 If I have one goal today, it's by the end of this, you all will understand what my title means. First to motivate this topic, I'm going to tell you a story. The story is a tale of two batteries, and the two batteries in question are lithium-ion batteries and flow batteries. Lithium-ion batteries, we all have lithium-ion batteries in our pockets right now in our cell phones. We all know what they are. They're a dominant force in a lot of industries today. Flow batteries, I'm curious, how many people know what flow batteries are? Raise your hands. We're a self-selected group of energy nerds, so number of hands were raised, but a number of hands weren't raised, and actually, when I talk to people in everyday life, pretty much nobody knows what a flow battery is. So that's pretty interesting, but let's go back to the beginning. Lithium-ion batteries were invented in the 1970s at Exxon Research Laboratories by M.S. Whittingham and colleagues. Flow batteries were also invented in the 1970s with a lot of work being done at NASA. So both of these batteries were invented around the same time. And then what happened? Well, lithium-ion batteries in the 1980s saw R&D. In the 1990s, Sony introduced the first consumer electronic with lithium-ion batteries, the CAM quarter. This led to a revolution in the consumer electronics. Phones, cameras, laptops, all used lithium-ion batteries. And then as time went on, they started increasingly being used in electric vehicles, starting to transform that market. And they're also already being used as grid-scale energy storage. They're the dominant technology for new installations in grid-scale storage and also into the future. How about flow batteries? Flow batteries in the 1980s saw R&D. In the 1990s, R&D, some niche applications. In the 2000s, R&D and some niche applications. And again, some R&D and some niche applications. So a very different trajectory for these two batteries that were both invented around the same time period. Although, flow batteries, there is a lot of excitement, including in our group, for the use of flow batteries in the future for grid-scale energy storage. But before we get to the future applications, I think it's worth reflecting on these two different battery systems invented around the same time with widely different trajectories. So why is that? Now, you can't pin it down to any one reason, but if I had to choose one reason, it would be this, it would be energy density. Lithium-ion batteries have about five to 10 times the energy density of flow batteries. That's why they're chosen for portable applications. And all of that use of portable applications increased their market share, it increased the manufacturing capability and the investment and it lowered the price and that led to it increasingly being used in grid-scale storage and other applications. So flow batteries don't have those same applications. And so they haven't received that same sort of cost reduction and scale up. So why is this and why are we still excited about flow batteries? To learn about that, we have to look at the architecture, the basic architecture of the two batteries. Lithium-ion batteries have solid compounds that store the charge and they're connected by a liquid that's ionically conductive. Flow batteries, on the other hand, have liquids that store charge. So the charge molecules are dissolved in these liquids and there's a solid ionically conducting membrane in between them. And the fact that these are liquids means that there's a lot of advantages to flow batteries. You can decouple the power and the energy, which means you can create a long-duration battery. You have the possibility for low-cost reactants because you have a much bigger space to play from. You don't have to always use cobalt, for example. And you could have a very long cycle life because you don't have a solid microstructure that's gonna degrade. On the other hand, the solids give you a very high energy density. So a few years ago, our group was thinking, what if we could make a flow battery that could have an energy density near that of lithium-ion batteries and somehow get the best of both worlds? The problem is that's actually really, really hard. And to see that, I wanna look at two different metrics that are critical to energy density of batteries. The first is the standard reduction potential and the second is the capacity. So what you want is a standard reduction potential of your two sides to be very far apart because that means that there's a very big energy difference between your two sides and you store a lot of energy for every charge you pass in between the two sides. And you want the capacity of both sides to be very high so that you can store a lot of charge in both sides. And lithium-ion batteries have both of these. They have a very far apart standard reduction potential so they have a very high voltage and they have a high capacity and that's because you're storing these lithium ions in these solids, the lithium would be green in these crystal structures. So you can fit a lot of lithium-ion batteries into these layered compounds. On the other hand, flow batteries and the conventional flow batteries use water as the solvent to dissolve the compounds that store the charge. They're constrained on both of these axes because you can't have the voltage be too far apart with a flow battery because if you go too far in either direction you either start making hydrogen or making oxygen out of the water. You start water splitting. You also can't have too high of a capacity because you need all of those water molecules around to create the liquid. So you, for just one charge storing compound you have all of these water molecules around rather than being able to fit a lot of lithium ions into this solid framework. So this suggests that in order to create a high energy density flow battery we somehow have to get rid of this water, get rid of the solvent. So how could we do that? How could we create a solvent-free flow battery? There's a number of different knobs you can turn to make a solid into a liquid. You can dissolve it in a solvent. You can heat it up, use a high temperature battery. You can modify a compound. Organic chemists can take compounds and put asymmetric groups, bulky groups, very long chains that are floppy and thereby lower the melting points of compounds. And you can also use eutectic mixing. And that's what I'm gonna talk about. Eutectic mixing is not new science. It's actually very old science. But it's something that we think is not paid attention to really at all in the context of flow batteries. And we think that it's something that actually might hold some promise. So what is eutectic mixing? What is the eutectic system? It's a system where the mixture of two compounds has a lower melting point than either individual compound. And you can see that in this eutectic phase diagram. So we have temperature here. So these lines show the temperature of phase transitions. And then you have composition on this axis. So going from pure A to pure B. So the melting point of pure A is here. The melting point of pure B is here. And the melting point of the mixture is lower than either of them with the lowest point being the eutectic point. So this is called a eutectic phase diagram. The most well-known eutectic is that of water and sodium chloride. And that's why we spread salt onto roads in the winter to reduce the melting point of the ice. So we started working on a eutectic between sodium and potassium. It's literally the combination of pure sodium with the melting point of 98 and pure potassium with the melting point of 64 to see what this looks like. Here's this video. This is a piece of potassium. This is a piece of sodium. Now these are very air and water sensitive. So we do this in an argon filled glove box. But you can see if you just mix them by, you just mash them together at room temperature, you create a liquid. It's a room temperature liquid metal that's being created, which is actually the stable phase at room temperature in this eutectic phase diagram. So we did some research to demonstrate that you could use this in a flow battery architecture if you use the ceramic membrane and we published that in Joule last year. And we're excited by this because if you use NAC, as it's called, sodium potassium, in a flow battery, you would have one side that has a very high capacity and a very low reduction potential. So that would be a good thing for energy density. But we still need something on the other side of the battery. So what will we put there? We were thinking of using quinones, which are organic redox active molecules. They're studied for flow batteries, both aqueous and non-aqueous systems. And they're studied because you can reduce them. You can actually put two electrons on each molecule, either with alkaline metal salt or with the proton to create a hydroquinone. And you can go back and forth. You can store charge this way. Now these compounds are solids at room temperatures with increasing melting points, as you can see. So we had an idea, what if we could mix two of them together and we can create a liquid that had a lower melting point than either one on its own? But the eutectic phase, this phase diagram is not in the literature. We couldn't find it at all. So we'd have to make it ourselves. We'd have to study it ourselves. But if it did happen to be true that we could do this, then a mixture of benzoquinone derivatives would have a pretty high energy density and be up here in this diagram. So it would be great as the other side of a high energy density flow battery. So we studied this with differential scanning calorimetry. That gives you peaks whenever you have a phase transition. And you can see these peaks show that we had a melting point that was lower than either of the two on its own, both through the charge and the discharge states, which means that we do have eutectic mixing going on in these compounds, which was exciting, but we also didn't have as much of a melting point decrease as we wanted. We're still around 100 degrees C for these compounds. So we still wanted to lower the melting point even further. So how could we do that? Actually, using eutectic mixing, if you start adding in more compounds and increasing the number of components in your system, you increase the entropic driving force to make the liquid. And so you can conceivably lower this melting point even further. But the problem with that is as you add more components, you now increase the phase space dramatically. So let's say you wanted to know the exact eutectic composition in this phase diagram by measuring every 3%. So you'd have to make 33 measurements. But let's say we had seven different compounds that we were mixing together and we wanted to know the exact composition of the eutectic by a 3% error, we'd have to mix roughly a million compounds. So that's not gonna happen. We need some model that could help us predict what that composition would be. So we used machine learning. No, actually we used a thermodynamic model. It is actually really a simple thermodynamic model that we applied here. I'm not gonna go over it, but it's a regular solution model assuming immiscible solids and thermodynamics everyone should study. I heartily approve. And the way we apply this is we took seven different compounds with melting points between 44 and 120 degrees Celsius and then we made every binary mixture of them and measured the eutectic melting point from the DSC and we plugged that into the model to get an interaction parameter in the model. And then we can plug that back into the model to get a prediction for an end component mixture. And to see how that looked in the end, if we took a mixture of all seven of these quinones, the model prediction is that the eutectic composition would be this different proportions and the melting point would be minus two degrees Celsius, which would represent a 70 degrees Celsius decrease in melting point from the weighted average melting points. When we measured the equimolar, so an equal proportion of each in the differential scanning calorimetry, we did see a peak at around minus four degrees Celsius. So that shows that the eutectic phase is there at about the temperature we predicted. But we saw a lot of other peaks, which meant that we still had a bunch of solids in the system that this equimolar composition wasn't the eutectic point exactly. But when we went to the predicted composition, we actually saw mainly one peak. There's a bit of a shoulder, so it's not exactly the eutectic composition yet, but it is a lot closer. And that's exciting because both it demonstrates that we can decrease the melting point dramatically by going to a higher number of components, going down to minus four degrees C here, and also that we can get pretty close to predicting what that composition will be just by this simple thermodynamic model with the number of measurements that we made. And that, and then eventually we might be able to use this in batteries. So to summarize, flow batteries have many advantages, but their big disadvantage is their low energy density, and that's because of all the solvent molecules that you have floating around that you don't wanna react in that take-up weight in space. And so what we've been working on is using eutectic mixing as another knob to engineer liquids that are solvent-free. So you just have the pure charge-carrying molecules as liquids by lowering the melting points via eutectic mixing. So I wanna give a few acknowledgments, is that as I do that, I'm gonna show you a video of the quinones being mixed. So these are three different quinone powders that have a eutectic point that's a little bit lower than room temperature. I wanna acknowledge Professor William Chu, he's the PI of all of this research in the material science and engineering department at Stanford, a number of people who have contributed to this research in various forms, formally and presently. ExxonMobil is our current funding source, which I wanna acknowledge and also our previous funding sources. And you can see this liquid being formed. Thank you very much. Thank you, Antonio. Thank you, Antonio. We'll have questions at the end, I think. Okay, all right. So our next speaker is Aisulu Aitbekaba. Aisulu did her undergraduate degree in Kazakhstan, received her master's degree from MIT, and now she's a third year PhD student in chemical engineering. Her research is with the Conelo Group, Mateo Conelo, and focuses on the design synthesis and application of novel materials for catalytic applications. And so today, she'll be talking on nanoparticle-based catalyst for CO2 hydrogenation to hydrocarbons, Aisulu. Good afternoon, everyone. Today I'd like to tell you about my research on nanoparticle-based catalyst for CO2 reduction to hydrocarbons. As you all know, CO2 emissions have been increasing since the Industrial Revolution, and in 2017, more than 40 gigatons of CO2 were emitted. One way to reduce the net CO2 emissions would be to react it with renewable hydrogen to produce high-value products, such as fuels and chemicals. This reaction at atmospheric pressure mainly yields two products, methane through the macination pathway and carbon monoxide through the reverse water gas shift. And among these two products, CO is more valuable because it can be used as a feedstock for other processes. My previous work on CO2 reduction was done at atmospheric pressure using ruthenium nanoparticles supported on Syria. And when we perform this reaction, the major product is methane with selectivity of more than 95%. And you can see it from the chart on the right because CO and methane were the only products formed. What we found is that by oxidizing these nanoparticles at very low temperatures, lower than 200 degrees Celsius, we can re-dispers nanoparticles into highly dispersed species. And these species no longer make methane. Instead, if we perform CO2 reduction over the single site, we make CO with selectivity of more than 90%. So what we found is that by simply changing catalyst pretreatment, we can change structure of our materials and thus get different catalytic activity. But as I mentioned, we are interested in making products beyond methane in CO. We want to make hydrocarbons. I just want to tell you that CO2 hydrogenation to hydrocarbons is usually referred to as modified fishtrop synthesis. Fishtrop synthesis is a process which converts carbon monoxide and hydrogen into hydrocarbons. And this is a well-established process that has been studied very extensively. And iron is one of the catalysts that is used for this reaction. What is interesting is that when CO2 is substituted, is used as a feedstock. Iron can also make hydrocarbons due to its water gas shift properties. Iron also has cheap price, which makes it appealing for this reaction. So it's not surprising that people have been interested in studying iron for this reaction. There has always been interest in how we can improve activity of iron-based materials by addition of various dopants. And specific attention has been paid to ruthenium. Because ruthenium is one of the most active fishtrop metals, its addition to iron is expected to increase conversion of catalyst. And by alloying iron with ruthenium, one can potentially induce different electronic properties which would result in different product selectivity. So if we check the literature and try to understand where the ruthenium promotion effect has been studied before, we can find studies that show indeed by adding ruthenium to iron, one gets higher activity and different product selectivity. We can also find studies showing that addition of ruthenium doesn't change activity or selectivity. And if we try to understand why people come up with different results, one potential reason is how these materials are made. Traditionally, such materials made through conventional methods, which can result in different structures of materials. For instance, conventional methods could result in ruthenium iron alloys being present together with segregated iron oxide nanoboticles, together with isolated ruthenium metals. So it makes sense that if one comes up with a different catalyst structure, one can get different activity. Another challenge with iron-based materials is difficulty to understand this catalyst using exitoo characterization. What I mean by exitoo characterization? It's when we perform a reaction, take catalyst from reactor, bring to a microscope and look at it to try to understand the state of the catalyst and try to relate it to its activity. The challenge with iron-based materials is that iron gets oxidized by oxygen, even in ambient air, so which makes it extremely difficult to relate property of a material to its activity. So a goal of our project was to use colloidal synthesis to synthesize well-defined ruthenium iron heterostructures and use exitoo characterization, basically characterizing materials under reaction conditions to understand whether ruthenium has promotion effect or not. So we made this so-called heterodimers, which you can see on figure A with smaller ruthenium nanoparticles attached to larger iron oxide, and you can see from EDS maps that the presence of two components was confirmed. So once we made these materials, we loaded them on alumina support and started for CO2 hydrogenation at six bar together with other control samples. Before performing the reaction, we reduced catalysts in pure hydrogen at 300 degrees Celsius, and what we found is that when we tested pure iron oxide, the catalyst made only two products, methane and CO, and you can see from CO selectivity axis that this catalyst had CO selectivity close to 100%. When we tested pure ruthenium, it also made only two products with significantly lower CO selectivity, consistent with ruthenium nanoparticles being very good at making methane from CO2 and hydrogen. When we tested the physical mixture, we also got only two products with selectivity somewhere in between, but when we tested heterodimers under the same reaction conditions, we formed hydrocarbons. So we got curious and we wondered, wanted to understand in greater detail what causes heterodimers active for hydrocarbon formation. And this result also elicited importance of proximity between ruthenium and iron oxide. So we brought our materials to SLAC, National Lab, and we were basically tracing oxidation state of iron in our materials during pretreatment in pure hydrogen. What we found is that, as I mentioned earlier, we performed our pretreatment at 300 degrees Celsius. We found that at this temperature, iron in pure iron oxide was mostly present as iron to oxide, and metallic iron can only be produced if we reduce catalyst at temperatures higher than 500 degrees Celsius. If we check the literature, iron to oxide is known to make CO2 from CO2 and hydrogen. So this oxidation state actually explains our catalytic results. But when we did the same with our heterodimers, we found that iron is present as metallic at 300 degrees Celsius in pure hydrogen. So what happens is that when we, what this result means is that by adding ruthenium to iron oxide, we promoted reduction of iron oxide. And this happens where a phenomenon called hydrogen spillover effect where hydrogen gets dissociated into atomic hydrogen by ruthenium spills over iron oxide and reduces it. So this difference in oxidation state of iron at 300 degrees Celsius explains our results. When we took our materials from reactor, we found that our heterodimers no longer looked like heterodimers. Instead, there were causal structures with, as you can see on the slide, with ruthenium being in the core and iron being outside. So what we found then is that when we start with heterodimer and perform reduction in hydrogen, hydrogen gets activated by ruthenium. And as a result of iron oxide reduction, ruthenium gets encapsulated by iron and we end up with these causal structures which are active for hydrocarbon formation. And this phenomenon is consistent with strong metal support interaction effect which was found in 1987 by Tauster. And this phenomenon says that when reducible oxide gets reduced by hydrogen at high temperatures, it partially reduces and covers metal surface. Since we knew that these structures are active, these causal structures are active for hydrocarbon formation, we wanted to know if we can change or increase hydrocarbon formation by changing the thickness of the shell. So what we did, we made these causal structures which was much thinner shell made of iron oxide. And when we performed catalytic measurements, we indeed observed fourfold increase in hydrocarbon yield compared to the original heterodimers. So conclusions from this work is that ruthenium promotes reduction of iron oxide via hydrogen spillover effect. Heterodimers upon reduction transform into causal structures which are active for hydrocarbon formation. And by tuning thickness of the shell, we can change hydrocarbon yield. Our future directions since we found this interesting synergistic effect when metal nanoparticles get encapsulated by metal oxide, we are now interested in encapsulating nanoparticles inside porous metal oxide. And schematically what this means is that we'd like to start with well-defined nanoparticles, deposit them on polymer, make a second layer of polymer on top of the first, basically have a sandwich structure with nanoparticles encapsulated in polymer, and then use this technique called nanocusting to convert this material into nanoparticles encapsulated in metal oxide, such as alumina. So nanocusting is a procedure where a template gets infiltrated with metal precursor and after certain thermal treatment, the original template is removed and its negative replica is made of metal oxide. With this, I'd like to conclude my presentation. I'd like to thank my PI, Matteo Cagniella, Professor, sorry, Simon Berry Group at Slack, our collaborators at Thermo Fisher Scientific and last but not the least, Precourt Institute for Energy for Funding our Project. Thank you. Thank you, Asulev, wake you up. Okay, and we have one more speaker for this afternoon, and this is Mike Howland. Mike is a PhD candidate in the Department of Mechanical Engineering. Under the direction of Professors John DeBerry and Sanjeeva Lele, his research focuses on the optimal design and control and predictive forecasting of wind farms. Mike will now talk on wind farm power optimization through wake steering. Great, thank you so much and thanks everybody for coming. So as was just mentioned, I'm a PhD student in the Mechanical Engineering Department here at Stanford under the direction of both Sanjeeva Lele and also John DeBerry, and today I'm going to be talking about wind farm power optimization through wake steering. And at this point, I'd like to acknowledge funding support both from Stanford University and also the National Science Foundation as well as a special thanks to Transalpha Corporation. So I just want to start with a quick motivation for the continued research and development into renewable energy technologies, and in particular in this talk, wind energy. We already know that wind energy has been deployed at scale and it's been done so effectively and in a low-cost fashion, but a recent estimate stated that just a 4% increase in the efficiency of wind farms in the United States alone will generate $1 billion in annual savings to the United States energy grid. And so as we transition to a low-carbon energy grid, we need to rely on technologies which are both efficient and low-cost and continuing to improve the efficiency of these technologies will help contribute to those goals. And so wind turbines are really efficient devices. When you take a single wind turbine and put them in a field or in the ocean and if it was completely isolated, that device is very efficient in an engineering sense. However, wind turbines are often placed close together in wind farms in order to save on land cost and also in order to share transmission lines and other energy grid infrastructures such as transformers. However, wind turbines are always controlled as if they're completely isolated. So they're not controlled considering that they're in a wind farm environment, but they're controlled as if they're just standalone in a field. And I've sketched what this looks like in a small wind farm where we have six turbines here and I'm showing a top view where the flow is from left to right. And you could see, as I've colored the velocity magnitude in this plot, as the flow is from left to right and every wind turbine is generating what's called a wake region immediately downstream, so immediately to the right of the turbine. And that wake region is predominantly characterized by reduced velocity, reduced momentum, and therefore reduced energy in that wind. And so you could think about it in conservation of energy. Essentially, there's some amount of energy coming into this wind farm and each turbine is extracting some amount of that energy and then there's less available for any potential downwind turbines. And then you could see that each wake is then hitting every successive downwind turbine and thereby reducing the power production that's possible at each downstream turbine. And so in the worst case scenarios of real wind farms that are already operational in the world today, this spacing where you have three to four turbine diameters apart between each other, as we show here, results in an efficiency loss of approximately 40% for that system. And so here at Stanford and in the Lele and Diberi lab group, we've been thinking about ways to actually mitigate these wake losses through operational changes at wind farms. So we're not interested in changing the layout where you would move the turbines away from each other because that would be prohibitively expensive and challenging to do, but instead we're thinking about control. And so there are only two ways that you could do control on modern-day wind turbines to potentially alleviate these losses without doing significant hardware changes. And that's induction control and yaw control. So today we're gonna focus on yaw control. And so as I mentioned, wind turbines are always controlled as if they're completely isolated. And actually the main feature of that is that the wind turbine is constantly turning to face the incoming wind direction. The wind direction is constantly changing in the atmosphere. And what we're interested in doing instead is something somewhat counterintuitive. It's instead to intentionally misalign certain turbines with respect to that incoming wind. And so I've sketched what this looks like in a two turbine scenario where we're again viewing it from the top and the flow is from left to right. And I've actually intentionally yaw misaligned this first turbine by the angle gamma with respect to that incoming wind. And what that actually does is it takes that wake region that I mentioned on the previous slide and it deflects it laterally along this center line, this dashed line. And you can see in this case, this center line is partially deflected away from that downstream turbine. And thereby increasing the downstream turbine's power production potential. And so what we're really interested in is if there's a case where now that we've yaw misaligned this first turbine, its power goes down as a result of the yaw misalignment but the downstream turbine's power may increase and thereby increasing the sum of the two turbine power collectively. And so it's thinking about the wind farm as a system rather than individual turbine. And so to think about how we might actually realize these gains at a real wind farm, we need to talk a little bit about how we model wind farms in this community. And so here I'm showing the most famous wind farm. You almost can't give a talk in wind energy without showing this wind farm. This is the Horns River Wind Farm off the coast of Denmark. And you can see very interesting physics here. The wakes of each turbine are very nicely shown by interesting humidity changes. But essentially we have complicated mesoscale structures in the atmosphere. We have every turbine generating a wake. The wakes are then merging and it's sort of complicated what's happening here. And so to actually resolve all of these physics in this wind farm, we would need to do a large-duty simulation it's called where we resolve all of these scales on a supercomputer simulation which would take on the order of days to months depending on the wind farm. And so if we're thinking about doing operational changes you'd have to then run a new simulation for every time you've changed your operation or your layout or whatever you're interested in studying. And it becomes very quickly computationally intractable. So instead what we do is we parametrize the key physics of that wind farm. So we take those complicated physics and we parametrize it in a so-called engineering wake model where instead of capturing all of the detailed dynamics of that wind farm, we're interested in capturing the main trends of power production at that wind farm. And so this is a sketch of what our specific wake model of interest looks like here. The two key features of this wake model that we've selected is that it's able to capture the velocity deficit behind that turbine, that wake region, but also that wake deflection as a function of the YAMAS alignment. And so here I'm sketching what the wake looks like behind this turbine which is YAMAS aligned by gamma again. So the wake of a turbine which is not being YAMAS aligned would follow this dash blue line. Whereas the wake of this YAMAS aligned turbine follows this solid red line. And then in terms of actually controlling this wind farm, now that we have this wake model, this is the protocol we take. So we take this wake model and then we rely exclusively on the historical data which is in the form of SCADA data. If you're familiar with energy grid systems, SCADA data is used for control by the energy grid operators and it stands for supervisory control and data acquisition, kind of a long name. So we use that SCADA data which is already recorded at every wind farm in the United States and we calibrate our wake model to ensure that we're capturing the trends of the power production at the wind farm of interest. Once we've calibrated our wake model we would then do a control optimization where we pick the optimal YAMAS limit angles for every turbine of interest at this wind farm. Then we would take that to the field, implement that in the wind farm, observe new SCADA data and then feed it back to the wake model in a closed loop system. And so fortunately we had the opportunity to partner with TransAulta Corporation in Canada and study one of their wind farms which is the summer view one wind farm in Alberta. And so I'm showing here on this left plot what this looks like if you look it up on Google Maps. And then here on this right plot I'm showing what this wind farm looks like when you normalize the coordinates by the turbine's diameter which is the relevant length scale of interest. And in particular in this study we're gonna focus just on these six top left turbines in the northwest corner called column B. And so they graciously provided five years of historical SCADA data for every wind turbine at the site in one minute average form. So quite a lot of data. And then we're able to take that data and input it all into what's called a wind rose which I'm showing here. And if you've never seen a wind rose before they're really helpful plots because they tell us two things. One they tell us the dominant wind directions at the site and so that is showing this view on the compass and it also shows us what the wind speed is. And those are the two main factors in determining where we expect wake losses and how bad we expect the wake losses to be. And so you can see from this wind rose that the predominant winds for this site are coming from the southwest. And if you could imagine the flow coming from the southwest you can see that the spacing between let's say this B column and the A column would be quite generous something like 10 or 15 turbine diameters. And we actually wouldn't expect significant wake losses. However you could see on this wind rose that there's a little bump coming from the northwest. And from the northwest the spacing of these turbines when it flows in this direction is actually more like three or four turbine diameters. And as I mentioned in the introduction at that spacing we would expect something like 40% efficiency loss for those six turbines. And so as I mentioned in that control loop the first thing we needed to do is fit our wake model to the historical data. So I'm showing what that looks like here where the blue is the historical SCADA data and the error bars denote standard deviation in that data for that inflow direction, that northwest inflow direction in five to six meters per second. And then the red is the model fit to that data. So it's doing fairly well at capturing the trends of the power production. And again at higher wind speed we can see we do even better quantitatively at capturing the power production of these six wind turbines of interest. So now that we've actually calibrated our model and we can see that we can capture the trends of power production quite well then we need to select the optimal YAMAS alignment angles. And that's actually a fairly challenging thing to do because it's hard to know what would be the exact best YAH angles to pick and if you have a large wind farm it grows unboundedly very quickly about how many different options you have. And so in particular we developed a new gradient-based YAH optimizer which actually utilizes tools from machine learning community. And so it uses the model form of the equations which we've developed and then comes up with a very efficient optimizer to select the optimal YAH angles. And so you could see for this inflow direction for these six turbines of interest with the different colors the optimal YAH misalignment angles being selected for each turbine. And what's interesting about this case is we can see that the power production normalized by the baseline power production for this wind direction has increased in this model prediction by over 20% which is very exciting. And so qualitatively what that looks like returning to this figure where we have that inflow from the Northwest direction and significant wake losses if we YAH misalign the turbines in an optimal way then we could see the wake of each turbine being partially deflected from every successive downwind turbine. And so we actually had the opportunity to test these YAH misalignment angles on the wind turbines of interest in Alberta, Canada. So these are the utility scale wind turbines. So each of one of those has a diameter of 80 meters so quite large. And we have YAH misaligned here each of the first five turbines by 20 degrees clockwise with respect to that incoming wind direction from the Northwest. And then the last turbine is not YAH misaligned since its wake deflection wouldn't benefit any downstream turbines. And so here I'm showing just a few of the field experiment results. And again, I'm showing the baseline SCADA data, so the historical data and the model prediction in this case as well as the experimental data that we observed in this field campaign. And the main features that are interesting qualitatively from this is that we can see that the first turbines power production has decreased as we expect but the downstream five turbines power production has increased quite significantly in this case. And then at higher wind speed for the same wind direction we see the same qualitative trends. And now quantitatively we can see that we've realized quite significant power production increases over the baseline in this experiment where we increase in the higher wind speed case by 13% power production and the lower wind speed case by 28% power production for these actual utility scale wind turbines. And so those are just a few of the highlights from this paper which summarizes the main results recently published on the cover of the proceedings of the National Academy of Science. And the main results I only mentioned the statistically significant power production increase today but we also noticed that through the mitigation of wake losses we reduced the variability in the power production so we reduced the intermittency of the wind farm and that's because wake losses also contribute to unsteadiness in the power production as a function of time. And of course there's always uncertainties in future work and research so we're trying to improve our yaw misalignment controller and we're also interested in knowing more about the fatigue loading so the mechanical loading on the wind turbines when you change the yaw misalignment angle to a non-zero value. And so with that I will take any question I guess we're doing questions together. Oh. Thank you very much Mike. Thank you. That's great. Antonio and Aisler do you want to come up? Three excellent presentations. We have about two or three minutes for questions maybe we can stretch that a little bit longer in the tradition that we have here at the energy seminar maybe students first if they could if they want to ask some questions. Yes. I have a question for Michael. Does your yaw optimization model inform us how to build new wind farms in addition to optimizing current one? Yeah absolutely so with that tool with the engineering wake model you can also think about layout optimization given as a specific to the model. Okay. Okay I have a question for Antonio. I was curious whether you saw whether you see flow batteries being used high energy density flow batteries being used only for great storage or whether those would also be potentially usable in electric vehicles or consumer electronics? Yeah great question. Thank you very much. Hello. Okay. Try to keep my answer short. Maybe. You know I think consumer electronics are a stretch but you know heavy duty transportation you know personally I think it could be possible you know there's cargo ships in China that are now being battery operator although they're carrying coal from one port to another but maybe it could be a flow battery. I have a question for Antonio. So if you were able to find that eutectic point with all of your charge carriers what do you think that the energy density of that would be at the beginning of comparing the looking I am versus the flow batteries? How much would you think that would improve? Yeah yeah great question. So basically if on that plot if we could have a NAC versus eutectic benzoquinone flow battery that's could get you maybe into the 500 watt hours per kilogram range which is actually double what current lithium ion batteries are but that's all theoretical so then you'd have to slap a percentage off of it for all the stuff that the battery is composed of so maybe all in you could get near current generation lithium ion batteries. Now lithium ion batteries are still improving so that I think you know 10 years later I don't think we could outdo lithium ion batteries with the liquid but if you can at least get into the ballpark then I think maybe it opens up these other applications too. Last question I think. Thank you. I have a question to the lady from Kazakhstan. I forgot your name, sorry. If I understand correctly you're reducing the process steps it would take to produce e-fuel or any other kind of migrants and how long will it take to commercialize that kind of technology? That's an excellent question. Well, as I said Fischer Tropsch is a commercialized process and it is used industrially on some scale. For CO2 hydrogenation to become commercialized I think we're still far from making this process feasible in the nearest future because there are many challenges that have to be overcome and the problem not on the inertness of CO2 but also the different difficulty in controlling product selectivities. So in our work we indeed showed that we can make hydrocarbons by reducing CO2 with hydrogen but ideally one would want to make a specific type of hydrocarbon for instance to force all of these and this selectivity challenge has to be solved first in my opinion. All right, well thank you very much. I think we have to wrap up now. So maybe join me one more time in thanking all three speakers and thank you.