Trend Detection Revisited: AI-based Predictive Maintenance from Factory Floor to Cloud

Trend Detection Podcast ·

Show notes

Welcome to the Trend Detection podcast, brought to you by Senseye Predictive Maintenance – which gives you visibility and insights into all your assets, from single machines to full plants to help you reduce downtime, increase knowledge sharing and accelerate digital transformation across your organization.

From the Trend Detection archive - a factory-floor perspective on AI-based predictive maintenance.

Tobias, Head of Maintenance and Improvement at Siemens, shares the predictive maintenance journey of Siemens’ highly automated factory in Bavaria.

The discussion explores how smart hardware, OT modernisation and AI-driven analytics were brought together in a live brownfield production environment.

It also examines how Senseye Predictive Maintenance helped maintenance teams focus on critical assets, act on emerging equipment issues and reduce reactive firefighting.

This episode was recorded previously, so references to the project and product reflect the position at the time of recording.

Transcript

Welcome to the trend detection podcast powered by Siemens, and now your host. Just to get things straight, Pablo, you're a product owner here at Siemens. Anya, you are the head of business line digital services, and Tobias, you are head of maintenance and improvement. And what we're gonna be doing in this session is we're gonna be talking a lot about one of our Siemens locations, and that is in calm. And um Tobias, maybe we can start with you. Maybe you can explain, maybe first of all, where Calm actually is for our international guests and what we actually do there. So, uh, our location in Calm is in the east of Bavaria, um, and is a part of our factory network within Siemens Electrical Products. And uh with a high degree of automation, uh, we are producing several products of our low voltage portfolio there, um, like contractors, motor starter protectors, overload relays, and so on, um, and round about 470 employees are working there, uh, spread across two production halls and around about 16,000 square meters. So it's, um, more or less very big factory. Uh, and belongs to our factory in Omberg. OK, cool. So, When we talk about maintenance then, in this particular factory, what is a challenge or some challenges that we are facing there? So when we talk about maintenance in our factory. And the maintenance of our manufacturing equipment. Um, and our major challenge, we have to talk about unplanned downtimes for our production machines, so I'll tell you why. Um, first and foremost, unplanned downtimes mean a loss of output, so you Don't get the output anymore. Uh, and secondly, and more or less a consequence of the first point, you suffer a loss of productivity and last but not least, maintenance costs are increasing. So I believe that many of our guests here in front of the stage in the booth or online on the live stream are familiar with the consequences of unplanned downtimes. Most obvious. Um, the employees, mainly in the production area, cannot do their work anymore when it comes to unplanned downtimes. So firefighting starts, internal discussions arise regarding the lack of productivity. Countermeasures have to be derived. So that's the internal view, but there's an external view too. Customers will be unhappy when they're not getting their products anymore. Uh, to the right time because of the machine broke down, for example, so a loss of image, a loss of trust may come up, uh, on the external view. So our conclusion was as follows unplanned downtimes lead to 3 to 5 times higher costs compared to planned downtimes. And if you put yourself in that situation I just described and imagine in addition. Uh, you're facing a lack of transparency on the health and the condition of your assets in the production machine. Your machines could break down at any moment, but you have no idea when or how. So we didn't feel comfortable with that situation, and we decided to act. OK, yeah, so a lot, a lot of consequences, negative ones, um, so yeah, what did you do? So as I said, we didn't feel comfortable, uh, with that situation, and we decided to implement a digitalization approach. Uh, that combines the real world on the one hand and the digital world on the other hand. So what did we do? Uh, we integrated smart hardware on the one hand, so the real world, uh, to make use of the data our machines produce, uh, and we implemented an AI-driven analysis approach. On the other hand, so. To play on the digital world, so, so that we can predict failures and downtimes before they really happen, OK. Now, Pablo, yeah, obviously digitalization, whenever we hear it, we think software immediately, but there's also hardware involved, right? What can you tell us, um, 11 of the things that you have to think that is that digitalization is not just about software, it's also migration or changing, adapting our OT systems, our automation systems. You, you have to think that digitalization starts in the shop floor. The shop floor has all the information of the real asset, the conveyor lanes, the pumps, the mixers, the crushing machines, all this information, these devices provide you information and The idea is to catch this information for you being used in a higher level. Without this information, it doesn't matter how smart your software is. The software is blind, and the idea behind it is the same like if you drive a car. Imagine that you would drive a car without cockpits. It doesn't matter how good are you for driving. You don't know the speed. You don't know if it's missing oil, if it is the revolution that the car is going or the motor is going. So all you need is all this information. For systems and this is uh where where they they start to to change our OT systems to catch this information and with the right hardware our applications became storytellers of what's happening on the real application. And bridging this between the OT to the IT level to flow this data up, OK? Now, I mentioned in the intro that we're gonna be talking about SIRIUS as well, so our SIRIUS product family. So, can you tell us which part of the product family is actually involved in this digitalization process? Sure, um, in CDU for years, for, for decades, SIRIUS was about controlling and protecting, yeah? Um, but it was part of our DNA, but The world is changing. The digitalization is making a tremendous change on the industry, and this change is generating also that we have to change our portfolio. Our portfolio is not anymore only about control and protecting for sure. We are still doing that, but it's about also. Gathering data, so the data flow, how this data can help to a higher level. So SIRIUS became a central part of the digitalization, not maybe because of the software side, but because of the migration of the Or Tite and the shop floor information that we can capture with our portfolio. OK. Now you mentioned bringing data up, so, so to the IT level, which, which part of the SIRIUS family now actually does that? Well, we have a, a complete portfolio like I said before, uh, actually it's in the screen, but we go from the standard contractors and circuit breakers. You can make them intelligent with a 3 year system intelligently module and connect to better ET 200 speed as well as the e-starter. Eric Froli was some minutes ago here and he showed a wonderful product. They start a full electronic starter also provided the same information like the 337 ILM soft starter, the well known SIMMO code in the process industry, very well known, and also we have full monitoring relays, so it is not about only motors. It doesn't matter which is the variable that you want to monitor in your factory in your application. We have the full portfolio for that. OK. So Anya, what would you say? How does all this data then actually help us? I mean it's great to have it, but what do you do with it? Well, so once the data is collected, it's being sent to either local or cloud systems for actual analytics, and the Siemens software turns the data into actionable insights which help the maintenance technicians. So the maintenance technicians use this information to optimize their maintenance tasks. OK, so which software is actually used then for this task? In this task we use sensor predictive maintenance, which is an AI based cloud solution. It learns from historical and real-time condition data to predict failures. We apply AI here in two disciplines. So first of all, the classic analytics capabilities that, um, yeah, monitor a time series and you can detect anomalies, trends, threshold violations, patterns, things like that. Um, and then the generative AI takes it to the next level by giving the numbers meaning, textual meaning, really describing what is going on, why is the case raised. It can suggest the reason and explanations, a root cause, and even a recommendation for the remedy. Mhm. So let's say I'm a user or a potential user of this. How can I now envision my interaction with it? How does it work? So the user is typically in the maintenance department and they can access the app via a standard browser either on a PC or on a tablet like this. OK, so the user doesn't have to be close to the machine to know about their condition, and that's the charm. All relevant assets, all onboarded assets can be um checked out and the condition is clear at all times and in the case you hear um it is highlighted where are assets that actually need attention, so the maintenance technician can go into the details of the case, check it out, see the evidence, which kind of data shows that there's something going on. And it's even possible to open a chat window to the co-pilot and chat, ask questions like in a conversation. And I'm inviting everyone here to check out the co-pilot functions over here in our industrial AI hub. You can talk and find out what's the problem and really get a step by step guidance through the problem that, that there is for the maintenance to solve. OK. And that, by the way, is why I and Christine keep reminding you to go and check it out, yeah? This is exactly the reason we want you to really experience it live. Tobias, let's go a bit deeper then into the calm factory use case. So what exactly did you implement there then? So we applied the solution that Pablo and Anya just presented to a highly automated production. Production line in our pre-production of the motor starter protector 3RB 2 in calm and uh the implementation uh took place during ongoing operations, so like a brownfield approach and was done in 3 steps. Um, first of all, we upgraded our cabinets by integrating the relevant, uh, serious products, the smart hardware, to make use of the data our machines produce. Uh, secondly, we expanded our programming standard and we integrated data preprocessing on the PLC, uh, uh, like a preprocessing routine, uh, to secure the data transfer from the shop floor, from the. Uh, manufacturing line to the cloud and, uh, finally we onboarded Sensi predictive maintenance, the AI-driven solution that Anya just presented, and now we are able to gain insights into more than 90 connected assets, um, in the production line. Uh, especially the critical motor driven components, um, that we try to have a look on the health and the condition, how they are working, and what stuff is there, uh, with the components. OK, Pablo, uh, Tobias just mentioned, the, the SIRIUS product family again. So can you tell us a bit more which hardware was used then in Calm? Well, in CM in particular, we use the 3RS. intelligent link module. This is a device you can find directly here on stand, but it's a device that goes over the standard contact or standard circuit breaker. It connects with both and easily you can start having not only control and monitoring and protection of the motors, but all the information, the data, that real-time data that it needs for the sensile level. Um, with, with this, this hardware makes very easy, the retrofitting of the machine because you don't have to change even the contractors. You can just snap in on these devices. And also we use the e-starter that was presented also some minutes ago. Also this starter was there and the beauty of this is that both devices, also the soft starter, the data structure of these devices is exactly the same. Therefore, if you use for one case the 3R 7 ILM and the next time you want to use the starter, the data structure, how you might handle that data is exactly the same. OK. Now, in order for this to work properly for everyone and everyone to be able to work efficiently with it. It needs to be connected to the entire system, right? So how does that work? Well, that's the beauty of our system. Our system is fully integrated to cymatics environment or to Tia portal environment. So therefore the integration is quick. Very quick. You, you don't need extra software. You just use the step 7 in the E-starter or the 37 ILM. You drag and drop and the communication, the diagnostic, and everything is automatically connected with our ecosystem, right? And just back to the devices again briefly. So obviously they're, they're measuring a lot of stuff to get the data, but what are the values that they actually measure? Well, our device measures the kick values of the motors, right? Currents, asymmetries, grounding currents, starting times, voltage. power factor, energy, all the, all the recurrent, how many cycles you have used the device, how many hours has been running the motor or also the contractors or the the starter, and with all this data you can collect all the crucial information that you need for the shop floor for the sens site level. OK, and, and Anya, we said before that the data goes up, yeah, to the IT level. So in this particular case then where does the data go from, from Pablo's devices? So the data goes from the shop floor to the top floor we like to say, and our industrial communication protocols make sure that the transfer is reliable and secure. And we use typical industrial communication protocols like ProfyNet, Ethernet IP or Modbus TCP, and then when it gets to the cloud, it's MQTT the typical way. And we perform the required analytics to turn the data into strategic information. So what does that mean? Data having motor temperature of 90 degrees. Well, what is this? It's data, but the context of it makes it valuable information. So if a motor is running slowly. This kind of temperature is, is OK, but if it's overheating and it's outside of the range, this is the point where the maintenance technicians need to be alerted, need to be informed, and this is what, what's, yeah, the anomaly detection here. Find the context of the of the usage of the motor with the actual condition data like temperature. And driving the attention to the asset when it's necessary is ultimately the goal of a predictive maintenance solution. And implementing or setting up these solutions, I mean, it sounds quite complex. So was there anything in particular you needed to take into consideration? How was the setup? Well, it's, it's a typical same way. It always starts with the critical asset that requires the monitoring. So once this is defined with the customer together, um, it's also necessary to define the necessary KPIs to measure the success, what needs to be improved or what needs to be reduced. This is the starting point to then determine, well, these assets, which kind of conditions do they have, which are the condition indicators that help understand it, and with this point you know which data points you need and then the next step is to figure out where to find them. Either they're already available in the PLC or sometimes additional sensors need to be installed, and this then determines the way the data needs to take end to end ultimately reaching the cloud where the analytics take place. OK, Tobias, I believe that this. Is all still in its infancy, um, so, so tell us, when did you actually start with this and which results have you already been seeing? Well, the go live was 6 months ago and we have already seen the first success cases which have all one thing in common. Um, by predicting failures, we were able to maintain the critical assets before they really broke down. So let me tell you one example. Um, in one example we saw anomalies, uh, in torque and the current consumption of a linear direct drive, which would have led to a production downtime of several days maybe, uh, and the corresponding financial losses that I mentioned a few minutes ago and just remember the, uh, all the consequences. Yeah, um, I, I told you a few minutes ago, unhappy customers, internal discussions, internal troubleshooting, and so on, all the productivity discussions. So our new approach by combining the real and the digital world prevented all these negative effects, uh, and consequences in more, um, more or less, uh, 5 to 6 examples that we had. Uh, in the last 6 months. OK, now with that in mind, what would you say are, are the key benefits then of combining Siemens hardware and Siemens software? Well, in my opinion, there are 22 big benefits. So on the one hand, the transparency on the health and the condition of the different assets in the production machines is much more higher than before. Only by using the data the machine produces every second. OK. And on the other hand, and that's the part of, of the AI solution, making use of the data leads to insights on the hourly, uh, in a, in a minutely frequent, um, insights on the, on the assets, on their condition. And predictive maintenance action actions can be derived before the machine breaks down. So in combination we are now able to avoid downtimes and drive maintenance efficiency and all these benefits that I told you, uh, were implemented through reasonable changes to the existing software and hardware environment in a brownfield approach. Yeah. So you launched it 6 months ago. What happens in the next 6 months and beyond, yeah, based on the positive experience we made with our pilot project, uh, we are now planning the next steps. So the further rollout is prepared in our factory in Ombeg and Calm, and we will onboard the next use cases. Um, but on the other hand, other locations have recognized the benefits too and are showing interest now. So, um, the knowledge transfer is on the agenda both internally to other production units in our factory in Ombeg and Calm and to the other Siemens locations, uh, which are showing interest. And finally, Uh, we will close the loop towards new production machines, so we will integrate both the hardware and the software in the very beginning of the life cycle of our manufacturing equipment, OK. Anya, could you maybe summarize some of the key takeaways of this sort of experience, if we can call it, in particular for our customers? Yeah, so we showed here that the Siemens factory is actually using the products it produces, and the Siemens products, software and hardware together jointly proved their value in in the setup. And the good news is we're not just keeping it to ourselves, we're actually enabling you as our customers to use exactly this technology. So you may benefit just as much in saving maintenance costs and avoiding unplanned downtimes, right? So, and it's not only software and hardware we provide, we also support in services. So our service experts guide you step by step through through the setup of the concept and then also implementing it. So we have typically um standardized blueprints of the typical scenarios that it's very quick to give a first um overall estimation how long that will take, how much it will cost to set it up, and then also enable it quickly to, to provide the value to the customer. So I'm inviting all of you to come by the sensor predictive maintenance uh model, have a chat with us about your specific uh situation, about your requirements and your operations and your production, and we're looking forward to finding the proper solution for you. Yeah, I think we even have it up here. Yeah, exactly. So, Pablo, where else can we find more info? Are you, are you everywhere? So yeah, we actually took control of the booth. So you can find sensor plate maintenance in the 85. You can find electrification, so the e-starter. Also the SIMIMOC code and the 337 intelligent link module here directly on the number 8 and in the middle where we crash together you can see the complete solution with SIRIUS and Sensi in the AI AI hub exactly and the final question, Tobias. I'm hoping everyone stays until the end of SPS, so until Thursday, and if they've got time on Friday, how far is it to get to Cannes from here? How, how long would you need? Let's say driving. Well, driving by car, I think, um, maybe 2 hours. 2 hours. So that's doable, right? So make sure to go and visit these guys in Cannes and then also in Hamburg, and then next time round we'll see where all your other solutions are. So thank you so much for being here today. …

Full transcript (WebVTT)