Name: In Process Laser Weld Monitoring
Uploaded: 2024-08-28T16:37:17.045Z
Duration: 1 h 4 s
Description: In Process Laser Weld Monitoring

Transcript for "In Process Laser Weld Monitoring": Alright. Seems that we are live now. So hello, everybody, and welcome to this, Coherent Amplify session on process monitoring. Today, you've got 2 hosts that will guide you through the session. It's myself, Roland, and Marcus right beside me. So maybe just a little introduction of ourselves. I'm with the company for more than 20 years now. I originally started, to study material science in Germany. I'm from Korea in Munich, and I studied also in Bavaria. But material science wasn't that interesting, but we had a nice group that was dealing with lasers in the early nineties. And so I really joined that group quickly and got into laser material processing, and, that's what I'm almost doing till now. So I'm always coming from the application side. I did, application development, process development, and now I'm the product line manager for marking systems, in Coherent Munich. And Marcus will also introduce himself. Welcome from my side too. My name is Marcus Donner. I'm the product line manager at Coherent Switzerland. My background, initially, I started physics and then, moved on into product management and sales in different branches, specifying and specifying production equipment and lines in branches like solid wood processing and photovoltaics before I joined the laser branch about 12 years ago. So I can't beat Roland there with this 20. Very good. Thank you very much for your introduction. So a little general information on our session. So we are streaming live today on 3 different platforms. That's Goldcast, that's YouTube, and LinkedIn. So our session will basically contain our presentation in the first half. And then at the end, we will have a q and a session where you can also always ask questions already now. Put it into the chat, and we will basically answer them after the presentation. There will also be a poll question, but this will only be available to, answer the poll for the GoldCast registered members. All the other ones on the 2 other platforms will, of course, see the result, of that poll. And, with that, I would start to share my screen, and we jump right now and immediately into the presentation. So you should see my screen now. Today's session is about in process laser weld monitoring, mastering your weld quality because we will mainly talk about welding. I will start the presentation, and then roughly in the middle, I will hand over to Marcus to finish with some applications on different welding processes. So first of all, little introduction on the outline of the presentation. After a short introduction, what we are talking about here and what the goal basically of process monitoring is. I will show you what we actually can offer with our SmartSense Plus style based process monitoring, And we will jump into process monitoring application for laser metal welding. We will basically divide, the applications into high power application and medium to low power application. Typically, we make the cut at around 1 kilowatt. Below 1 kilowatt is typically called a micro application. Above 1 kilowatt is typically called a macro, welding application. And then we'll summarize the whole presentation at the end, of course. So little introduction. Of course, every laser manufacturer, every manufacturer of a system like we are in Coherent, we, of course, try to build a system to the highest possible standards in order to give you a very good processing tool, for highest possible productivity and uptimes and and yields. But we all know that, Murphy's Law exists, and I'm not quoting directly, but I can say issues will happen. There is possibility of degradation of components. There could be a pollution of optics, contamination of optics due to the process. There could be material issues, for example, the wrong material batch, not the right composition anymore, part accuracy not being precise enough. And, of course, there could be operator errors for part misplacement on the tool, and those issues will happen. So the goal, of course, of everybody is that to notice those issues as early as possible. Otherwise, they will go downstream in the process flow and, basically, cause more costs and more problems down the down the road. And worst case, of course, they will even ship to the customer, and you have to call it back, at some point. So the the whole goal is basically to check and control the process as good as possible in order to avoid exactly those, issues going or wrong parts going down the processing chain. This is a processing chain. Basically, the blue arrow line is a timeline on the left hand side. You basically can imagine, okay, now the part is arriving at the machine. Somebody is scanning, for example, the travel card with it, and then you load the part into the machine. And the first process control is typically covered nowadays by vision, so we check, okay, is the part in the right position? Do we have to move the object that needs to be lasered? Is it actually the right part? Because we can also do a validation check. After the process, same thing can happen. You always can use a camera in order to check maybe the marking or to check the weld quality. But, of course, also, you can do something while the process is happening in order to get an immediate feedback after the weld. Was it okay or was it not okay? And this is basically what we will talk about today. So what we propose here and what we wanna present here is based on dial based monitoring, but can also include acoustic monitoring, which is our SmartSense Plus solution that we can add to many different lasers in our portfolio. And so what is it? That based in process monitoring called SmartSense. It's a monitoring optics. Basically, we catch the process light that's happening while you do a welding, while you do an ablation type process. So we basically catch that light and guide it onto a fiber into a detection box, and there, we will split it up into 3 detectors. One will cover the visible range. One will cover the laser wavelengths, typically around 1 micron, and one will cover something like a relative temperature range between 121700 nanometers. We can also combine our detector box with acoustic sensors, for example, for airborne, microphones or contact microphones. If you wanna listen, for example, to the part, while it's welding, it could be quite interesting to also listen to it after the weld in the cool down phase because you might hear the cracking, also going on with the acoustic sensor. And then, of course, we wanna decide basically on those signals we are collecting, was it okay process or another okay process? And, what the whole goal basically in our determination is that we get an immediate feedback, in our tool that is, yeah, divided into 2 stages. So what you can do for okay, not okay process labeling is using the pathway. So, basically, you record a couple of good wells. Let's say, 10, 20 wells, in this case. And then the pathway will basically be determined by the lower level and by the upper level. For example, in this image on the right hand side, the red curve is the upper level, and the blue curve is the lower level. So if the next one, like the white curve, is right in between, we can call it a good process. But, of course, if it's violating that pathway a lot, it will basically come back with another okay process. Then it would be interesting to learn or to see why it was not okay. And for this second stage, we can use a classification method using AI, using machine learning, so we can train certain failure modes. And they will be basically considered and evaluated and given back to the operator with a certain percentage level, of probability. So, basically, one is telling, okay, not okay immediately after the process, and what and the the second second stage is basically telling, okay. If it was not okay, it could be, that typical trained issue, with a certain probability. So that's what we can do, in our software because software is a really important part of our complete SmartSense solution, which, of course, contains hardware and software. And, in this software, you've got different settings. This is showing the engineering mode. So this is basically for the process developer in order to set up the process, and you can train your, pathway components. Basically, here, you can get a result here, but you can also train your machine learning, here. And, later on, Marcus will show you specific part. And, yeah, software is really important for that kind of process monitoring and has to be really fast in order to give a feedback as soon as possible after the weld. So, wanted to say, SmartSense is really flexible because it's our own development, so we can change components, not only electrical, but also optical components quite easily. As I've see shown on the first page, we can go up to 5 megahertz sampling rate, which is really high. Mainly not necessary for welding, but really interesting especially for, short pulse applications with nanosecond lasers, and picosecond lasers as well. So, of course, we've got applications where we connect, this optical sensor interface with a fiber, that looks through the galvo head onto the part, and it basically collects the information and the light, going backwards. We connect this to a 10 watt, infrared picosecond laser. We connected it also to a 20 watt UV nanosecond laser for some ablation and cleaning applications. And, of course, we can connect it to a 10 kilowatt, so 10 orders of magnitude higher, laser power, for example, with a CW arm laser beam path here at the end of this beam path. So we are really flexible in terms of wavelength, and power levels that we can work with. So what are typical welding issues? This is typically what we see when we talk to customers and evaluate possible issues. Of course, gap can always be a problem. The gap can be horizontal or vertical, depending on the part fix, but, of course, this will influence your welding quality for sure. Too low, too high laser power. Very often, we see that most of the problems are coming from out of focus processing because something, just moved, accidentally. Also, no shielding gas because either the shielding gas bottle could be empty or, somebody has accidentally moved the shielding gas nozzle, so it's blowing into the wrong direction. Surface contamination, we will for sure see in the beginning. The cover glass is something we are asked to determine basically during the process. And, of course, also wrong welding materials, either they're upside down on the sandwich welding or they're just made of the wrong material. And just a little introduction, that's an, a setup we have shown at various exhibitions throughout the last years. So this is just showing basically the influence of a gap, and, of course, 2 different gap heights, on our signals and our evaluation, basically, using AI. So what you see here is the sandwich structures. We've got 3 metal sheets. You've got stainless steel. And in the middle sheet, we basically placed some cutouts. So depending on the position where you're welding on the top, sheet, you either hit full material if you, for example, weld in this range, or if you move over and you hit the gap, then, of course, you weld through a gap. And by variation of this middle sheet in 100 and 2 micron thicknesses, so this is 4,008,000 thickness, we can basically have 3 different states that we can determine. So we've got the no gap stage stage, which is, that, of course, giving a good weld, and, of course, we've got the stage where we have a gap in between, and you can see it's already causing problems, a void in here, a sagging of the weld. And this is, of course, not the state we wanna have, and this already happened here at a 100 micron gap. So, this is what we want to have. No gap. Of course, this is the desired process, but, of course, we will determine also the other ones. So the setup was, using an acceptable 4 30 machine where you can see that, the optical sensor interface here is connected to a complete Galvo head beam path. We've got the camera in in here. We've got the laser coming, at this position, and then we basically, get back all the light through the whole, setup onto our fiber and into the detector box. So that's the setup of this welding trial. So we determined and collected, the signals. So this is the not no gap signal. This is something we want to have, and it's not quite obvious. This is the 100 micron gap, so not much difference to the, human eye. But, of course, as you can see, it's already red. So, the machine evaluation, the software already determined, okay, this is deviating heavily from our no gap condition. So what we did, basically, we learned and trained these three conditions, no gap, 100 micron gap, and 200 micron gap through our AI machine learning algorithms, and only on 4 samples. We got back basically a very good match between the true class and the predicted class. So, basically, this is 89%, 84%, 97% on predicted class. So it's a really good match between what we expected and what we actually got. And then I'm just going to show you how that looks like, in our machines. So this is the display of the machines. So we moved to a certain position on this sheet setup. We welded a centimeter long, and immediately, after that, basically, the result came back. Okay. This is a 100 micron gap with a certain, probability. And, of course, the higher the probability, the larger this ring is filled. But, of course, it's red because it's it's determining a gap because we want to have a green no gap, issue. So, basically, this shows immediately that we could train three conditions with only 4 samples and get to a good quality, result after the evaluation. So that was my little introduction to that. I'm going to stop sharing now and basically hand over to Marcus to guide you through some more applications, starting, of course, with the higher power level. And I'm sharing. No. You're off. Thanks, Roland. And I'll share my screen. Just a second. There we go. Good. As, Roland already announced, I'll give you an overview of a few applications, across power levels from a few watt up to multi kilowatt level. So I'd like to start with a high power application. Today, we have a lot of demand from the, EV industry, from from, e mobility. So it's busbar welding to terminal of prismatic batteries. So this requires welding of about 2 millimeter aluminum, busbars on top of, another aluminum contact. And these are welded with a with a a coherence highlight FL arm laser, the articulated ring mode, where the power level in the core and the ring can be modulated individually. So welding this, it's, you see, welding depth is about 2.5 to 3 millimeters. The weld is applied in 0.4 seconds and at a laser power of about 4 kilowatt. Goals here, we need to be very fast and productive, achieve a high conductivity, and, of course, a mechanic high mechanical strength. To prove this in production using SmartSense Plus, Here, we have an example of the three signals of the individual sensors. As Roland already indicated, we are monitoring the plasma and the visible range. We are evaluating the laser back reflection and, also the relative temperature in the material. So this is a typical typical, for example, signal at the beginning of a weld. First, the laser is reflected and then couples in stable into the material, and you should see a really stable signal all along the weld. Where the plasma usually, is a lot more unstable because it moves above the work piece and, is, of course, also affected by extraction systems or cross jets below the processing end. So you see the the good signal, the signal pathway is comparably narrow, in between several work pieces. Compared to this, a weld where there was some dirt in between the plates. This can happen during positioning. And, immediately, this causes actually virtually an explosion in the weld creating a cavity, and it's clearly visible on all three signals. Another example, just dirt on the surface, not as severe as the first example. So this is not the dirt. It's the the burn marks caused by the dirt on the surface. Also here, clearly visible on on the three different wavelength. So the next few application examples are, all in lower power ranges, so below 1 kilowatt micro applications as Roland called them. And it's another example from, battery manufacturing. Here, welding of tap to tap. So these taps here are 0.3 millimeter thick. So we need a very stable penetration depth in order not to to damage the the lower layer. You see again, in the 3 wavelength ranges, the the pathway. So the pathway is usually quite narrow. So the signals are very stable on all three wavelength where, again, in this case, a bent tap caused a very obvious deviation that would be detected as not okay. And, also, a misaligned tap, causes, deviation in the same magnitude. Both would be clearly identified as not okay. Here, we have a video of the line in production. You can see the the operation on the video in the bottom left corner. Here in the center, we have the view, of the work piece. So this is actually the production mode that displays only the information that is relevant on the shop floor. Relevant for an operator or, well, personal overviewing the process. So every wealth position is clearly indicated by, okay or not okay. And then in the end, also, the the complete work piece is then qualified. At the bottom center here, you can you can, select the different, signals that can be seen. So you can flexibly vary the the interface that the the operator sees. Another example for a micro application is welding of razor blades. So we have very thin, stainless steel blades that need to be welded with individual spots at a laser power of only 80 watts to to the supporting base material. And so you can see every single blade is indicated here as good or not good, and every single spot in the weld is monitored. Again, we see all the individual steps and the the complete result of the work PC. This actually needs a a very high sensitivity of the system at those low power levels with very low noise levels to be able to clearly identify a good or not good weld. Another application where we are welding, pieces directly on well, electronic components directly onto a PCB. So that means we have on top a layer of about 100 micron copper and weld these on top of 35 micron copper tabs directly on on, PCB. So that means under the 35 micron copper, there's directly the the supporting plastics that is, of course, not allowed to be damaged. We are doing this using, a fiber laser, a single mode fiber laser combined with our PH20 SmartVault Plus processing head that is capable of controlling the energy input into the material very precisely by controlling the speed of the laser spot on the surface very precisely and within a single a single pattern, even at frequencies of 2 kilohertz repetition rate of the pattern. So the the weld happens in 10 within 10 milliseconds. And you see we have a very shallow welding depth as the material is dead thin. Still a quite high laser power, but, that's so with that, we achieved the high speed. So how do the signals look like? You see here a good weld, the top view, and this is after a successful peel test, proving that the PCB below is not damaged. Just a very slight print on the surface is visible, but this is acceptable by accepted by the customer, and it's not burned. So that's no defect. You can see the signals after the initial peak in the back reflection and also in the plasma. We have a very stable signals in a good process. Increasing the power level to have to create obvious defects, you can see here slight burn throughs that can happen in very localized small, positions. Again, the whole spiral takes 10 milliseconds, and you can see, how fine the resolution is regarding the signals. You can see every single incident reflected in the signal. Of course, if we increase the power net notch, it's very obvious to find that. The more difficult ones are these here, where from seen from the top, it looks okay. Maybe there's a slight discoloration, but that's it. And then after the peeling test, you see there's a lack of fusion in a in a certain area of the spiral. And during the weld, this is clearly indicated by sharp peaks on all three signals. Sorry for that. Process optimization. Roland earlier introduced the engineering mode, and you saw the production mode in the videos of the the different applications and production lines lines. Based on the engineering mode that Roland introduced, that shows the details of the measurements, it's possible to optimize processes because based on the signals, you can you can clearly identify whether the laser couples stable into the material or not. So that means you can determine exactly the point where the process becomes stable at the minimum energy input that is required, which is very important to minimize the heat affected zone and the heat input in the complete work piece. So here, in this case, we welded also with an with an ARM laser, with different profiles in ring and core. And this was the starting situation here. And the result was in the laser back reflection a very unstable signal. So the process was not stable. Then optimizing the profiles, coming up with completely, individual profiles for ring and core and and different, timing in the modulation, We achieved a comparably stable signal in the laser back reflection and achieved a very nice weld in, actually, foil copper foil stacks. So these can be foil stacks of up to 50, might be 100. There's a huge variation in the market where it's required. Today, it's about, 30, 32 foils. Some manufacturers are talking a lot more. And this is an example where we proved that we can reliably weld the foil stacks. So here, just a spot. In other cases, a unique approach with the PH 20 smart cloud plus welding the complete edge of the foil stack. So process optimization, for process optimization, it's a very useful tool in r and d labs and r and d departments too. So I come to the end of the presentation. I hope the, the application examples were interesting for you. So the summary, here we have thanks to the process monitoring, issues are detected before they cause additional, cost down the production chain as Roland already indicated. And doing so, SmartSense Plus can cover a large variety of welding processes ranging from nearly 0 to a multikilowatt level. And then based on the the excellent sensitivity and the clear signals and minimum noise levels on top, in combination with the higher sampling rates, SmartSense Plus is also suitable for laser marking, cleaning, or structuring processes based on ultrashort pulse lasers. With that, I come to the end of the presentation, and Roland and me will be available for questions. Thank you, Marcus. So before we go into all the questions, and then trying to answer all of them, so thank you very much for all the questions that have been put into the chat. We wanna do just one quick poll, basically with you, especially, of course, the GoldCast attendees, and you have to go to the go to polls tab, tab on the bottom left side in order to participate in that. So the question basically for that poll is, do you consider process monitoring as an important aspect for future laser systems? And we can see already a strong indication that even considering the maybes, that this is very important topic. Of course, it was about to be expected because, we see all the advantages, and we, of course, we see also that everybody has to reduce scrap rates in order to save money and also avoid callback situations for parts that have already been shipped out to customers. So, of course, the earlier we noticed basically that there's something wrong in the process, the the cheaper it will be in order to fix that. And so if we look at the votes, which seem to stabilize now, it's, 90% are considering one just change back, maybe. I can understand that for that for some applications, it's less important if there are no high quality requirements. And, of course, the examples we picked were from electronics industry, were from automotive. And, yeah, we are used with high quality branches. But, of course, if, mechanical strength is not a 100% required and you lie allow high tolerances, then you might not need it. So, yeah, thank you very much for participating in that poll. So it shows the importance of the topic. Thank you again, for that. So we would now go through all the posted questions. Marcus and myself will pick some of them. I think I will start right at the beginning from a question on open. I think it was gone. Is it gone now? Yeah. Okay. Can you operate open welders? Of course, you can open class 4 welders, but, of course, you have to secure the room and make it laser safe. So, of course, we, at Coherent, sold class 4 lasers until 3 years ago, but you have to take the precautions then to the bigger room and make that room safe and, of course, not accessible to everybody. Then you can all also work with an open, class 4 laser in your production. So Roland, actually, there's a lot of questions in the chat, so we might have to pick and will not be able to to answer all of them. So one question that I see here is that, that you say in in the slide I showed with the prismatic battery cell application that a defect length, wouldn't exactly match the data length. That's, of course, possible because, the, the effect on the surface can be a lot different from, from the, the signal. And you saw, for example, in the, in the example with welding on the PCB in copper, that on top, you didn't see it at all. So, therefore, there can be quite a variation, and it's not identical. So there's also an early question. What data size are we talking about for an, for example, one second weld? Of course, that's mainly depending on the resolution you are choosing, basically, on our software. For a weld, you don't need the 5, megahertz sampling rate. Of course, you can come in with much lower values, but I would think it's in a mega megabit range already also for that kind of, short weld. Would you agree, Marcus? Yes. Definitely. So we can we can also, determine what frequency measurement frequency is required for what application in order to reduce the the, amount of data generated in production. So there's a a 4 year analyzers integrated that checks in what frequencies there's still information content on the signal. And so you cannot you can choose you can choose that frequency and, so the, well, data will be minimized. K. Next. You wanna take one? Yeah. I just had one here. How SmartSense Plus differentiates from comparable competitor systems based on the same measurement principle. You notice in the slides that the clarity of the signals and the sensitivity, are very good. So we can also measure effects in very, very thin material and at very low power levels and with ultrashortpulse lasers. And we focused in designing the the hardware, the measurement hardware on really minimizing the noise level and succeeded. And we also adapt, SmartSense Plus to the laser optics. So they are they are properly aligned and we can also adjust the size of the measurement spot to in order to have the the maximal, signal to to noise ratio. So I think on the measurement side, it's a lot more sensitive than the systems out in the market. K. Another question. Can you extract data outside the SmartSense platform, including acoustic sensing into, for example, CS suite date, to do your own data analysis? I think the short answer is yes. The data will be stored, of course, in that format. And another question, how do you propose to evaluate the AI model? I mean, what we did, of course, as most people are doing, we, first of all, trained the model to this, for example, what I showed, 100 micron gap, 200 micron gap. And then when we have trained it and that we only used, as I said, only 4 trials to get this model trained. We then, of course, would, run another 10 or so, of the same kind, like a 100 micron gap, like a no gap, like a 200 micron gap, in order to evaluate the I AI model. And, of course, in production, you can always still add, basically, events or signals, to this AI based, bin where you say, okay. Okay. This was now again a 100 micron gap. I will put this signal to the, 100 micron trained model bin, and then, of course, it will enhance the quality of that as well. So here's another one asking how do we calibrate the system to validate its output data against quality requirements. This is, of course, a little bit tricky, because every there might be a might be a quite a variation of quality requirements dependent on what tolerance, what, pull strength variation you you can tolerate. And, what we do have is we have on the AI side, we can train, error characteristics. And then once you have, let's say, you have work pieces that are that are close to the requirements or just exceed them, you can put this data in a certain bin. So doing that, we can train we can train kind of warning levels and error levels. So the system is able to differentiate between you get close to your limits and this is not okay, but it needs to be trained. Another nice question. What happens in the case of an uneven gap? So, basically, I think this is also coming back to one what I showed, with our little test on 102 micron gap, training. Actually, I can tell you exactly what happens because we use that 3 sheet setup, at the show a couple of times in order to save material, of course. So we welded a couple of times onto those, and then we got some distortion already, on those parts. So what you can see, if you already had a 100 micron gap, you could see, of course, it was lifting and moving towards more or less 200 micron gap. And on the signal evaluation, basically, on the AI evaluation on our, system, we saw that the probability for a 100 micron cap is going down, and the one of the 200 micron cap is coming up. So, basically, this was a really nice side effect that we could see when we had an uneven gap, something that is not really trained. It would show probabilities basically in between. So Here's here's another one. Does SmartSense Plus work with any laser and processing optics? Right now, we focus on coherent systems, and and optics, but there are systems where that could include a coherent laser and the process optics of a different manufacturer, and there's the request of SmartSense. So we will try to accommodate that. We cannot do it with any optics because one question, of course, is what are the codings of the mirrors used in the heads? So what signals, actually reach SmartSense Plus? So in our, processing heads, we optimized those codings for a for a maximum signal strength in the in the 3 different wavelengths. Of course, working with other processing heads, We can only test that and see whether we can work with the signals. K. Another question. Can we detect or control the porosity generated by laser welding in the welding path? This is a really good question. Not easy to be answered because it depends when the porosity is actually happening. Is it happening while we still see some lights? So exactly, when the process is happening, is it happening afterwards in the cool down phase, due to some cracks? Then we would not see it. But, most of the times, I think we can we can see the porosity, and we have also videos on that. But, typically, of course, this needs to be tested. Right, Marcus? Yes. Definitely. Yeah. I have another one comparison to to OCT. Mhmm. OCT, of course, gives a measurement of the penetration depth during the process kind of measurement because the the data and the the individual data points, they have a standard deviation. And once it comes to very thin materials, for example, the the PCB welding that I showed. The the standard deviation is in the range of 50 to 100 microns dependent on on the systems. And so they they lack the resolution in very thin material to give a clear indication. So that's the difference there. We, on the other hand, we can, of course, not give a measurement of the depth, but we'll see the effects during the process that cause a deviation of the penetration depth and will indicate any deviation from the the stable process. So it's not a quantitative measurement. Yes. But we will definitely see any deviation. K. There's another one. You mainly showed welding and mentioned ablation type applications. How about cutting? Yeah. That's an interesting topic. We didn't really show, results on that. But, of course, in cutting, we also can see some very nice, application areas, especially for, through cut detection, with that kind of technology. So if we are looking at the laser back reflection, we can determine, if we actually cut through or if we get a lot of reflection, of course, we know we are not cutting through. This is important for some areas. It basically needs a different setting of our software because what we have shown so far is basically collecting data and giving you a result afterwards. For typical cutting applications, the cutting path might be quite different. You might not be able to train it, so you need a different, approach basically in order to look for the level, of your signals and determine a threshold. If it exceeds that level, then you might stop your process at some point or get a warning to the operator or just, yeah, print out a report at some point. So this is something, yeah, that's quite interesting as well. I just oh, well, ending in 15 minutes, although we have some time left. I just spotted another interesting one, an application related question. Can this be used in a continuous process well? We integrated SmartSense Plus actually in welding of tubes. So the task is to identify defects that are somewhere in the weld, and we actually call this the continuous mode. In the continuous mode of SmartSense Plus, you can define a length of the measurement, time. So you can say, I want a measurement every 0.5 seconds and then evaluate that to give you a good or no good indication per section on on the the seam. So this can be either applied to continuous processes like welding tubes or in case you have very long, big work pieces with 2, 3 meter long welds where it's hard to, analyze the signal over the full length in one go. Also there, we would we would cut it down into shorter sections for analyzers. So I think you can continue. Most of the questions probably go more in your direction. We answered already fuse. Let's see. Back reflection. How do you account for back reflection of the laser beam when using IR on copper? Yes, that's a point. But, for example, if you lose high brilliant single mode lasers with a very, very high energy density in the spot, This jumps the threshold, in copper where the laser couples into the material. And once you are on the material and it's heated up, you can achieve a stable process with IR on copper as well. And you saw the example welding very thin copper pads on the PCB. So we control the penetration that's very stable in that process. So it's just about knowing how exactly you apply the beam to be able to weld it with IR and copper. Then another question. Does SmartSense automatically optimize core and ring power dynamics, or the results you showed are obtained by trial and error by the operator controlling the machine? Yeah. That's a a good point. Marcus, you wanna take that out of your try? A second. Where where was that? 539. 4 539. Where's that? K. Let me answer it. I mean, it doesn't really optimize automatically, yet. This was a trial and error. So this was basically, an example for process optimization. So, also, we use in our labs in order to have a look at the signals, see where there's a lot of reflection going on, and then we can, basically counteract by cracking up the power, for example, in order to get a better penetration. And this was basically the example we showed on on Marcus' slides for, process optimization in the lab. What we, of course, can do is during those tests, we also determine how stable the process is and how big is the process window So that we and then, of course, you target a big process window ideally. And, so we have a clear indication what we can achieve and how robust the process is in production later on. Yeah. Monitoring system has to monitor coherent laser products or could scale to other main laser competitors. That's also, of course, a nice question. At the moment, we limit it basically to our own coherent laser sources and, and setups. So it's not a product that you can buy, as is, but only in connection with a laser. But, of course, there's always something we can think about if the volume was big enough. Right, Marcus? Yes. I mean, I partially answered that already already earlier. If you just use it and want to combine it with whatever system there is with a laser from, you need an interface. You need a mechanical interface and an optical interface. As I said, optical interface is important to, make sure that you have enough signal strength in in the important wavelength. And for for other processing heads, we cannot guarantee that. We have no idea because we don't get data about the codings. There's another one regarding the AI, to give a few more details, how it works. Any AI works only as good as it's trained. So if you want to if you want to, classify errors, of course, you need examples. So initially you might start welding and you have, errors that happen only very seldomly. So you start to teach the system with with the good work pieces, of course. And once you have an error, you can, you can pick this data set and assign it to a certain class for AI, depending on how many error classes you want to generate. It might be that it takes you another 2,000 work pieces before the error happens again. So still, you have to pick this and assign it to the to the class and that if you want to, let's say, classify 5 errors in the beginning, you need to do some sorting and, enable the AI to pick up the characteristics of that error class. And so over time, it improves with the classification result. Then one comment, right, in the beginning on slide 9, you missed quality of the welding fixture and also effect of operator skills and trainings. Yeah. This is alright. Thank you very much for that. But, basically, it will also be ending up in another error. So, basically, if the fixture is not right, you will either get a gap or a wrong positioning, out of focus working, positioning. So, and also operator skills will basically cause some of the already mentioned issues on welding. But, of course, it's a good point that this will basically influence the weld quality. Thank you. Copy. Ready. Does this work for keyhole as well as conduction welds? I I just came across the same one. Yep. You want me to answer? Yeah, please. Yes. It works for both. Of course, it's a completely different process, and the characteristics and the signals look somewhat different, but it works for both, processes. Yeah. Cost of the system, of course. We can always talk about offline. Go through your application. We will not cover that at this point. And I think we covered all of the questions. Right? I don't think so. Then I would say we there are no more questions coming in. Then I would say, thank you very much for participating. Of course, you can always watch, the Amplify session again. It's recorded. And, yeah, thank you very much for participating tonight. Hope to see you next time maybe with a different topic. Thank you very much from my side too, also for your interest reflected by the huge number of questions. Thanks. And, of course, the poll which showed, okay, we are on the right track. We have to continue to put more process monitoring capabilities onto our tools. Thank you very much. Then we will close the session. 5 minutes to an hour length. Thank you very much. Thanks again.