Driving Success in Automotive: A Global Manufacturer Story with Peter O'Hara and Jan Trnovec
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.
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, Gugan Sanga. Welcome to the trend detection podcast, powered by Siemens, and now your host, Gugan Sanga. In today's episode, we'll dive into how that automotive manufacturer has achieved its success, the technologies they've implemented, and what other manufacturers can learn from their journey. To help us dig deeper into this success story, I'm joined by some fantastic guests who have valuable insights into the processes and technologies that have driven this plant's success. Jan and Peter, thank you for joining me on today's episode. We're looking forward to hearing your insights on what has made the plant a stand out in the automotive industry. Please take a moment to introduce yourselves to our audience. Yeah, hello folks, Peter O'Hara. I am a customer success manager here at Siemens working with uh Sensai Predictive Maintenance. I've been with Sensai for just over 7 years, and my aim in life is to help our customers to use the application, add more users. Celebrate any wins, add more machines and slowly take over their entire facility, going site to site until everything the customer does is monitored by Sensor. Predictive maintenance. My name is Jan Tarnovitz. I'm a technician, a specialist in Ciprin, which is a daughter company of Siemens. We provide uh maintenance and engineering services to our customers, uh, mainly in automotive sector. Um. My work is mainly to reduce the downtimes, to reduce the price of the maintenance, and to uh achieve that our customer. can make as many products as he can. Perfect. Thank you both for your introductions and I'm really interested in this topic, so looking forward to hearing your insights into the automotive industry and you know, key successes around some of the asset types. Um, so what I wanted to do was to kick off this um podcast and ask um a first question around the asset type. So I understand the asset type is a high pressure pump injecting foam into car bodies for soundproofing. So if you could just provide a bit more background just for our audience so they can understand, that'd be great. Yeah, start off just because I can do that at the high level and Jan obviously knows his equipment way better than me. But it's an it's a really interesting use case for us because I believe this was the first time we've, uh, we've monitored a, a pump injecting foam into the car body for soundproofing. And for one area when we are talking to potentially new customers, one of their first questions is always, have you monitored this particular machine in this particular environment before? And I would always like to say, no, but that's not necessarily important, unless of course we have, because we monitor like many thousands of machines across many thousands um. Uh, well, not many thousands, hundreds of customers, and the main thing is we use the same algorithms to monitor data that means something to the condition of the machine. And those algorithms are self-learning, and so it's about feeding them with the correct data that will tell the application something about the condition of the machine and ideally removing the process that the machine is undergoing, and that can be a little bit tricky. But getting something about the condition of the machine. So in this case, we're looking at a process that has 3 stages to it. We have a pumping mode, a circulation mode, and then a pressurization stage. And for the circulation, we are monitoring uh flow rate, pressure, the RPM of the pump, and temperature. And then when we get to the pressurization stage, we take the time to get to half pressure and the time to get to full pressure. And one of the beauties of this application is that Jan chose the data himself and decided what to feed into Sensi predictive maintenance, and then very quickly within the deployment, we had this first success that we are talking about. So perhaps I should hand over to Jan to just tell me a few more things about this um injection pump. Yeah, as Peter already said, uh, there are like high pressure pumps which are feeding or pump pressurizing to medium to mediums which are mixed together in jets and then create the foam. Uh, I think the most important thing, uh, in this success was that we had the right data which described, uh, Which describes the machine, uh, the, the failure mode of the machine because we knew that the pumps go wrong, uh, and these breakdowns cost us or the customer some time to replace the pump during production time and so on. And so therefore, we looked at the pumps closely and found out that you can actually see the, the washing out of the pump in the way how the pump is building the pressure. So in the, pressurization stage. And In this press pressurization stage, uh, where the circulation valve closed and the pump starts to build pressure. Uh, we can see that if the pressure isn't rising linearly, but slows down as the pressure rises, then we know that the pump starts to wash out and which, which is then followed by a breakdown. Uh, so we simply started to monitor via Sensi predictive maintenance. The time, as Peter said, needed to reach half of the set point and time needed to go from from half of the set point to the set point itself. Uh, these two times we called half time and full time. And we programmed all the logic inside, uh, in the PLC. To, to get these two times regularly for each pressurization stage, uh, then we connect it to the sensei. And voila. After about 2 months, we get a warning from Senzai. Uh, that the, the time is changed. The, the, the, the, uh, full time was changed on one pump and we could see that, uh, really the, the curvature of the, of the rising of the pressure was changed. And we could see that the pump. Is going to be wrong, like it's, it's, it's going to break down. Uh, what we didn't, we didn't, we didn't know that, uh, how much time we actually have. Uh, so we saw that the bump is changing, uh, but we didn't know how, how long does it take. To how much time we, we have to the actual breakdown. Uh, so in one case, we actually leave it. And uh then we also knew the exact time when the pump will go wrong, and when the, when the pump will break down. Uh, and now we like 100% know or 99% know where is the limit we can go, uh, to, to, to use the pump as much as, as possible. And uh that's actually interesting because uh we get a warning, uh, like a month and a half before the actual breakdown. That's quite nice. Like you have a month to plan the, the, the, the, the replacing, the place of the pump. That was really, really nice. And then, so you, you didn't add any more sensors to, uh, to the pumps. This is just harvesting data that exists. Yes, of course, yeah, yeah, that was also really nice that we didn't have to. Mm, change something on the machine to add sensors and so on. We just used all the data we already had. We used the data from frequency converters, about the speed of the pump. We used, uh, the pressures which are being monitored, uh, also for the technology purposes of the machine, for the regulation and so on. Uh, so we don't have to add anything. We just collect it or, or, um, summarize the data which we already had. In the BLC and then we write uh the files in the visualization and then send it. To Sensi. And so what process did you go through to go from, obviously you had some data, but then you'd have to look at it at fairly high resolution to choose what to send to Sensi, and then, so how did you go through that process? Yeah, we simply used an an PLC analyzer, uh, to see in high resolution the data. Uh, where we, where we looked at, at the behaving schemes of the, of the machine, how the, how the pressure rises, how, how it slow down, uh, how, how the system works simply. Uh, we tried to, to, uh, understand what's going under the skin of the machine. And then, uh, we found out that you can simply, uh, We, we, we, we. We managed to find out the the simplest way how to get the important data because that's, I think from my point of view it's the, it's the most important thing to feed the right data because if you don't have, if you have only some data, the sensei uh is a perfect tool, but it, it isn't some, some wizard inside with a, with a glass. Uh, sphere telling you now it will break. It calculates, yeah. And we often try to guide our, our customers through a process of having the correct people in the room at the early stage of a deployment, so they know maybe somebody who has uh ITOT type experience who is able to get the data, but also will have a maintainer there who will be able to say. what failure modes are important to the customers and also some management to decide which machines are important, etc. It sounds to me that we have the perfect person on the project because you can analyze the data, you have the skills to interact with the PLCs, you have the skills, the analytic skills to look at the data to decide what is important for these failure modes, and then to pull that out. So this was a Beautiful deployment, but if, uh, if we could, um, scale yams and put them in every factory, I think we would have a much easier life. Yeah, but it wasn't just me. Like I've had a, I had a colleagues that helped me, of course. I had in particular one colleague, uh, a technician, which uh knew the machine really, really well, and he could, he was able to tell me some insight, uh, what Where can I look? And then when I looked, I found it. Yeah, that's, that's the always the thing that, um, you need colleagues, you need, you need the experience, uh, of probably more years, uh, you need also the experience of, of the team behind Sensai, of course, uh, which you can tell, uh, who, who can tell you what is the Sensai able to do, how you can monitor this and this. And also Um, as you mentioned, you didn't have any experience with, uh, for example, foaming machine or, uh, this type of machinery, but, uh, when I reached to you, um, uh, with, with, with particular, uh, types of breakdowns and what type of data I have, uh, you helped me on the, on the other side. OK, we can extract it in this and this way so we can monitor it back then and process it and at the sensi site. So that's always, um, it's not a one person. Uh, I will agree it's not one person, but it's the skills you need to make a successful project involve knowing the, the failure modes that you want to capture, but then. Having the ability to get the data. Now sometimes that is pulling it from a PLC, sometimes it's, I need to add an additional sensor here and having that bit of experience as well, but yeah, it's um it's nice to see. Uh, a Siemens colleague, you know, fully embracing the, the process of getting good condition monitoring data into the application and then letting the application kind of run with it, and then I think with early day success in a project, we always need a couple of pieces of luck, so one is, you know, an early. Machine failure, obviously, you never want the customers to have machine failures, but when you look at. The likelihood of a project succeeding, you need to have machines, uh, enough of them, so the chances of one of them failing early on, uh, within the first couple of months, um, that certainly helps, because if everything runs as normal for quite a while, people are like, yeah, well, why are we monitoring this? And then as soon as things start to happen, then you can. Uh, help people to see that actually, yeah, this is, this is good, this could be successful, and then you get them to think, well, where else do we have a problem? Where else can I harvest data or where can I put some sensors, and then pull that equipment into the application as well. Yeah, exactly. Like it doesn't make sense to, to, to monitor a perfectly well and working uh machine which doesn't have any breakdowns. Like you can monitor it for years and nothing happens. Like that's, that's the also, that's also an important thing to monitor things which do break down and uh which are also important because. Uh, every, like in every facility, in every, every plant, there are machines which aren't important, and they do break down by, but you just Um, somehow no. You do not care. Like they are not important. And then there are machines which are really important because they stop the line, for example. Uh, and these are the important ones. These are the bottlenecks which you have to analyze also from the management, uh, management, uh, uh, level. Uh, and these need to inspect, needs to be inspected properly and if possible, then. Um, also monitored via. There are possible solutions. Of sensor which is one of them and um from my point of view one of the best ones. Um, that was really useful information just to understand background around this asset type. And I just wanted to move into the next question just around, um, you know, what role does the foam injection pump play in the automotive process. Yeah, um, it's used for dampening noise, noises or noise, noise noise cancellation simply. So the foaming is injected uh in uh pillars of the vehicle. Uh, for example, into A pillar or B pillar in the pillars above the, of the, uh, of the axels and so on. To cancel noise. So, and, um, one has to, has to mention also that not every vehicle does have the pillars formed. Um It's not like the standard feature. Uh, so that's also the, that's also possible, the possibly the reason why, uh, since I hadn't yet. Uh, the, the experience with such, such a machinery because it's not absolutely common machine in automotive, uh, world. But, uh, some, uh, customers use it. Uh, and, uh, what I see, uh, it, it's used more and more often. Exactly, because of noise cancellation, it's quite a. Good way to for for noise counseling. Uh, and as the technology evolves, Uh. It will be probably used more and more. That sounds good. So, One of the questions. We had was. How did sensor predictive maintenance, uh, detect the trend and how reliable are the alerts. So, what I would say, I think I'm probably repeating myself here, but we use a fleet of algorithms that learn the normal behavior for a machine, and I think that this is. Different to a lot of other products out there, so we don't have people behind the scenes making models for all the different machine types that we, that we monitor, so. The early days learning is all unsupervised, and what I found interesting in this case is that the data for, I was looking at this, yeah, yesterday or the day before, you know, the data for this machine, it stayed stable for about 4 days, and then it started to trend up, and then as you said, since I alerted the user. Uh, after, uh, about a month that this data was trending up, but I, I thought that was quite interesting. People often ask me, you know, how long is the learning process, and the short answer is it depends on the type of. Algorithm alert that we're using, but usually we quote about 48 hours, and I found it interesting that, like I say, with this, I could see the data was nice and stable for the 1st 4 days and then just slowly started to trend up, so I, I thought that was quite interesting. And then one of the other ones, sorry Gag don't know if I'm uh stealing your thunder with this one, is the role of the co-pilot, so we haven't really mentioned that yet, but with the use of co-pilot in industrial applications now. Uh, the co-pilot did spot this failure mode. Uh, I wondered what you thought of that. Yeah, um, thanks to the expertise we have done in the first stage, the copilot, copilot actually didn't tell me, told me anything which I Uh, haven't already knew, but the thing is that, uh, after replacing the pump and so on, we, uh, insert the, the input of the, the, the all the necessary data into Sensi, and then on the next case, the co-pilot already. Uh, pointed out that uh it's probably a uh, a pump which has to be replaced, which was also the right answer. Uh, so whenever the the the system will be used by someone who doesn't have as much experience as, for example, I have, the co-pilot can direct him what to do, and I think that's a really, really, really good thing that it can use the information which is in put it in and to show it to the next person. Who Doesn't have to, doesn't have to have uh such deep level of the knowledge of the machine. Yeah, I, I agree with that, Jan. I just wanted to highlight, I think that's one of the key benefits with the co-pilot. Um, you know, you're able, it's designed to take that knowledge and you don't necessarily need to have a high level, um, you know, amount of knowledge to understand. So it helps to support and just break that information down. So, um, sorry, I just wanted to jump in. Yeah, Especially when we're scaling, this is the thing, and in, so I've been in this industry for 7 years with uh set aside predictive maintenance, but. You, early days of the project, you are working with knowledgeable people about their machines, and then you go to scale, and then sometimes you go to a different plant or a different site and. That knowledge can be hard to replicate, and so the fact that the, the co-pilot will tell you exactly what's going on, but also, you have the ability to ask it the, the silly questions because we have seen in other projects in the past where the project gets moved on to people who are perhaps not as data literate and perhaps don't understand the, uh, what this data means for their machines as much as the original. Uh, people on the project, so it's really nice that you can, uh, you can interrogate it, you can ask it questions, as I often say, it takes my role, customer success management, it takes me a while to build rapport with people, and I know I've cracked it when they turned around to me and say, I don't understand what this means, but people will ask the co-pilot that sort of question way sooner because people don't mind. Um, you know, asking the co-pilot the questions that they really want to ask, and sometimes they just say, oh, I was just testing it to see if it knows, and it's like, yay, that's, that's what it's there for, just ask it away. What does this mean? Why is it doing that? I think that's a really beautiful step forward, and it's great to see it out in industry. And, yeah, nice and quick in this case as well. Yeah, I was going to say as well, Peter, just touching on that point, I think we had a discussion maybe a couple of weeks ago, but you know, around the co-pilot, people generally seem to be more open to asking questions. And like I said, it's so intusive. Like it can give you that information without talking to Um, you know, a human as such, if that makes sense. If you need the answers there and then, it's very quick. So, um, definitely helps with efficiency. And as Jan said, you know, he closed the case and then the next time the co-pilot sees something, it's feeding back the words that Jan has used to the users, and so you can. You know, you spread that knowledge throughout the organization because the co-pilot will tell other people and uh it's, yeah, it's a very powerful tool for um upskilling the, the users and the workforce and sharing knowledge. So moving on to the next question, I'm really interested to understand how critical was the 40 day early warning in avoiding downtime. Um, normally, we need like, mm, only days to warn us because, uh, when we get a warning during the week, we Prob, we always are able to, to manage the, the replace or the, the repair uh during the weekend when the line stops. Uh, the thing is that if you have such an, an, a long, long time period, you are, you, you are able to, to manage it, uh, not only that you repair it, uh, as the machine, but you get into the stage that you can manage how many stocks. Do you have because all the things you need to replace or or the or the spare parts you, you want to. …