in, in particular. Um, and I think it stems from sort of 22 problems. The first is there's, it's an aging workforce. So in that same Deloitte report, they identified that 25% of manufacturing employees are 55 or older. So that's a big portion of, um, you know, manufacturers, maintainers who are going to be heading towards retirement. And obviously through retirement. These organizations are losing the skills, the experience, all that knowledge, um, from these experienced, uh, individuals. And then on the other side of this problem, over the last sort of two or three decades, there's been a declining interest in manufacturing, mechanical engineering, and, you know, all of the related, uh, topics. And it means that in the job market, you just don't have the people, um, who have the skills, uh, to, to backfill these, um, these positions. So it does lead to quite a a troublesome situation and um in terms of efficiency, what it means is that you're having people who are less skilled at maintenance or less experienced. Um, the maintenance obviously takes longer, um, it's more prone to mistakes and that will have a knock-on effect to, um, up time, mean time to repair and just general efficiency of the assets and, and the sites. And I guess if we're, because I mean, even when I sort of started within the sense our team that the talk was a lot about unplanned downtime. I mean, that used to be the focus and maybe and it's still a very important metric, but what we're saying is this is sort of shifted, I guess in the past maybe 123 years towards, you know, listen. Becoming a real issue now. Yes, yeah, very much so. So, you know, it's downtime and it's still an issue and making sure that you're doing precise maintenance is, is still an issue. But yeah, it's certainly how, how do you go about capturing all of the knowledge, um, especially as people are retiring, you want to capture it before they, they retire, and then making sure that that knowledge is accessible and you get value from it. Um, and it's, it's a difficult problem, um, and something that, you know, over the last maybe 2 and a bit years Sens Ii has been focused on, um, uh, and you know, we're seeing great, uh, benefit from that for our, for our customers. And I, and I guess the, um, the, the maintenance co-pilot to start with has been a big step towards that. So maybe for, for people maybe not 100% aware, we could talk about sort of the development of that over the past year, year or so in general and where we are today. Yes, so, um, this was, uh, I think we released it probably a year ago at this point, um, and there's sort of multiple elements of it. First of all, within Sensi now when we raise an alert to, uh, a user. Um, so that we believe, uh, an asset needs attention or it needs maintenance. Um, we obviously provide, you know, details of what we've, we've seen in the data and graphs alongside it. Um, but now you also have an interactive co-pilot you can talk to and it's like any chatbot that you're used to, um, but within, within Sensei, and it has the, the context that it needs. So it understands the asset you're looking at, the case, um, you know, what we've seen in the data. Uh, historic cases which are of, of relevance, recent work events, all of this information it has, has available to it. So you're able to ask questions about, um, you know, what, what's the situation, what should I do next, um, and it can help guide the user, and this is coming back to this, um, this issue of, you know, having maintainers who may be less experienced, um, so we're trying to provide guidance to them. And uh just one more thing that I think the more interesting topic is obviously it's nice having that interactive co-pilot to talk to, but it's even better if you don't have to talk to the co-pilot, right? So when we raise cases, not only do we present the information we've seen, we also then provide prescriptive guidance alongside it. Um, and you know, that we suggest what are the potential root causes to this problem, so that, you know, the engineers can, the maintainers can look at the, the data, look at the, um, potential root causes and know what to, what to do next. No, no, absolutely. And what, what you talked about there within the sense applications, there's already a vast amount of data which the co-pilot could use to provide that prescriptive guidance, whether in, in an interactive way or in the, um, in the cases, which again, I think you raised on one of our previous, um, when we launched it, it was very important. You sort of told that story of the, you know, that there's a great summary here. So people like me, I could go in there and understand. Stuff straight away, which I thought was really cool. Exactly, exactly, really cool functionality. But all that knowledge within the app is one thing. But, um, and back, I guess to our first question, and we talked about the knowledge that's actually being lost, um, due to retirement. So there's knowledge within the app, but there's also all this other knowledge that's leaving the organization. So what, what type of knowledge is sort of, you know, leaving with retirement. People changing industries, etc. Yeah, a lot unfortunately. So I, I think this is, um, you know, the individuals who are retiring have been in the industry for decades. They've built up loads of knowledge through that time and especially that that generation, actually people's tenure at their current employers are normally longer than, you know, what, what we might have in our generation. So you might be talking that they've actually had decades of experience in that factory, right, working with them assets, working with the processes. And they know it inside out, really in depth, and you are, you are losing all of that, um, and it's, it's almost to the level where you get that intuition, you know, when something is breaking, you know, these these individuals will know what the likely root cause will be, they will know how to fix it efficiently, they'll know how to stop it in the future. So, um, it's, uh, it's all of that knowledge that will be lost and you know, it's often called sort of tribal knowledge within organizations. It's that internal knowledge which builds up over years. Um, which is just, you know, hard earned hands on experience, which is very hard to, to capture and to replace. Yeah, and, and I guess also, I guess that point as well. So it's, I guess it also restricts collaboration as well, um, on, on that side of things. So it's it's actually a barrier to collaboration because again, if people don't have the knowledge and how are they supposed to share that, you know, across their organization, but I guess there's other barriers too to collaboration that we should, that we could raise as well. Yes, certainly. So it, it, it might be worth sort of talking through the, the types of knowledge we're dealing with. So, um, on one side you've got a certain, um, You know, type of knowledge which is documented, it's explicit, um, and, er, these are things, you know, on generally you have sort of manuals for your assets. So say you buy a Siemens drive, you'll get a manual with that, and that will talk you through the maintenance you're supposed to be doing. It will have troubleshooting and all of this information, the components, um, so that's, you know, one form of very explicit, uh, knowledge within, within an organization. But there's much more local knowledge as well, which is typically documented. So they, most organizations will have standard operating procedures, which are essentially like step by step guides, you know, when X happens, you do ABC, um, and this is local to the organization, normally local to a particular factory, sometimes even more local to like a line. Um, in some situations, but most manufacturers will have this information available, um, and I think the, the difficulty is more the accessibility of it. So say when you're, you're trying to fix something, you, you're trying to do some maintenance, um, you're probably under time pressure, especially if it's like a reactive situation. Um, are people going to be able to find them documents, uh, in the first place, and even if they do find a document, some of these documents are hundreds of pages, right, they're very intense. So you actually have to find the information within the document that you, which, you know, provides value. Um, so this sort of accessibility of that, that knowledge is, is problematic. And then on the other sort of main type of knowledge is more that sort of tribal knowledge. So it's that inbuilt, um, in a company knowledge that gets built up over time experience. It's understanding actually these types of assets fail in these sort of ways and actually, you know, you have to tweak this and turn that to fix it, you know, it's that the sort of knowledge that gets built up, um. And this is just hard knowledge to document, right? It's hard to sit someone down and say, what are your intuitions, write, write them down for me, er, it's, it's just difficult. Um, I think the closest manufacturers have a sort of maintenance systems. So these are databases where you log the work and maintenance which happens on an asset over time. So you do build up a history of, um, of what's happened to that asset. Um, and that's incredibly useful. The issues with maintenance systems is, firstly, they're notoriously, they're notorious for not being very, uh, high quality. So you know, sometimes records aren't captured or actually the information captured isn't detailed enough to be of use. Now that's not always all the time. I, I should, um, I should comment because actually a lot of our customers do keep good records. Um, but also then you've got another accessibility problem that you've got a mountain of records for, for the asset. You, you're under that time pressure, you want to get something fixed. Um, you know, it's hard to dig through that, that pile of records to find situations in the past which might give you some guidance. Um, and you know, in an ideal world, you wouldn't just be looking at that single asset history, you'd be looking at every asset which is similar in the, in the site or in the whole organization and picking out actually that, that incident two years ago in the site across America is very similar to what I'm having now. Maybe I can learn from that. So, you know, accessibility is. It is one of the key issues, um, and like any industry there's, there's silos between teams and plants and you know, for a lot of global companies, language barriers, there's shift work. Uh, you know, people are busy, very, very, very responsible for keeping the machine, the factory running, right, rather than writing down knowledge. So I don't know, it's a tricky problem. And like I say, having the time to actually read through manuals and find the exact part they need. So what, what we're hearing here is there's, there's these, all these pockets of knowledge, whether it's people, whether it's manuals, we could talk about a manual in a dusty cupboard somewhere or a manual uploaded into, you know, a file sharing page, but there's just all this knowledge. Dotted around, what you want to do is bring it into one place together, which leads us to perfect introduction to to knowledge builder, which we're introducing in Sensi predictive maintenance. And I guess let's just say, how does it work, Dan, I guess is a simple question. Yeah, so, um, yeah, Knowledge Builder is one of our sort of newest features related to, to knowledge, but obviously, you know, like I said, you know, we've been doing this for a couple of years now, so there's others which we can talk through in a bit, um. But KnowledgeBuilder is all around um capturing that explicit documented knowledge into the system. So users are able to take their manuals, their procedures, upload it into Sensei. um, Sensei then processes that information and makes it available er for us to use. And the way we use it in is um in, in two ways, uh, and it relates to the co-pilot we were discussing earlier. So when you're having an interactive discussion with the co-pilot, you can ask questions about what you're doing or the maintenance you want to perform, and then the co-pilot can refer to the documentation you've provided, uh, to provide, you know, the user guidance. Um, but as I said before, it's nice to not have to ask for this information. So alongside that, we're also, um, proactively looking at the documentation and, um, identifying, um, information which would be useful to bring into cases. So when we are providing that prescriptive guidance, we're now supplementing it with the, um, that explicit knowledge which is captured in that documentation. So you know, the step, the troubleshooting section of your manual, we're, you know, we're bringing that into, into cases or the procedures that you're supposed to do in, in response to certain situations, we can bring that into the case as well. So we are solving that sort of accessibility problem um by, by essentially serving up the information people need. Um, and we're also trying to address that, you know, the, the guidance side of things, so when you, when people do need more guidance, we can actually give them, um, that guidance upfront, sourced from, from their own documentation. I guess, uh, I guess the first question people would be interested to know is what, what types of results, I know we talked about manuals and, and different things, but are there sort of specific examples you could give which would provide the most value, you know, to enhance the, you know, the out the outputs. Yeah, so you, you can upload pretty much any, um, file, um, so you know, text files, um, PDFs, uh, office-based documents, audio video. Um, and that's housed within, um, Sensi for people to make use of. Now right now the co-pilot can only make use of, uh, text and PDF, um, files, but we're obviously going to broaden out, um, support for other, other files in the future. Um, and what's important to understand is actually when we're processing these files, we use, um, vision models, and that what that means is if you're uploading documents which have a lot of flow charts or diagrams or tables of information, we can, um, consume that information and make it available. Uh, and you know, er, many people might have er scanned er paper documents and we can make use of that as well through these vision um technologies. So you know, we can accept quite a broad range of documentation. And that obviously complements the knowledge we already have inside of Sensai. So, you know, the typical way we work is we collect um measurements from people's machines and we obviously track that for a very long time. Um, so we're sort of building up a history of that asset within Senseii. Um, and then we complement that with, you know, integrations into maintenance systems, so we're bringing in information about what work has been performed on assets. Um, and then we also collect our own information so people can leave comments in Sensai if they can provide feedback on, on cases, uh, we collect manual measurements. There's a whole, you know, load of information that we have, and that information is much closer to that more tribal knowledge. It's more of that sort of history of an asset over time, what, what fails and what do we typically do with that, with that asset. Um, so I, I think knowledge builder is really interesting because we're, we're taking what I would consider the more tribal and then we're adding in the, the explicit documented and that sort of combination of all of that in one place, um, being able to pull out, pull on that information as people need it and surface it when they need it. We're really, you know, tackling the core problems that lie at this sort of knowledge problem, uh, for customers. Yeah, and I guess actually specifically, so combining all this knowledge makes a lot of sense and it's continuously learning and improving from all these different sources of knowledge, which is fantastic. But the, the output of this is obviously prescriptive guide, you know, like suggestions on, you know, what, what they should do, what should they investigate, how does it Sort of fit, fit in with that. How does it produce that output, I guess. Yeah, so it's, um, it's very interesting. So, uh, when we, when we spot something in the data, um, so this might be an abnormality that we've seen a step or a spike or a sort of a trend, um, we'll raise that to a customer in a case. So the cases start off with describing what we've seen in the data. So we describe, um, the actual measurements which have changed. And then we, we take them patterns we've seen, we'll look back, um, in the asset's history, er, but not only the asset in question, it's any asset which is similar across the organization. So that might be across sites if you've got very good sort of uniformity of, of assets, and we'll try to identify cases or situations in the past which look similar to what we've seen now. Um, and when we find them situations in the past, we'll gather up all of the knowledge around them. So there might be comments from people, there might be work events, there might be feedback on cases, and we'll gather all of that knowledge together and then we'll from that we'll try to determine what the root causes of that problem might be. And obviously the history is a good guide to, to what that the issue may be. And now with Knowledge Builder, we're supplementing that with the explicit documentation. So as well as looking at that history, we'll bring in information from the manuals, from the procedures, er, to try to identify, you know, in, you know, the manual might say if you're, if you're suffering this situation, then this is what it might be and how to solve it, and we can obviously bring that into the case. er, likewise with the procedures, if we see, um, an incident that has a procedure associated with it, we can obviously bring you information about that procedure into the case as well. …