Name: Tips to Laser Welding Success
Uploaded: 2025-01-23T17:36:44.313Z
Duration: 46 min 44 s
Description: Tips to Laser Welding Success

Transcript for "Tips to Laser Welding Success": Hello, and welcome to this webinar on laser welding. I'm Jeff Shannon, director of marketing. And joining me today is doctor Ronald Meyerhoffer, product line manager for laser systems. Just a bit of housekeeping to start before we get into the main event. Presentation will last about 25 minutes. You're really encouraged to ask questions. You know, we like to be a interactive session here, so please use the q and a tab, to do that. That means we can answer the question perhaps at the end of the session. If we're unable to answer them at the end of the session, we will answer them after the session. So for those people, this is being recorded. It's also being streamed on Goldcast LinkedIn, and YouTube. And for those who are gonna attend, MDNM West, which is coming up in a couple of weeks, feel free to swing by our booth. Doctor Myhawk will be there and can answer any further questions you may have. Just one other item, we will be having a poll, towards the end of the presentation. Only the GoldPASS audience will be able to participate. However, both LinkedIn and YouTube will be able to see the result. Okay. So that's all the housekeeping. So let's start the presentation. So, Roland, would you like to bring up slides, please? Of course. And the next one? Yeah. So you should see it now. Right? Yeah. I see it. Yeah. So it's the this is this is the outline. You know, we're talking about the the so the the ways to get success in laser welding, you know, both the high power and low power. Obviously, you know, there's there's a number of steps. We start with material selection all through to the process. What we're doing today is looking at the process side of it. So and what's happening in the area, particularly recently, which is really gonna help, provide, a better idea of what's going on and ensure your your yield goes up in your well quality. So we're gonna cover vision. We're gonna cover in process monitoring, which is a pretty hot topic right now. And, also, the question is, you have all these data, so what are you gonna do with it? And so there's various ways we can manage this data and and make the access to to you. Okay. Next one, So, you know, how how do we get here? You know, what's been happening over the last few years which is really enabling what we're doing here? So, you know, everybody, I'm sure, is out of industry 4.0. This is basically an intersection of, you know, cheap computing power, all that GPU stuff, you know, artificial intelligence, machine learning, and sensor technology. You know, essentially creating all this data, the artificial intelligence is making some sense of it, and then and then the cheap computing essentially powering everything. And what it does is it provides us with a sort of new kind of regime in our in our systems, you know, which is really being developed through the software and the software architecture. So we've got the ability to look at the the system information. We can look at the system control. And, obviously, you know, it's enabling things like in process monitoring and and vision and things of that nature, which, you know, if you look at the future benefits of some of these items, which, obviously, Roland's gonna talk them in the future, there's there's a lot of upside for you from a from a production and manufacturing perspective. Essentially, it gives you the reliable production you're looking for. It gives you, manufacturing data, which is also super useful. And, also, it keeps, keeps the the process optimized if you have a variety of machines in the same path. And so I'll I'll, step aside, and I'll hand over to you, Roland. Thank you very much. Good morning also from my side. So just a little introduction also here what we are talking about. Of course, we all know that, laser manufacturers or system manufacturers try to build the best system, as possible, but there's also always something that can happen. For example, if you have a quite dirty process, your optics could damage, it could be contaminated. Sometimes the part fit up is not okay. There are too many tolerances in there. It could also be a wrong positioning in the machine, from the operator error, focus position check could fail, and alignment could also be wrong. Therefore, the goal, of course, of this whole thing is to really avoid making scrap. Of course, next thing is not to ship any scrap because this is really expensive at the end of the day. And so we try to, prevent that with different methods and also collect a lot of data for comprehensive documentation afterwards, and of course, to, also use those technologies to minimize downtime and to reduce or eliminate inspection needs after the process is done. What I'm going to use throughout that presentation here is this timeline where we show basically on the left hand side, this is the, time when the part is actually presented in the machine and, going to be ready to be processed. This middle section is the, actual process that's ongoing, that time when the welding, for example, happens. Then there's a time, of course, for post process inspection after the weld, right, immediately after the weld in order to get an information if it's okay or not okay. Typically, then some datas are sent out or collected in a report in order to also show the traceability from the front end to the back end, and collect all the data and document it. Then for a longer time frame, basically like a longer interval, there are regular system checks also done like power measurement also. All those data at the end of the day somehow needs to be stored in order to have a further analysis on it. How can vision process monitoring that Jeff talked about already, help us in the different stages? For example, in the pre process control area where the part is just placed, we use a lot of vision in order to see where it is. During the process, in process monitoring technologies can be applied to collect data from the process while it's happening. When it's done, you can again apply vision processes to see, okay, was it okay, was it not okay? And combinations of those of those, of course, are also possible. And then on the longer timescale, like a longer interval, okay, you can do, your data management, information, collecting it and displaying it, for example, somewhere, on premise or somewhere else delocalized. First of all, when looking into the vision aspects of it for the pre process control as well as the post process control, For example, part position presence check, you can do a focus correction also using vision, and of course, the visual process quality inspection after the, where all this done is also an option. Just checking, okay, what are we doing with our software? Because software is the key to all of that because it is our goal to make it as easy as possible for you to set up your complete workflow. This involves vision, process monitoring, inspection, and data management, of course. You need a powerful software package to really provide that in an easy and intuitive way. That's why we have developed the laser framework software environment. The camera based vision and pattern recognition inspection can adapt, of course, the laser object needs to be lasered to the actual position of the part. So typically, we have got a camera cube in between the laser and the, galvo process in this case, and we can look through the camera. We can, select the places where we wanna look for an edge, for example. Then, of course, we've got the live image to see, okay, what is the pattern, what is the gradient that we are looking for, and go from there. I just wanted to show you that in a little video here that we took for welding, in this case, we are welding 3 pins into holes and basically are welding from the left hand side here. I'm going to go through that. We've got the 3 positions here and we'll monitor that as well using the Exactware 230 platform for that. This is showing the interior. We've got the fiber coupling in there, you've got the process monitoring in there, and the camera down here looking through the galval onto your part. Next step, of course, is setting up the recipe in our laser framework software, typically by drag and drop of elements from the right hand side. In this case, we just want to have a vision process before the laser process happens. So we jump into the program. We apply, in this case, a one point interactive search. And we don't have to do it on the whole thing. This is a metric, so we just have to apply it on the master cell. So we have to do it only once. Also, we do 3 wells in this case. So we're guiding that operator through the process. In this case, of course, you can also filter out shorter lines because we just wanna see that hole and look for that edge shape, and then we can teach it. Now it's taught, and then we can start processing our parts here. We can add other steps like adding a live image for the operator or also result display of the metrics later on for the operator to see if everything went worked out well. The dry run without laser, just a quick thing to check for the operator, is everything in line? We just could run it without laser action to see if everything is aligned. Then, of course, we execute it. This is the operator view again. There's welding action, of course, going on and you can see it also through the processing window and just see how that welding happened. Of course, the operator then can check the results after completion. He's going to have a look at the result of the matrix display. It's all green, so no problem there. You can check the stored images of the vision system, and that's basically the workflow we are providing with the laser framework software. Then we are on the other side, so the process is done. After the weld, there are camera based technologies that we can use to also characterize the welding. Was it okay based on a vision topic? These are these camera artificial intelligence machine learning algorithms that you can train certain photographs or basically images. You can say, okay, this is a good well, that's how it looks like. For example, if you just switch off the shielding gas, you really have a burnt and contaminated surface and you can train your vision system towards that and once it shows up, will give you a certain probability, of course, for that error. And, of course, also for WELD interruption, this can be trained. So there are possibilities, of course. The image should fit into the TTL image to have the the whole weld covered. You are not able to see any weld defects underneath the weld. Some things can be covered with these vision based systems, but of course, not all of them. If you look into the in process monitoring, it will tell us also something, but immediately after the process is done. You get a result immediately after the welding process. What we are using here is a diode based monitoring. We can also combine that with acoustic monitoring, for example, just to attach a sensor, a piezo sensor to the part. We can look for the body noise out of the welding process. What we are using, have developed over the last couple of years is called SmartSense Plus, our dial based in process monitoring tool that is going to be attached basically straight above the welding areas or all the back reflected light is going to be fed into this fiber. It goes into a detector box and there we split it up into 3 different areas. We've got the visible wavelength range, we've got the laser wavelength range, and we've got a little relative temperature range between 1217 nanometers. As you can see, they're all totally different depending on the process. You can see that the plasma is rising steadily, whereas the laser wavelengths actually is a pulse shaped down. And of course, for the temperature, it takes a while to really come to a certain point, and then it's steadily growing. So those are distinct three distinct, photodiodes that we are using in here. And we can, as I said, combine that with acoustic monitoring. The goal is to have a immediate feedback basically after the process. Was it okay or not okay? What we are using for that are 2 different stages we can offer. Basically, it's the pathway stage, is the standard one. You're collecting a number of samples, and you define an upper value and a lower value. And once the white one is in between, you, label it as good or the system is labeling it as as a good, process. If it's violating a lot, of course, it will come back as a bad process. But, of course, we can also have a second stage where we train certain failure conditions, like out of focus welding, like, a big gap. And we can train the signals to that, and, then the algorithms basically are telling you the probability for a certain failure. Condition, nice for an analysis why something went wrong. SmartSense, for us, it's our own development. Therefore, we really have the highest flexibility. We can combine it, for example, to a 10 watt picosecond laser, in this case, attached to, this last beam bending mirror. We can combine it with a 20 watt, UV nanosecond laser, again, also in the beam path here. And of course, we can go with 3 magnitudes of power higher, and, combine it with a 10 kilowatt, in this case, arm laser or battery welding, it's then attached here. So that's the flexibility we have, in our system to really cope with a lot of different processes. This is my typical example for welding of a gap, of detecting of a gap. Basically, that could happen in between like a horizontal gap. So, we had 2 sheet 3 3 to 3 sheets of metal. The one in the middle had cutouts. And, basically, if you weld in the outer section, you hit the full material. If you weld in that area, you hit the gap. And the gap could be 100 or 200 microns, thick. And, therefore, you can see okay. If we had, the 3, layers in a nice spot, you can get the nice weld. If you hit it, for example, in the gap area, you can see there's a void showing up and, of course, a gap. And so we train those, conditions, in our system, the exact word 430, where we are looking basically again through the galval head onto the belt. And those are the signals, the no gap signal and the 100 micron gap signal for the human eye. Not that much different, I would say. But, of course, we can apply artificial intelligence in order to train those three conditions, no gap, a 100 micron gap, and a 200 micron gap. And we did this only on 4 samples, which is quite, astonishing that it all already worked. And we get, a correlation up to, 97% for the 200 micron gap and already 84% out of the laser back reflection. This is already pretty good, predictions basically of what was happening. What we then did is basically throwing that on the machine. We went to a certain position on the sample. We didn't know where. Of course, we did this 2 centimeter long weld and it immediately came back. In this case, we hit a 100 micron gap, obviously. That's how it's made on a machine. Then of course, you can get a green signal if it's the value you wanted to have. In this case, of course, no gap, and you will get a red signal, for the operator in case this is not the wanted result, and not the trained result. Another application here is a pretty tiny one because it's razor blades, so the whole total thickness of, the blade and the frame is like 200 microns. So we only use 80 watts, CW power and modulate the beam. So we are talking about 12 spots within 150 milliseconds. This is our production mode in the SmartSense tool suite. Basically, you get an indication if the weld was okay, if it's all within the pathway. This is really a fast feedback in our production tool for those tiny, tiny welds with only 80 watts. If we go up, for example, to applications with 500 watts, for example, welding of a battery tapped to a terminal, we can have a look at the signals because it's a pretty, yeah, line welding. There's not much going on in the in the pathway. It's, this is the upper lip, limit, and this is the lower limit of the pathway. When the next one is in between, it will show up green. But of course, there could be something that's going wrong. For example, if the tab is bent, then we can see, okay, it's violating already the visible range quite significantly, but most prominent, of course, is the temperature increase when there's no good contact between the parts. Then you can see in the relative temperature signal that there are 2 really distinct, events where we see a displacement of the 2 foils. We can definitely determine that as another k, welding. Another one, misalignment feature here is showing up also significantly already in the laser back reflection area. You can see there's a lot of deviation from the normal path as well as in the temperature, which goes up at a certain point and then goes down at another point. It's really nice information you can get out of those signals. If you again have a look at the production software tool for that application, battery tap to terminal, you can see on the left side is a video of the whole stack. It's a household batteries in this case. Sorry. And we are welding, the TAP 2 terminal with, CW, and we are welding the pulse mode. At the end of this, stack, there you can see the signals over here. And it's basically indicated if there is a successful weld, with a green bar at the position you have determined, in your program. Also quite nice feature here. If it would, for example, show up as a warning, it would be yellow. If it would show up as a not okay valve, it would be a red bar, in this case. More examples on boost bar to terminal welding. In this case, we talk about a 4 kilowatt laser welding application. This is the okay signal pathway, so you can see, okay, it's not as straight as a as a line anymore. It's, going up and down because we'll be going around corners for that square weld shape, we are doing here. And, especially in the temperature, you can see it's, heating up in the corners, where the speed is not reached fully, therefore, heating up in the in the edges. But, of course, things can happen. For example, there could be a dirt between the two plates, and therefore, you can see there's in the visible range, there's pretty strong plasma signal over here as well as the temperature is going up in that area, and you can see also dirt on the surface in this corner. It's showing up, of course, later in the signal. This is something we can typically detect with SmartSense plus benign process monitoring. Just the last example on that one is, basically a case where, SmartSense is used for process development. We can see on the right hand side that, we are doing welding with the ramping of the power. On the left hand in the left hand video, the power is basically, ramped down from the center to the edges. And on the right hand image, we are basically increasing the power when we go to the center. So this is more a more a flat weld, whereas this is, basically a more deeper penetration weld that you can use. Those are high speed videos that show really the process. But of course, if you use SmartSense in this case, you can see all the signals for that kind of belt. The green signal is actually we plotted, the programmed power level, so the ramping, basically, so we're really ramping fast up and down, but not down to 0. The white curve is the plasma signal, the visible light. The red curve in this case is the laser back reflection and this is, of course, in the beginning, showing a stronger signal because it starts with the in coupling of the beam into material that is not as effective in the beginning. As you can see, there will be a lot of reflection, for example, at this point. And, once you reach a certain power level, you can also see that more laser power is absorbed in the material. You can see that the red line, is going down and then, of course, left at a certain level. The temperature is a yellow signal in the background that is, rising, of course, LLE once you perform those, laser welding. So it's really also a very good feature for process development. If you look at your signals, you can see, okay, we're not coupling into the beginning very well, so we might not ramp down or not start the ramp at 0. We might start at a little bit higher power in order to have a more effective coupling into the material. So also that, was done in this case with a Smartwell Plus, hat, in this case, the 20 PH 20, and we took the signal at the end of the beam path here and into the fiber. Lots of signals, lots of possibilities to determine the weld quality immediately after the weld. That's something we are using more often now, also for all kinds of processes, not only welding, but also ablation type processes or even cutting. It's a really nice tool. What is left, basically, on the right hand side is now, what to do with all those data that we collect and also, basically, what needs to be done in order to have a system check-in a longer sequence, in a longer period of time. Check the system wellness, the beam quality, the power. This can be done with a lot of data that we get from our systems. This is an acceptable 430 welding system. You already collect quite some data like gas pressure. You can have a distance control in there. Optics contamination can be determined and detected. Internal power meter of course of the laser power used, is something we can track continuously and of course more system condition indications like temperature and signals. All of those data need to go somewhere at some point. But this is part of the system check. Another aspect that customers are asking us is basically, okay, how was the system utilized basically over one day? Has it been productive for the highest part of the use? Has it been in maintenance, in idle, in setup state? We basically collect those data and times, and at the end of the day, we are sending out, for example, to a Cloud or to a customer database. We are sending out the information. The system productivity for that day was, I don't know, 90% production, 10% idle, and then these data can be collected by our customers. Or, of course, we can also come up with solutions that we provide a landing page or landing spot for those information on our system. And how could yeah. Industry 4 point o's has been a topic for the last, several years. What is coming up, of course, is the OPC UA communication way that is more standardized now. Many suppliers are moving towards that communication way, and we can imagine, for example, having an OPC UA server, for different components, in the system. And, this could be in the laser, this could be in the processing head, this could be in the process monitor, and those data are being sent then through a gateway, which is a safe way to block it from, from the outside, either to the cloud, to a server inside the factory, wherever you want to have those data, where they can be collected and where they later on can be analyzed. And, just a demonstration of that, could look like that. This is a cloud based landing page where you can also define your data monitoring, your dashboards, and you can see how the system was utilized, how many parts have been produced, and also, basically, what kind of power level, the laser was showing, over the day. And this can be then used later on, to be analyzed and optimized, through all kinds of, technologies like AI as well in this case again. So Lots of things, lots of data can be collected and will give us more information about the process and hopefully help us to improve the quality. That was my quick ride through my presentation. At this point, we would like to start the poll questions. Thank you very much. So are we bringing up the poll questions? Here we are. Now it's retargeting. Honey, it was on local database, by the way. Let's give it a few seconds more to see more data coming in. So it's stable now, so probably a good time just to, to pull it down. Oh, no. Few more. Alright. Alright. Actually, I'll I'll have a question for the audience. Why why don't you wanna use a secure file? So if you wanna if you wanna post the chat on that, that'll be interesting to see that. Security. I'm sure security will. I think I think that's generally what we're what we're seeing, Robin. Right? It's generally most customers are just looking for a local database connection. Absolutely. Absolutely. Yes. I mean, we just have to look at the hand shaping between our system and their system and the format of the file transfer and answer the top line, which which is something we do routinely in there. Yeah. Most of customers do not really like to put the the system directly into the Internet and into the data connection. So sometimes, of course, for for maintenance, it could be, reasonable, but I think most of the customers would like to have it on premise or in the local network. Yeah. You want to just put the summary slide up, Roland? Yes, I will do that. So, well, thanks a lot, Robin. That was super interesting. I mean, there's just a lot of good stuff in there. You know, being being a welding guy, there's still a lot of good stuff. And I think, you know, the easy vision I mean, I remember engineers spending weeks looking at these cases in a corner of the room, trying to get it all right. I would just click on an icon, drag it across, and then and we're good to go. So I think the thing about the vision is is that, yeah, it's been around for a long time, but the implementing use is just so easy these days. You don't even have to be super qualified in order to have credit to work. You can have all the embedded libraries within the software and ready to go. The one I think is really interesting is this post world inspection. I've I've used this personally at at a beta level, and and it's pretty impressive. So, it'd be interesting to see how that progresses. And there is a question about that which I will bring up is is what is the time penalty for that, which I think is a really good question. I think the other thing for in process monitoring, again, this this has been around for quite a while, but now we've got the ability to really extract the the correct meaning of data from from all of these signals that we get. And because label side and acoustic, which previously couldn't be looked at because it was just too complicated, we're now at kind of 4th dimension, which gives us a little bit more redundancy in accessing whether the the world is good or bad. And I think, obviously, Roland highlighted the fact that there is a good chance now to talk about hierarchical errors. So, you know, you don't have to, you know, be fumbling around in the dark on your on your latest system as to what has gone wrong. You can start to have a little bit more of an informed decision about, okay, it looks like it's gonna be this. Let me go and have a look at that, and that can save you hours and hours of downtime. And the and the data management, as I said, we've got we've got lots and lots of data. So, you know, what do you do with it? Terabytes of data. I think that's still, you know, an open debate. But I think you you kind of answered the question that we we've been seeing as well as that most likely, it it's a low local database, connection, which seems to be the way people are going. So with that, I'm gonna go to some of the questions. And, having so, Robert, I did see your question. Thank you for that. So on the on the postcard inspection for, so doing the the optical postcard inspection, Merlin, Robert's asking, you know, essentially, what's the time penalty for doing that on a on a cycle? Got got a measurement, which is probably too relatively quick, you know, as long as it's in the in the view of the camera. Maybe you can talk a little bit about that. Yeah. I mean, basically, to, to see what is what is necessary, what kind of technologies need to be applied before and after, the actual process, which actually, as you said, have a lot of influence on the total cycle time. We really have to see are those absolutely necessary. But you can think about, 50 millisecond, 100 millisecond impact, basically, on easily on those image collection and and analyze analyst, down times, I would call them. And, therefore, it needs to be judged, okay, if I only have a, 50 millisecond weld, that is adding a lot of, time, doubles the time, basically, just for the image collection. But if I have a, I don't know, a 10 second valve, 50 milliseconds might not hurt that much. Therefore, it needs to be really, looked at the exact application to judge if it's feasible to apply those technologies or if it's just not. But, of course, it depends also on your your quality overall. If the, if the quality is already pretty good and you're don't need to check it every every pulse or every, every part that, of course, it also not exactly necessary. Yep. Yep. So okay. I think this is applying yeah. So how how does your technology compare to already existing process technology? Obviously, you know, monitoring's been around for quite a while. That's fair enough. I just wanna also point out, this is a complete in house developed system. You know, we're not we're not purchasing this from a third party and doing and doing some additives to it. It's been developed by us. You know, we consider ourselves welders for the welding community. So Just fwise. Anyway, Roland, maybe you can, indicate why we think it's such a good thing. Yeah. This is one of the beauties of that system that we are really flexible towards, really from detecting processes in the in the in the single digit watt area up to the kilowatt, region. And, of course, we also have, a really high sampling rate, which is also outstanding. So for example, to collect data on a on a femtosecond laser process, you should be in the range of 100 kilohertz. So we can go up to 5 megahertz sampling rate if necessary. Of course, this is collecting a a lot of data, so it had to be stripped down at some point. But, I think the resolution as well as the high data acquisition speed are distinguishing factors here. Yeah. And just just for that, there was there was a comment about, is there a speed limitation, on what the monitor can see? Essentially, it's it's happening at 200 kilohertz. So, you know, you can you can do you can figure out what the spacing would be if you're running at whatever speed, and 200 kilohertz is a it's pretty quick. And I think that's that's by far the past on the market, though. Right? Yeah. Yeah. Absolutely. Yeah. Yeah. Yeah. Okay. Another question here. Can you set the world stop when parameters fall out of range, continue with inspecting the terminals of what is acceptable? So, basically, can can can you have an can you have an output from the monitor which says, you know, it's our limits, stop the wells, have a pause, and then the operator can go and have a look to make sure everything's okay. So I need to see where we are with those, questions. Yeah. It's, Joe. He's about Yeah. And stuff. Okay. I gotta I could take this question. So does the SmartSense system have the ability to learn how to adjust laser output due to welding items with high reflective surfaces? That is an interesting question. It's not that easy. Of course, we will, we have an output, basically, so we could modify or influence the power level, like, for example, done in in parameter controlled systems, we could do that also with SmartSense. But typically, the, welding we are seeing is not impacted by only 1, parameter, like just the power would help. Sometimes you have to also change the frequency in order to avoid overheating. So it really depends on the application. For a CW weld, I think it can be done because we have the the the 0 to 10 volt output that could then, use as an analog input into the laser power, side. But, something we can surely talk about. And just and just going back to that previous question, I think I think the question was if if the if the limits have exceeded, you know, you wanna have an output that says the limits have been exceeded, and that goes that goes to the machine to say stop process. Right? Yeah. Yeah. So, for example, if if your signal's exceeding a certain threshold value, yeah, I think we can we could implement something like a control, out of that. Right. Okay. But then do you how do you handle yeah. I guess you get this question a lot. The only CT comparison, you know, how how do we how does it compare to OCT for welding? Yeah. I mean, it's a different approach. OCT actually looks into the keyhole and gets information down to the depths. We, with this diode based technology, we have a relative information. We don't have an absolute, value, basically. So we always have to train, certain conditions. We have to collect, I don't know, 10 good samples in order to see, okay, how the 11s looks like. And, so it's it's all relative data. But, of course, typically, it's also the more cost effective approach. And if it's sufficient to, especially not for deep penetration valves, OCT also hits a certain limit for resolution. And we are talking here about mainly micro welding, then SmartSense is absolutely fine in collecting, the right signals on that. I think I think also as well, I think the smart sensor has a bit more, error identification range as well. I think the OCT, is a little bit more one dimensional on how it can do thought analysis or thought thought analysis. So I think I think there's, although I think that side of it is still is still developing rather than right. It's still the ability the ability for us to do, you know, may may be a hierarchy of full analysis rather than, you know, this is not the right one of it, I think. But, and there's a little bit more flexibility there. Okay. So I understand through the list here. So then another question here. We talked we talked about welding. Of course, this session is about welding, but I also mentioned, that there are more applications we can target, for example, for process monitoring. As I can connect, our SmartSense, detector to almost any laser we are using. You can also look into, marking applications, into cleaning applications, into, drilling applications, or cutting. And this is also quite useful on on that end. And our frequency, detection frequency is high enough also to cope with those, high frequency laser applications. Just just another question here. Can the system also recommend parameters, based on the material? I think I think we're using it as more of a as more of an offline tool at the moment for process optimization. I don't think we're closing the loop on that. Right? Nope. Yes. That's correct. Yep. Okay. How many samples to train deep learning defects detection? And I think we're looking at continuous welding. So if you're doing if you're doing a continuous pipe weld, I'm assuming something like that, how how does the sensor work for that in Liggett keep essentially forever, or do they have to stop at some point? We have 2 modes, of course, in there, or we can see 2 application modes. Of course, there are processes that are quite short and they're contained. After a second, the laser welding process is done and it's always the same. Then we can, of course, train exactly that process. For processes that are ongoing basically continuously, we cannot do that because we cannot wait, I don't know, 3 hours. And, after that, we will, give a signal, oh, there was something wrong. Therefore, we've got also got an online mode where we continuously basically track little elements of that continuous weld and immediately then feedback after that. For example, also for processing expensive parts, it's important to know if something does go wrong already in the beginning and not wait until the end. Otherwise, you're wasting too much material in some cases. Therefore, we have 2 modes that we can apply here in the software. Okay. Just one last question was the was, you know, the the the AI involvement in, the in purpose monitoring. I think that was, again, that was something that we developed in house, but maybe you can just talk a little bit, to that. What what's the AI actually doing there? Just a little bit more, behind that. Okay. So what I what we can basically do is, as I, as I showed you, is that we have the pathway, which is the standard, evaluation method. We're collecting good samples and then compare the next one to it. If it's violating, it will be bad. But of course, you can train, also certain failure conditions because the signals might look pretty specific. We might not see it with the naked eye, but with AI algorithms that you can apply in our software and there are various ones and you can also set the parameters for those algorithms, you can then train certain failure conditions. Once it shows up like that, it will basically flag it as, okay, this was exactly or it was with a high probability this trained failure condition. And that's what I showed also with my, gap, welding where we hit either full material or the gap. And so we trained, those three conditions, but only with 4, experiments, which is quite astonishing, as I said, that it already worked that that well. Yeah. Okay. The better the the more the better, of course. Yeah. Yeah. And then, obviously, if you have if you have a production line and you know what specific faults are, obviously, you can target those faults, and focus on those. So if you have 2 or 3 that sometimes a little bit more than than less than, yeah. Okay. I think I think we're I think we're done. I think we've got a call out for time. So, thanks everybody for attending. Thanks everybody for for listening in. So just another reminder for those of you who are gonna be at MDNM West, Yeah. Just come by. We'll we'll have it up fairly quickly, just the the last. We'll actually have the monitoring system there and, the vision, and, also, you can see the the laser framework software. And if you haven't seen a laser framework software, it's actually really cool. You know, it it's it's much more drag and drop. You don't need to know g and m code. You know, it's a b stored operator thing. Very, very user friendly. It'd be I I highly recommend you you just sort of streamline to see that. Okay. Well, thanks, everybody, and and, thanks for attending. Bye for now. Thank you very much. Thank you very much for attending.