Steven Hakupyan | Cycling Aerodynamics Analysis Model Validation

If everyone is ready, we can start uh with the last presentation today. I will uh present my research about cycling aerodynamics analysis. Uh I will start by u giving you a little bit of background how this all came about and what problems we were trying to solve and within what frameworks and methodologies. Um to start with the background uh I will uh talk about two things. The first one is why is it important uh for the industry? Why is cycling aerodynamics information uh important for the field? Um first of all uh when riders go into certain speeds uh particularly higher above 30 km/h most of the efforts that they use goes to uh pushing air out of the way. And uh in this context any uh even small aerodynamic uh gains and optimization may help them to self in save significant time in competitive settings. Uh but also for enthusiasts as well who want to enhance their performance as well. Uh my motivation came in because I'm interested in endurance sports and um actually my very first job that I worked was fixing bikes. Um and everything came on together beautifully to uh help me to uh contribute to this project as well. And the most important thing I think that motivated me was also uh the uh way I I perceive engineering. It is just a way to help explain things that are happening in nature. And this was a perfect opportunity to apply this to a practical field. Um the general pro uh problem that we're trying to solve is to of course enhance uh performance of the cyclists but also uh these intermediate three problems were aimed to solve which one uh first of all uh the first problem was um offering an alternative to high-cost wind tunnels which are the common practice for aerodynamic analysis and also uh integrate real world uh validation um to the study as well and also the add an intermediary step of uh coming up with a uh CAD model of the uh scan to uh also make the workflow reusable and not have to go through entire process every single time. Uh the overall framework was to build a realistic digital cyclist model and uh simulate that model using CFD and also validate those results in real uh track testings. Uh the methodology that was used was um as follows. The first step was the 3D scanning which proved to be the most challenging uh phase and um actually it could at the end divide be divided into lot more uh steps and proved to be the most um complex one. Uh the second one was after the model was available we would go into mesh generation and making the CAD model. Uh from where we would be able to also define the outdoor boundaries in the CAT environment and then be able to transform it to uh the be able to transfer it to the CFD environment and simulate the airflow and get the erodynamic information that we need. Uh with the last step being the validation phase, we would also try to come uh compare do the same experiments in real life and be able to compare the simulation uh results with the uh real life results. Uh a little bit uh more uh deeper um and detailed information about the steps. The first step was the digital uh making the digital twin of the cyclist um using a scanner and it was aimed to uh use a multiple riding positions uh with different equipment to get as much information as possible and the expected output was to get a highly accurate digital representation of the cyclist and the bike. Um the next part was when this post-procession of the scan uh data was done, we would be able to make the uh 3D mesh out of that scan and uh also make the 3D model out of it in the CAD environment to then be able to uh reuse that model and modify the model not do the whole scans uh once again to generate different meshes for different positions and setups. The third step was actually the CFD calculations where the finalized model was imported into CFD software and the airflow was simulated around the model and key aerodynamic parameters were calculated uh including drag force and pressure distribution uh for different cycling positions. And the last step was going to be the field testing and validation. As I already um talked about it, the um test was going to tests were going to be conducted in controlled conditions uh trying to achieve uh the same power, the same speed um and the same track length uh to control and get as much compact information as possible and uh try to achieve an acceptable threshold with the comparison of the model. uh validating the accuracy of the simulation model from uh starting from where we can only uh lean on the model to use it. Now uh about the results and the discussions on what we were able to uh achieve um again the uh vel drone the the real track testing was only going to be available for us uh from March that's why we started with the u mock model and we scanned the human and tried to test the whole u pipeline with that model um and then I will talk about the actual scanner that we use and what limitations we faced uh and uh what alternatives we tried to come up with And um uh after that I will also show you the rest of the worksheet validated with a simpler case. Uh this is the human model trial case. Uh the very first result we got with the scan. Uh where we can see that we have a lot of overlap regions. Uh even the wrinkles on the uh clothes were visible and this was uh also um showed that we we can get a lot of details from the scanner. But it uh also showed us that the post processing takes a lot of time and uh it's not that easy uh to do and it requires expertise uh and at this point we started to face the limitations we had with the uh scans. Uh we used the farm M70 scanner which was available uh in the university and it was tested in various cases uh which was uh mainly practical architectural or object scanning applications. Um and after the very first scan when we were able to go to the valadrome and conduct the scan with the cyclist and the bicycle method, we saw that um we have the first limitation which was uh using the highest possible resolution of the scanner versus the person being able to uh keep the composure and just keep the u pose for a long period of time. That's why we have to find an optimized version of the resolution so the person was uh could realistically hold that position steady. But uh the result showed that even in that case very small movements including even breeding um distor the geometry and was actually captured by the scanner uh making uh that models unreliable for uh future use. Uh at that point we already started testing alternatives and trying to see what else can be done to achieve the final result. Uh simultaneously I was when we were doing the scannings with the farro scanner I was also uh doing the scans uh with my phone using a polycam the polycam app uh which uh can see the result in the middle and um the limitation with that was that uh the app could not uh distinguish the difference between the spoke area and the ground and it projected the spokes onto the ground and uh spokes are a very important and uh vital part of the geometry that contribute to the aerodynamics of the whole system And um the important to have that geometry present. Um another alternative method that we tried was photoggramometry uh which was um uh us taking pictures of the uh whole system u 500 and more pictures from different heights and different angles to be able to capture uh every single detail in the system. And um with all the results this is what we were able to achieve. I will go uh through layers uh with this thing. The very first layer was uh scanning with the three methods the scanner, the photo polycam app and the photoggramometry using the whole system of the rider and the cyclist. Um and that uh the limitation there was that the movement of the cyclist distorted the uh geometry and we were not able to get a complete watertight mesh at the end. Um to eliminate that we tried to uh scan uh the bicycle separately as uh the uh intended use for the scanner was more about uh being an um stationary for stationary objects. We tried to scan only the bicycle to try to get better results and again we did the same with uh polycam and with the uh photoggramometry as well. Um the best result actually visual best result was from photoggramometry. Uh as you can see on the right uh part below we have the point cloud and uh upper part is the mesh. Uh the problem with photoggramometry is that even though it visually gives you better details, it doesn't give you the actual scale and the dimensions of the model and uh we were not able to in that sense also achieve a uh ready model to go uh to with the rest of the capsule. Um one last thing that we also tried to maximize the use of the scanner was um to again scan the stationary bike but using enhanced tricks basically that came from the expertise of the people who helped us with the scanner. Um to explain this I will quickly explain how the scanner works. Um it's a device that fires a laser beam and a rotating reflexive mirror is uh basically uh distributing that beam throughout the room and catching all those point uh points eventually giving us a point cloud. Um even though if you want to for example scan only this u uh table, you have to scan the whole room and then clean up everything else to just have the uh table. But uh to that also contributes to the error of the accuracy for example for this table. To eliminate that what we did uh is we first started with different positions. Uh we started to run a low resolution scan just to identify the angle that uh the for example this table and in this case this bicycle took up in that room and then we ran a much higher resolution scan that was reasonable within the settings uh comparing the data sheet and the actual uh settings that is are shown now there. Uh we uh did four um high resolution scans from different angles to try to achieve um as much detail as possible and this 2x 3x 4x actual u uh sessions show us that each point from the scanner is test checked and tested four times. We actually used 4x and u it didn't uh make sense to go higher in resolution because the scans would take like uh hours from each position and uh they uh after discussing with the professionals we understood that in terms of results it wouldn't give us anything more than we could have achieved and the top left picture is actually the very last scan that we had. Um the problem are are also with u reflexive uh coverings as well. But we tried to eliminate that by applying um chalk or on the materials that were reflective. But uh at the end again um um the spokes and the small geometries around the pedals uh were not uh fully visible. And u those are also things that are very important for contributions. And after discussing with the expert we came to the conclusion conclusion that uh this first part is actually achievable but definitely first of all this is not a oneman job and requires a lot of expertise and um not uh we we should not think about using different methods but also using all of these methods together um at once. So doing both the photoggramometry and the far laser scanning because uh all these methods help each other to uh get the final accurate result. Nefaro in terms of uh size and scale um and photoggramometry in terms of the details uh that we could see here and uh all those contributions even though after using all of these methods uh together we have to still do a lot of post-processing work which requires expertise um in working with these kind of models with a lot of software such as edgesoft netshape reality scan uh the softwares that we used after the mime and cfd and uh a great expertise in CAD modeling as as well to be able to uh reconstruct the missing geometry. Uh to validate the rest of the capstone, we actually took a uh obtained the model of a uh road bike wheel and uh simulated the rest of the process with that model to show that having the um uh model present the rest of the capstone will work. Um this first step is uh the first step of the meshing process. When you have the mesh ready from the scan, you put it into the CAD environment to define the outer boundary. And this big box is basically the air pocket in which uh the model is going in which the air flow is going to be simulated around the model. Um a little bit about the meshing process and how important it is. Um after the CAD CAD model is ready, we transfer it to the mime software which is the meshing software and the geometry reconstruction software. Um important part is that uh in that process we divide the whole geometry into small computational cells upon which the CFT does uh the calculations with the governing equations on each of those computational cells and um in this case as you can see we divide it into boundary families as well and uh the uh how we choose these boundary families is also very important because it basically depends on uh two important things. The first one is the size of the geometry of the part that you're trying to uh reconstruct. For example, the small pressure tube needs to be represented with more uh computational cells to be able to accurately reconstruct the geometry. And the second uh part is uh the intended calculations. For example, if you want to understand how the airflow works on the spokes, you have to again make them uh the computational cells more dense there to be able to uh extract more information about that. And just to give you a perspective, only this uh wheel took 9 million computational cells and on a uh 8 GPU core computer, it took 24 hours uh to calculate the uh to do the whole CFT calculations. And if we had the model, the full model, it would take around 60 to 70 million uh computational cells and just proportionally we can understand how complex the calculation also is. Um and now about the results uh it will be divided into two sections quantitative and qualitative results. The first part will be the qualitative and the visual results that we have. Uh I uh there are a lot more but I uh only will talk about three important ones that will also be very important for uh the cyclist model as well. The first one is pressure. As you can see we have high and low pressure zones here represented by the red and blue colors and uh those actually represent u drag sources. uh and to understand it we have to look at the pressure between two concepts. The first one is um the stagnation pressure which is the pressure where the air flow meets the uh geometry at first and then after meeting that geometry how well the pressure is uh able to recover and if that pressure recovery is low it contributes to the aerodynamic losses. And in case of the full uh cyclist bicycle model, we will be able to uh see which body parts or which equipment um contribute to the uh pressure uh high pressure losses and uh be able to tweak those and understand how we can uh perform uh enhance the performance uh that way. The other uh uh visual um uh basically result that we got is a velocity magnitude which just shows the uh flow direction in in all three directions and the total air flow around the speed and uh what weight and uh uh flow slowdowns appear in terms of the geometry. And for the full cyclist and bicycle model, this will be again an important uh measure because it will show how uh the uh positions and the equipment uh contribute to the uh wake and um if we can achieve a smaller wake, we will uh see that that position is better for uh gaining uh more speed for example or enhancing again performance. And the last part is the uh turbulent to laminar viscosity ratio which just shows in general how the air flow is around the uh object and how um the intensity of the wakes are uh because of this geometry um and how the turbulent the flow is. As you can see in this case we have the turbulence right below the wheel and um trying to make this uh ratio smaller will actually help us to again uh g uh have gains in our dynamic components. The last last important qual quantitative measure was the CDA result uh which uh this is the convergence of the uh drag force calculation uh uh which is around 0.2955 and then using this drag force calculation we were able to get the CDA answer which was around 0.016 016 m squared. And I also compared this to several studies to show that um as you can see the measure is uh very close to the similar studies and it was adequate showing that uh if the geometry was present the uh rest of the CFD worksheet produced reasonable uh results. And as this is a numerical calculation, it is important to um check that all of the variables that were calculated actually converged. And um uh this is done by looking at the residual plot. And as you can see all of these calculations converge at uh uh we did the accuracy of the fifth or sixth uh decimal place. But this does not mean that the result that we got uh is actually reasonable. This is just a uh uh proof that the numerical calculations was carried on uh uh rightly. uh to understand if the numbers make sense, we have to go back to the entries and uh the definitions of the surroundings and everything else understand if the uh number was reasonable. And just to try to conclude everything that I presented here, the initial plan planning stage included uh to obtaining the full cyclist and bicycle model and um we were expected to also get the watertight 3D model out of it and then run the CFD calculations which then would be validated in the field. Um but as uh I already presented it was discovered that the scanning phase was much more complicated than we expected and uh included um a lot more expertise uh to be achieved. But uh I think this is a great starting point uh for future work because uh the rest of the capstone is actually u working and uh someone coming in and trying to with the team uh of course trying to solve this first part will result in a u a great uh uh project because this is actually very applicable as uh we found out um going into the valadrome and um everyone that was seeing us doing the work the kids and everyone they were very interested um already had ideas of what they can test, what they can um what we can scan for them to understand whether those equipment or those body positions were aerodynamically stable uh and uh um efficient or no. Uh thank you. Um and I think a lot u alternative ways can be done for the first step as well. uh one that we discussed was um not by scanning but also trying to obtain the geometry of the model and scan just the person and try to combine uh those methods. But one more uh working method is just as I said to combine everything that we did and um so with some post-processing obtain the watertight model of the uh full cyclist and bicycle system. Uh at the end I would like to these are all the references but I would like to thank everyone who uh contributed to the project because um first of all um my supervisors and then the um Armenian cycling federation because this was a collaboration with them and they were very kind enough to uh help us and give us space and equipment to test and scan and also um to the professionals of Alexi that help us with the scanner and um all the post-processing. uh compared with that. Thank you. And I will be happy to answer any questions that you might have as a cyclist and dynamicist. I'm very interested in your work both uh and uh I I think you uh mentioned in the end and very uh honestly that the scaling actually consume too much of your uh resource whereas you should have done uh with the uh ready available in cat models and uh uh also also you could play a little bit with the geometry variations of uh average cycles to try and >> uh you know establish a base point and then uh go on from there. But uh overall uh well it's it's a very uh everyday problem for every cyclist because when we cycle we constantly feel it uh air drag and uh everyone tries to distinctively obtain a efficient position. Uh but competitive uh they pist as well. Uh but I think uh you have to uh continue this work because it's it's an auto field. It's auto field especially a lot of equipment being optimized with the bicycles being optimized. Uh lot a lot of work can be done here chain drives are being changed the the gear drives etc. A lot of work is there for you. Uh you have to continue this study. Very well done. >> Thank you. >> I have some question. >> Yes. First, uh I wonder why uh Armenian Cyclist Association is interested in this result because um in competitions it is important to uh make um as much of the performance uh enhanced as possible because those uh aerodynamic gains that you get give you uh efficiencies in time and uh in those competitions >> it is understandable but what would they change uh having this result, >> optimize their results basically and >> optimize >> uh yes uh to understand their equipment, how it works, how uh cyclists can make their positions better and etc. >> Okay. Uh second uh uh again it is not clear why to spend this much time on uh the scanning digitizing when there are software with moniking design where you can uh create any posture you want and there is a creoparametric with a mannequin design. They have even Japanese mannequins. They have Dutch monnequins with the different heights and and you can create any posture because there are as many joints as in human being. Uh so why you spend so much time? because it was meant to be personalized for the uh riders at the association at the uh cycling federation. And there are different age groups uh with different equipment with different uh skin suits. And in that case, those mannequins wouldn't work because there are specities that those mannequins don't have. And um being able to have and produce a method, a model using scanning, we could just personalize it for every single person and get more accurate results rather than just using general models. and understanding how can they contribute to overall uh performance. If I may start to answer that question because I was part of the problem statement definition team. Uh our our team tri club each of us cyclists spend like uh 4 hours in average >> uh per person multiple times to tune our bikes and our positions to be able to be >> put a gun. Right now in Armenia that's done manually. It's expensive. The idea was to replace the how do you call it tunnel weight tunnel or manual testing with the software? Of course, we didn't achieve the results but the demand is there. >> Okay. And the last one, what software CTF software? >> CFD++ >> CFD+. >> Yes. Okay, >> any other questions? >> Great job. Good job everyone. >> Ocean, do you have anything to add? Oshim, we uh scheduled in a way so that I vote for the mannequin. >> I vote for the mannequin. >> In order for this project to work, we need a much more detailed uh version of the geology that doesn't change from scan to scan. meaning each time that we scan get a different set of step >> we get a different you know and in order to get a drag data it's almost impossible to do that um so is right I think the way to do this is to actually find the geometries cat geometries of the bikes that are being used and mimic the characters of the riders using a mannequin and then find the drag datas using that rather than trying to get the scan. Now the scan can work but in order to get the geometries from the scan to actually be CFD ready you probably need somebody who has many years of CAD experience which is almost impossible to have in a student at at this level. So it requires a lot of expertise in all of the different stages and I think starting from a viable cab is probably a much faster and viable process. >> Thank you Ocean.