WEBVTT 00:00.000 --> 00:10.000 Okay, that's fine. 00:10.000 --> 00:16.000 Hello, welcome to my presentation. 00:16.000 --> 00:27.000 My name is Paula Fonda. 00:27.000 --> 00:31.000 My name is Paula Fonda. 00:31.000 --> 00:35.000 I'm from the Technical University of Orange Lake in Germany. 00:35.000 --> 00:37.000 I'm currently my colleague, Basti. 00:37.000 --> 00:39.000 He's got the flu. 00:39.000 --> 00:43.000 I work through his life as well. 00:43.000 --> 00:47.000 The topic that we want to present you today is an open source workflow for 00:47.000 --> 00:49.000 Huawei analysis. 00:49.000 --> 00:53.000 We want to do that by connecting the open source Huawei designer 00:53.000 --> 00:55.000 and the simulator of Urban Mobility. 00:55.000 --> 00:59.000 So this has two open source tools, which we think are really cool. 00:59.000 --> 01:03.000 In research, it's really helpful to work with open source. 01:03.000 --> 01:07.000 Because you can just deep dive into the software 01:07.000 --> 01:09.000 you know exactly what's happening and that's pretty good. 01:09.000 --> 01:11.000 It's also more reproducible. 01:11.000 --> 01:15.000 And well for research, it's also helpful if it's not so expensive. 01:15.000 --> 01:21.000 First of all, let me introduce you to the workflow of Huawei analysis. 01:21.000 --> 01:25.000 So first of all, we have to build an infrastructure. 01:25.000 --> 01:27.000 So there's some kind of infrastructure design. 01:27.000 --> 01:31.000 There will be some switches, some sightings, there will be some signals. 01:31.000 --> 01:35.000 And of course, this will depend on the kind of the policy that we have, 01:35.000 --> 01:39.000 the kind of demand and the operational mode that we want to drive in. 01:39.000 --> 01:43.000 And then on the other side, we have the rolling stock. 01:43.000 --> 01:47.000 So these are our vehicles and they come with certain driving dynamics. 01:47.000 --> 01:53.000 And then if we put the vehicles on the tracks, we can estimate the driving time 01:53.000 --> 01:59.000 and we have the demand. So we know how often we want to have this train driving on that tracks. 01:59.000 --> 02:01.000 And from that, we can create a schedule. 02:01.000 --> 02:07.000 So there will be a departure time and we will can also do a capacity analysis. 02:07.000 --> 02:15.000 So we know how many trains we can follow each other so that there will be a service. 02:16.000 --> 02:19.000 Well, all of this is basically static analysis. 02:19.000 --> 02:22.000 We can do this in theory. 02:22.000 --> 02:26.000 But then, well, if we go to practice, we all know how that goes. 02:26.000 --> 02:31.000 I think either of us has had delays when we are riding a train. 02:31.000 --> 02:38.000 And to kind of take some of the bottlenecks upfront, we can do a microscopic operational simulation. 02:38.000 --> 02:41.000 And then we can add disturbances to our scenario. 02:41.000 --> 02:44.000 And from that, we get operational results. 02:44.000 --> 02:49.000 We can see how resilient our system is and how resilient our schedule is. 02:49.000 --> 02:54.000 Well, it would be really cool if we would have this workflow in open source. 02:54.000 --> 02:59.000 And there's already been some work done by the SNTF with the open source Huawei designer. 02:59.000 --> 03:04.000 And there's a lot of these functions that can be done in the open source Huawei designer. 03:04.000 --> 03:10.000 But we are missing the other the right side of this workflow. 03:10.000 --> 03:16.000 And we thought about gapping this with the sumo the simulator of open mobility. 03:16.000 --> 03:22.000 So what we want to do is kind of combine these two tools for our analysis. 03:22.000 --> 03:26.000 Well, just a really short example on the open source Huawei designer. 03:26.000 --> 03:31.000 I know there has been talks about the open source Huawei designer here and there will be more. 03:31.000 --> 03:34.000 So I would keep this very short here. 03:34.000 --> 03:37.000 We have an example here from his student work. 03:37.000 --> 03:40.000 So he put a bunch of passenger trains. 03:40.000 --> 03:43.000 And then he tried to fit flight planes on this. 03:43.000 --> 03:47.000 And he did this with 160 kilometers an hour for the flight trains. 03:47.000 --> 03:51.000 And then he was able to fit only one or two per hour. 03:51.000 --> 03:53.000 But then he added some sightings. 03:53.000 --> 03:56.000 And he said, well, let's maybe do that differently. 03:56.000 --> 03:59.000 Maybe let's go with 160 kilometers an hour. 03:59.000 --> 04:01.000 And let's just let him drive a little faster. 04:01.000 --> 04:05.000 Well, and then we can build this in serious. 04:05.000 --> 04:09.000 So he was actually able to fit five trains per hour. 04:09.000 --> 04:12.000 Just looking at the capacity analysis and do a time schedule. 04:12.000 --> 04:15.000 Looking at the time schedule constraints. 04:15.000 --> 04:18.000 But then the question is how does this look and practice. 04:18.000 --> 04:21.000 And what if we added disturbance to the system. 04:21.000 --> 04:24.000 And this is what we want to look at in sumo. 04:24.000 --> 04:27.000 So we checked out what sumo could do. 04:28.000 --> 04:31.000 Because simulation is a valuable mess up. 04:31.000 --> 04:35.000 So we can kind of wherever you understand the operational behavior. 04:35.000 --> 04:38.000 We can see the impact of disturbances. 04:38.000 --> 04:43.000 And also we kind of can make changes in a virtual environment. 04:43.000 --> 04:46.000 Have like a virtual playground, a virtual test of our resilience. 04:46.000 --> 04:50.000 To see if we want to do something and play around. 04:50.000 --> 04:54.000 Because we researchers then we can test this in a virtual environment. 04:54.000 --> 04:58.000 And not have this in a real network. 04:58.000 --> 05:00.000 Well, there is software for that. 05:00.000 --> 05:03.000 There's like open track, there's well, there's looks. 05:03.000 --> 05:07.000 There are commercial tools that are perfectly made for this. 05:07.000 --> 05:12.000 So they come with very specific railway modeling capabilities. 05:12.000 --> 05:17.000 And you can use these tools tools to do this task. 05:17.000 --> 05:18.000 But they're licensed. 05:18.000 --> 05:20.000 So you have to pay for that. 05:20.000 --> 05:24.000 And you don't really know what exactly is happening down in the software. 05:24.000 --> 05:26.000 And how things are modeled exactly. 05:26.000 --> 05:29.000 So we want to use sumo instead. 05:29.000 --> 05:32.000 Sumo was developed as a German aerospace center. 05:32.000 --> 05:35.000 And it's covered under the eclipse foundation. 05:35.000 --> 05:37.000 And what it's also. 05:37.000 --> 05:39.000 So that's what we all really like about it. 05:39.000 --> 05:42.000 It's an agent based microscopic simulator. 05:42.000 --> 05:46.000 So basically any vehicle that's moving in the simulation. 05:46.000 --> 05:49.000 You can access all the information and all the paths. 05:49.000 --> 05:52.000 And you can get all that data. 05:52.000 --> 05:54.000 Originally it's been designed for road vehicles. 05:54.000 --> 05:57.000 So it's coming from the car domain. 05:57.000 --> 06:01.000 And it's been used for traffic forecasting, 06:01.000 --> 06:04.000 VTV communication, things like that. 06:04.000 --> 06:08.000 But the good thing about is that because it's from the road domain, 06:08.000 --> 06:11.000 they're used to working with very high amounts of vehicles. 06:11.000 --> 06:15.000 So it's very powerful simulator. 06:16.000 --> 06:19.000 And sumo has added the feature of intermodality. 06:19.000 --> 06:21.000 So it's not only cars anymore. 06:21.000 --> 06:23.000 But we can do a well ways. 06:23.000 --> 06:25.000 We can do trumps and subways. 06:25.000 --> 06:27.000 We can also do bicycles and pedestrians. 06:27.000 --> 06:32.000 So there's a lot of possibilities to simulate the transportation. 06:32.000 --> 06:34.000 So we've been checking out sumo. 06:34.000 --> 06:36.000 And the case study. 06:36.000 --> 06:42.000 And to see how we can we represent, for example, a light real system in sumo. 06:42.000 --> 06:45.000 And for that, I took the corridor A in front of us. 06:45.000 --> 06:48.000 This car is lines U1, U2, U3 and U8. 06:48.000 --> 06:53.000 With the upper part, this is all about surface, the blue part. 06:53.000 --> 06:56.000 And the red part is in the tunnel. 06:56.000 --> 06:59.000 And then in the south trains we were. 06:59.000 --> 07:02.000 And then they go back up to the north. 07:02.000 --> 07:04.000 And I'm ordered two different scenarios here. 07:04.000 --> 07:06.000 So the first one was the conventional scenarios. 07:06.000 --> 07:09.000 Twins are driving in a pig's block system. 07:09.000 --> 07:11.000 And they have a non automated reversal. 07:11.000 --> 07:13.000 And then I tried a different scenario. 07:13.000 --> 07:16.000 It was a new system called communication based train control. 07:16.000 --> 07:19.000 And these trains go with moving block. 07:19.000 --> 07:22.000 And they do an automated driverless reversal. 07:22.000 --> 07:27.000 And I used sumo to see how these different techniques would turn out. 07:27.000 --> 07:31.000 And then I added, of course, delay because that's when it gets interesting. 07:31.000 --> 07:34.000 And well, first of all, about the worst thing. 07:34.000 --> 07:38.000 What we do in the schedule planning, this is just driving from A to B. 07:38.000 --> 07:41.000 And we now, how the passengers take that train. 07:41.000 --> 07:44.000 But then also, of course, we have to do the worst thing. 07:44.000 --> 07:46.000 We have to see how do the vehicles we've heard. 07:46.000 --> 07:49.000 And which driver is on which train in the end. 07:49.000 --> 07:53.000 So making the reverse was always part of the planning as well. 07:53.000 --> 07:56.000 And this is the reverse thing, not in front of a soup on hoof. 07:56.000 --> 07:58.000 As we can see, two different paths. 07:58.000 --> 08:01.000 How the train may reverse there. 08:01.000 --> 08:03.000 And then there's two different kinds to do that. 08:03.000 --> 08:06.000 So that would be the non automated reversal. 08:06.000 --> 08:09.000 Well, the driver actually drives into the reverse thing area. 08:09.000 --> 08:13.000 Get out of the vehicle, walk to the other side, get back in, 08:13.000 --> 08:15.000 and then drive out again. 08:15.000 --> 08:20.000 The automated driverless reversal, the driver actually gets out at the last station. 08:20.000 --> 08:23.000 The train does everything by itself, and the driverless manner. 08:23.000 --> 08:28.000 And then the driver just gets back on when the train returns to the station. 08:28.000 --> 08:34.000 I did a little video of the sumo simulation to give you an idea how that looks like. 08:34.000 --> 08:41.000 So the trains just arrive, they return, and then they go back to the noise. 08:41.000 --> 08:44.000 Well, then I added delay to that system. 08:44.000 --> 08:47.000 In this case, 300 seconds. So that's five minutes. 08:47.000 --> 08:50.000 And then I checked out how the conventional system would do. 08:50.000 --> 08:54.000 And how the CDC system would do if we added delay. 08:54.000 --> 08:56.000 So they were driving the same schedule. 08:56.000 --> 08:58.000 This is the actual schedule that's been driven there. 08:58.000 --> 09:00.000 I got it by a GTFS. 09:00.000 --> 09:05.000 And then I checked out how they could cope with the CDLA. 09:05.000 --> 09:10.000 And we can see that the CDC system can recover from the delay much faster. 09:10.000 --> 09:13.000 It's due to the movie vlog, but also due to the reversal. 09:13.000 --> 09:17.000 We can see the delay is dropping faster. 09:17.000 --> 09:23.000 Then I also looked at how much the delayed vehicle affects the consecutive vehicles. 09:23.000 --> 09:26.000 This is subsequent vehicles that follow behind it. 09:26.000 --> 09:29.000 So this is a 600 seconds delay, a 10 minute delay. 09:29.000 --> 09:32.000 And this is the average delay across all stations. 09:32.000 --> 09:36.000 And we can see this is the first one, the vehicle that has the delay. 09:36.000 --> 09:41.000 And then while the U-1-3, this is the first vehicle which is not affected by the delay anymore. 09:41.000 --> 09:45.000 And again, we have the two different systems that I'm comparing. 09:45.000 --> 09:54.000 And I can see that in the conventional scenario, my delay is much higher for the consecutive vehicles than with the CBTC system. 09:54.000 --> 09:57.000 So we can see also there's an improvement by this new technique. 09:57.000 --> 10:02.000 And this is something that I can model in sumo and that I can get the results from. 10:02.000 --> 10:05.000 And the schedule has been the same for both. 10:05.000 --> 10:09.000 It's just the operational mode has been different. 10:09.000 --> 10:12.000 Yeah, this is also a video of sumo. 10:12.000 --> 10:14.000 This is Frankfurt City. 10:14.000 --> 10:17.000 So I have the Hop-Bacher and the Constable-Bacher. 10:17.000 --> 10:19.000 Anybody has been Frankfurt. 10:19.000 --> 10:22.000 And then there's this great train here. 10:22.000 --> 10:23.000 And it's actually broken down. 10:23.000 --> 10:25.000 So this has just stopped operation. 10:25.000 --> 10:28.000 And we can see a jamming here behind it. 10:28.000 --> 10:30.000 Of course, the vehicles are bunching. 10:30.000 --> 10:32.000 They can't go anywhere else. 10:32.000 --> 10:33.000 There's no sighting. 10:33.000 --> 10:38.000 And then at some point, this luckily, this vehicle continues its journey. 10:38.000 --> 10:42.000 And we can see how the other vehicles then pick up it again. 10:42.000 --> 10:45.000 And how they try to re-establish the schedule. 10:45.000 --> 10:49.000 And then what we want to do is, well, we look at the initial vehicle delay. 10:49.000 --> 10:52.000 But we can also investigate the second day of the delay. 10:52.000 --> 10:56.000 And well, how could things be different if we added a sighting? 10:56.000 --> 11:02.000 Or if we had a different schedule. 11:02.000 --> 11:05.000 So we can check out different possibilities. 11:05.000 --> 11:09.000 And how that influences my dynamic behavior. 11:09.000 --> 11:11.000 So we think sumo is a really cool tool. 11:11.000 --> 11:13.000 We can do some stuff with that. 11:13.000 --> 11:17.000 So we did a little deep dive into what sumo, how sumo works. 11:18.000 --> 11:22.000 So sumo implements a node edges model in XML. 11:22.000 --> 11:24.000 So if edges files have node files, 11:24.000 --> 11:27.000 or I can also combine it to dot net files. 11:27.000 --> 11:35.000 And if I have a track section, that is a bidirectional edge in sumo. 11:35.000 --> 11:40.000 They have a geometry, and they also have a speed limit. 11:40.000 --> 11:43.000 They can also do arc radius via geometry points, 11:43.000 --> 11:46.000 but it's kind of implicit. 11:46.000 --> 11:47.000 Now I have switches. 11:47.000 --> 11:51.000 Well, they basically nodes with several attached edges. 11:51.000 --> 11:56.000 I do have stations, and they can model a dweltime. 11:56.000 --> 11:57.000 Then I have signals. 11:57.000 --> 11:59.000 They are just basically nodes of type will. 11:59.000 --> 12:03.000 So it's attached to the node, not to the edge. 12:03.000 --> 12:05.000 But they are only main signals. 12:05.000 --> 12:07.000 They are no distance signals. 12:07.000 --> 12:12.000 So this is something that is, if I want to compare to a legacy scenario, 12:13.000 --> 12:14.000 this is a bit difficult. 12:14.000 --> 12:19.000 So basically the trains behave as if they had in cap signalization. 12:19.000 --> 12:22.000 I mean for the fast 20, mostly do, 12:22.000 --> 12:25.000 but if I want to do like a commuter train, 12:25.000 --> 12:28.000 they don't have that yet. 12:28.000 --> 12:30.000 And then we have a vacancy detection, 12:30.000 --> 12:32.000 which is also kind of implicit. 12:32.000 --> 12:35.000 Sumo knows the exact position of the train, 12:35.000 --> 12:40.000 and it's freeing the occupancy of the tracks based on that knowledge. 12:43.000 --> 12:46.000 Well, there are some railway specifics, which I'm missing. 12:46.000 --> 12:49.000 So if we want to use this for a railway analysis, 12:49.000 --> 12:51.000 that's the question, how can we improve that? 12:51.000 --> 12:53.000 How can we add this? 12:53.000 --> 12:56.000 And then also if we wanted to use it together with the open source, 12:56.000 --> 12:58.000 where we design a, there's also the question, 12:58.000 --> 13:00.000 how can we map this to the open source, 13:00.000 --> 13:04.000 where we design a to that railway JSON format? 13:06.000 --> 13:11.000 Also about the results analysis in the commercial tools, 13:11.000 --> 13:14.000 but if we're showing you earlier, where this looks, 13:14.000 --> 13:19.000 they have very ready statistics that we turn in the end of the simulation. 13:19.000 --> 13:23.000 There are certain performance metrics that are well established in the railway domain. 13:23.000 --> 13:26.000 So if the railway engineers look at these tooling say, 13:26.000 --> 13:29.000 and now exactly what they get out of that software. 13:29.000 --> 13:31.000 For example, there's like train graphs, 13:31.000 --> 13:33.000 there's speed resistance, 13:33.000 --> 13:38.000 there's always an analysis of the capacity of the headway. 13:38.000 --> 13:42.000 So these are metrics that the railway engineers will look at. 13:42.000 --> 13:45.000 And then we use sumo, we don't have that, at least, 13:45.000 --> 13:47.000 we don't have it out of the box. 13:47.000 --> 13:49.000 So the war data is all available. 13:49.000 --> 13:52.000 We have a lot of XML output from sumo, 13:52.000 --> 13:56.000 and also I have a precise control on any vehicle, 13:56.000 --> 13:59.000 and there's an API, which I can hook myself to, 13:59.000 --> 14:07.000 and then I can get any elements paths, any elements characteristics. 14:07.000 --> 14:10.000 So basically what we need to do is, 14:10.000 --> 14:13.000 what I did for my results earlier that I've shown you, 14:13.000 --> 14:16.000 I did this with Python scripting. 14:16.000 --> 14:21.000 So the information there, it's just not nice and ready. 14:21.000 --> 14:26.000 So you kind of need some programming skills to get this information out of it. 14:26.000 --> 14:30.000 And well, yeah, programming skills, I must have, I'm not a programmer, 14:30.000 --> 14:32.000 I'm from the railway domain, 14:32.000 --> 14:35.000 and so also this is something that's toppling, 14:35.000 --> 14:42.000 and so it would be good to provide some metrics 14:42.000 --> 14:44.000 also for the railway engineers, 14:44.000 --> 14:48.000 which are not that good at programming like me. 14:48.000 --> 14:53.000 So in the end, this is the workflow that we are imagining of. 14:53.000 --> 14:56.000 This is how we would like to use the open source of railway designer 14:56.000 --> 14:58.000 and sumo together. 14:58.000 --> 15:03.000 So we would start our designing infrastructure in the open source of railway designer. 15:04.000 --> 15:08.000 We would add it to infrastructure, we would draft the timetable, 15:08.000 --> 15:12.000 and then we would export this and use it in sumo. 15:12.000 --> 15:15.000 In sumo, we would do the microscopic simulation, 15:15.000 --> 15:19.000 we would test the dynamic behavior, we would look at the bottlenecks, 15:19.000 --> 15:23.000 and then of course there needs to be some kind of results analysis. 15:23.000 --> 15:26.000 So we will look at the delays, we will look at the capacity conflicts, 15:26.000 --> 15:29.000 and how resilient our system is. 15:30.000 --> 15:33.000 And in the end, we want to do then we work in the open source railway designer, 15:33.000 --> 15:36.000 so maybe it would be helpful to add a siding, 15:36.000 --> 15:40.000 or to maybe make some modifications at the timetable. 15:40.000 --> 15:44.000 The left side, I think, is already pretty established, 15:44.000 --> 15:46.000 these tools are available, 15:46.000 --> 15:50.000 but on the right side, the data transformation from the open source 15:50.000 --> 15:51.000 railway designer to sumo, 15:51.000 --> 15:55.000 and also the results analysis, there are still some gaps, 15:56.000 --> 15:59.000 there's anybody who wants to contribute to that, we're happy, 15:59.000 --> 16:04.000 that we work on this to get this connection, 16:04.000 --> 16:07.000 that we can use the both tools and combination, 16:07.000 --> 16:10.000 and close the gap to close this workflow, 16:10.000 --> 16:13.000 and then it's iterative manner. 16:13.000 --> 16:16.000 This is our context, a few, free to reach out, 16:16.000 --> 16:18.000 we'll be happy to get in touch with you, 16:18.000 --> 16:20.000 and have further discussions. 16:20.000 --> 16:22.000 Thank you. 16:23.000 --> 16:28.000 Thank you. 16:28.000 --> 16:31.000 Thank you. 16:31.000 --> 16:33.000 Any questions? 16:33.000 --> 16:35.000 Yes? 16:35.000 --> 16:40.000 You gave an example of changing the operational mode in the performance. 16:40.000 --> 16:42.000 We have one on the transport system. 16:42.000 --> 16:43.000 Yeah. 16:43.000 --> 16:50.000 How you implemented any of these in a practical real transport system? 16:50.000 --> 16:53.000 Well, actually, thank for studying that at the moment, 16:53.000 --> 16:56.000 so they still use the fixed block and the manual reversal, 16:56.000 --> 16:58.000 and we have to talk in 2030, 16:58.000 --> 17:00.000 because then they plan to be done, 17:00.000 --> 17:02.000 and then we can actually compare these, 17:02.000 --> 17:06.000 but I don't have the results of that yet, but. 17:06.000 --> 17:07.000 Yes? 17:07.000 --> 17:08.000 Yes. 17:08.000 --> 17:10.000 You mentioned, 17:10.000 --> 17:15.000 you mentioned the occupied parts of the track, 17:15.000 --> 17:18.000 and you also use this software to, 17:18.000 --> 17:21.000 with my regards, improve the hardware a lot, 17:21.000 --> 17:24.000 so that, for example, adding one or two switches, 17:24.000 --> 17:29.000 you can double the volume of trains on a specific part of track, 17:29.000 --> 17:33.000 or you can come to your waiting time, 17:33.000 --> 17:38.000 and complete normal timing and get a lot. 17:38.000 --> 17:41.000 Well, basically, you can change the infrastructure, 17:41.000 --> 17:44.000 and one two different scenarios, and see how they compare. 17:44.000 --> 17:48.000 You can do that. 17:48.000 --> 17:49.000 Yes? 17:49.000 --> 17:50.000 Yes? 17:50.000 --> 17:54.000 Yes. 17:54.000 --> 17:56.000 It will, however. 17:56.000 --> 17:58.000 I would really love to do that. 17:58.000 --> 18:13.000 Unfortunately, I'm missing the data to do that. 18:13.000 --> 18:16.000 But what we're actually doing is, 18:16.000 --> 18:18.000 we are comparing it to the commercial software, 18:18.000 --> 18:20.000 because at the other level of engineers, 18:20.000 --> 18:22.000 I actually have some work on going 18:22.000 --> 18:24.000 that's comparing it to the commercial software. 18:24.000 --> 18:27.000 I hope they have done some validation as well, 18:27.000 --> 18:29.000 but it will be really cool to do that. 18:29.000 --> 18:32.000 So I listened to the talks earlier about the delays, 18:32.000 --> 18:33.000 and how to gather the delays, 18:33.000 --> 18:37.000 and I will try to maybe catch an event and compare that to my zoom-o results. 18:37.000 --> 18:38.000 Yes? 18:38.000 --> 18:41.000 That will be cool to do. 18:41.000 --> 18:42.000 Yes? 18:42.000 --> 18:43.000 Okay, last questions? 18:43.000 --> 18:44.000 Yes? 18:44.000 --> 18:45.000 Yes? 18:45.000 --> 18:48.000 You know, so, by now, the zoom is responsible for 18:48.000 --> 18:50.000 stimulating or the trains behave, like, 18:50.000 --> 18:52.000 of the x-a-way, this area, and the tap. 18:52.000 --> 18:53.000 Yes? 18:53.000 --> 19:11.520 And what's important to zoom-o xi to are the next 19:11.520 --> 19:14.520 young actual operation of all that stuff? 19:17.260 --> 19:18.980 Well, I know that, the cow world, they do 19:18.980 --> 19:20.900 microscopic simulation of the vehicle, 19:20.900 --> 19:22.340 and then they give out the acceleration, 19:22.340 --> 19:23.780 and they can hook it a sumo. 19:23.780 --> 19:26.220 So it has an external appie, it's called Tracy, 19:26.220 --> 19:28.260 and with that, you can have interaction 19:28.260 --> 19:29.940 and like a co-singulation. 19:32.580 --> 19:34.060 Thank you. 19:34.060 --> 19:35.640 Thank you.