 So my name is Philippe. I'm happy to be here today. Who knows about LIDAR here? Everyone? LIDAR? Okay. For those who don't know LIDAR, you know this, right? This small equipment to measure distance in your apartment. So we have 19 meters here to do to the wall. The LIDAR is exactly the same thing but it acquires one million points per second and it rotates very fast. You put this on a train or on a car and you get the full environment measured in 3D. So that's LIDAR. Today we are going to talk about artificial intelligence too and the topic is minimizing non-quality costs. Before we start we are called the cross product and I have a game for you. At the end of my presentation I will ask you why are we called the cross product? If you can answer this question you win this telemetric LIDAR. Okay the question is why are we called the cross product? So I'll start with this scandal that happens in Spain this year in the 21st century. Most of you must know this. There was 300 million euros spent on designing trains that don't fit the tunnels. So basically you build a train but the tunnels are too small for the train. They spent 300 million designing those trains before realizing they were too large for the tunnels. The secretary of state in Spain for transport was fired. The president of the Renfe is a railway operator in Spain was fired. This is a very big problem. Why did this happen? How do you know if a train will fit into the tunnels? For this you have to do what we call clearance or clash detection. You have to simulate the train into the tunnel and verify that it fits. The way to do it you probably know this equipment. We have several of those around here today. It's called a trolley. It has a LIDAR and it measures the environment. The problem with this equipment you will need four persons to walk the tracks. You will acquire four kilometers per night. That's for 1000 kilometers means a year of walking. The cost will be about 1200 euros per kilometer. Worst thing you have to anticipate two years in advance this operation because you have to intercept the track so there's no train on the track. You have to plan these two years in advance. So the question is this is not a very good performance. Maybe it's the reason why the Renfé did not do it. Probably not. But can we have a better service, a more performing service for doing the same thing? The first thing we can do is use the existing data. This is a French train. I'm French from Paris. This is a French train that has a LIDAR on the top as you can see there. This is a Regal LIDAR and this train will go on and on all the time on the network to acquire systematically all the data. This train runs 200 000 kilometers per year. It runs six times per year on the same track so we have a lot of data. The question is what do we do with this data because this data will look like this. This is the point cloud. So the point cloud has been generated by acquiring one million of those points per second. So what you see here is a tunnel, right? Right? There's a tunnel. There's a track probably somewhere. But you can see it because you are human. Actually it's just a list of points. X, Y, Z. A list of points. A computer cannot see anything here. There's no information. It's just a list of points. So that's where the cross product comes in. So why are we called the cross product? Remember you can win this at the end of the presentation. So what we do is that we are able to find the rails in the point cloud. Automaticity. Fully automatically. And then calculate the axis. Okay? That's basic operation people do on railway. Find the rails and find the axis. Second, we can classify the point cloud. So here you have the tunnel, the ground, and the rails. We'll see a different picture later with more classes. Once we have done this, we can circulate a gauge. So we take a train gauge. Dynamic. So that gauge has rules. It will change its shape based on the train speed, based on the railway curvature. And then you can see where the train will clash. For this you need zero people. Right? This is fully automated. It's an algorithm that does this for you. Just provide the point cloud and we do this automatically for you. We can process this 100 kilometers per day. Actually 1,000 or 1 million kilometers per day. Why? Because it's on the cloud and you know the cloud has infinite resources. So this could be very fast. The cost that's our price actually 200 euros per kilometer. That's our price. And anticipation, no anticipation. You have the data, just process it. So our promise is that for clash detection you can do it 100 times faster and 6 times cheaper with us, the cross product. This was validated by SNCF. SNCF is a French railway owner and help with CEDEC. Now we do a bit more. This is what we vectorize on the point cloud. So you can see the rails, the axis, the platform edges, and the wires. This enables a lot of measurements that can be done automatically on the track. We do the same for highways. So we vectorize the horizontal marking and the safety barriers. One thing we can do once we have the polylines is to reverse engineer those polylines into alignments. Straight alignments, curves, alkylothoids or sparrows. This is purely mathematical and we have algorithms to do this automatically for you. So yet you can reverse engineer your tracks. Interesting topic. This is our most advanced algorithm where we can scan to beam an electrical stop station. Meaning that you provide the scan just a list of points and we calculate the model, a BIM model. Because here we not only incorporate the scanner at this input, but the list of expected objects. In this kind of environment all the objects are known in advance. We know what we are looking for and we have a model for those objects. So we will take the point cloud, classify it, segment it into objects, and then find the best match with 3D models. Last thing we do, clash detection for fire. You know for avoiding fire to propagate you need clearance between forests. And we have done this exercise to verify that you should actually cut trees so that the fire does not propagate.