首页  >>  来自播客: The Limiting Factor 更新   反馈

Prof. David Howey: The Charging Curve // + Does LFP 'Like' to be Charged to 100%?

发布时间 2024-05-15 22:02:08    来源


David Howey runs the Battery Intelligence Lab at the University of Oxford. In this interview we do a deep dive into battery charging ...



Welcome back everyone, I'm Jordan Geesege and this is The Limiting Factor. I've been running this channel for about 4 years now and one of the biggest knowledge gaps for me from a software perspective was what determines how fast a battery charges. Now I finally found someone who I think can help shed some light on the topic. David Howie of the University of Oxford who specializes in modeling and managing EV and grid storage power systems. Before we begin, a special thanks to my Patreon supporters, YouTube members and Twitter subscribers as well as RebellionAir.com. They specialize in helping investors manage concentrated positions. RebellionAir can help with covered calls, risk management and creating a money master planned from your financial first principles. Hi David, it's nice to finally have a discussion with you.

I've been looking forward to this interview for quite some time. It's great to meet you, Jordan and yeah, nice to chat about batteries. So let's start with the basics. How is the state of charge and a battery determined in, for example, an electric vehicle? Great, yeah, it's a great question, you know, very fundamental question. I've been telling people to be worrying about for a long time actually since batteries and consumer electronics came together 20, 25 years ago. And I guess the short answer is wouldn't it be great if you could just stick a sensor in and measure the state of charge directly? And in some types of batteries you can, right? So let us battery, you can actually just sample the electrolyte, measure the density of the electrolyte and get the state of charge from that. Back in the old days, that's what people used to do, the very old days. Lithiumine, it's not so straightforward.

And if you've got an application where the current is quite small, what we call the C-rate, I'm sure you know what that is, but you know, the current is pretty small, you're just charging the battery for many, many hours, maybe like your mobile phone. Then to an approximation, you can just measure the voltage of the battery, right? Something a little bit on which lithium-ion chemistry you are using, okay? But as soon as you start doing things with more current, the electric car, maybe you like driving fast and accelerating, get lots of spikes in the current and this causes the voltage to jump up and down and so that method kind of becomes less good and we have to start compensating for all these jumps.

And so, you know, the method we use in the lab, the sort of gold standard method if you like, to measure safety charge, is called Coulomb counting, which again, I'm sure you and many listens will have heard of and this is what it says on the tin. So basically, you count the number of electrons going through your external circuit, which sounds very complicated, but it just means you have a current sensor and you take the measurements of current instantaneously and you just add them up over time. Mathematically, we integrate the current. And why this works is because, you know, in a battery that is degrading very slowly compared to how you are using it second by second, there's a correspondence between the number of electrons going through the external circuit and the number of ions going across the battery internally.

And so it's like bookkeeping or accounting, we can basically, you know, indirectly count how many ions moving across the battery by counting the electrons in the external circuit. And in some ways, this is the only way we can measure the state of charge. So it's actually kind of a, it's kind of a complicated issue because we're always having to kind of operate blindfold if I can put it like that. And a lot of the effort in battery management systems over the last 10, 15 years has been around how do you do this, but in a real application where the current sensor might not be super accurate and you might have a very dynamic driving profile.

And in order to do that, you need to bring models into the BMS. And those models are basically used to correct for errors in the current sensor and noise and stuff like that. Because when you're adding up measurements of current over time, you're also adding up any errors that creep in and that has to be reset, if you like. And so that's the problem with Coulomb counting. You can never measure the current perfectly. So you get this drift issue that you have to correct before. Okay, so there's, it sounds like there's, there's a gold standard way of measuring the state of charge, but so you can do it through Coulomb counting or voltage. What I'm picking up is that you're always, it's always a best estimate and a best guess as to what that state of charge actually is in reality. We kind of have to use these other variables as proxies to understand what the state of charge is.
为了做到这一点,您需要将模型引入电池管理系统中。这些模型基本上用于校正当前传感器和噪音等方面的错误。因为当您随着时间累加电流测量时,也会累加任何可能出现的误差,这些必须被纠正。 这就是库仑计数的问题。永远无法完美测量电流。因此,您将面临这种漂移问题,必须在之前进行校正。 好的,听起来有一种衡量电池电荷状态的标准方法,您可以通过库仑计数或电压来实现。我理解的是,您始终需要估算和猜测电池的电荷状态。我们必须使用其他变量作为替代品来理解电池的电荷状态。

Yeah, that's correct. Without getting into too much detail, you know, what we're trying to measure is the state of lithiation of the electrodes in a lithium ion battery. And we could say that voltage is a way of measuring that, but only at equilibrium. And unfortunately, and what that really means is we have to discharge the battery really slowly under very controlled conditions charged very slowly. And by slowly, I mean really slowly, you know, maybe over 10 hours, 20 hours, 50 hours, something like that.

So it's just not practical outside of lab to do these very, very slow measurements and count all the Coulombs in a very sort of controlled way. And that's why we need all of these engineering approaches to kind of bridge the gap between that and something that actually works second by second in a real vehicle. So for instance, an electric vehicle, I've seen a number of different ways reported to estimate the state of charge. What is typically the way that they estimate the state of charge in a vehicle rather than say in the lab? Is it the voltage or?

Yeah. So what you do, well, let me just take a step back. So you could argue that in a phone, you could measure the voltage, right? Maybe for various reasons, maybe you'd go beyond that, but you could argue that it's slow enough charge, sorry, discharge, not charge that you just measure the voltage. In a vehicle, what you do is you would do this Coulomb counting technique. So you would have a current sensor, you measure the current from the pack, you'd add that up second by second over time. You need to know the initial condition, you need to know the starting point, otherwise you know, is it full or empty? Where are you starting from? It's all relative to where you start.

But then what you do is you would also be measuring the voltage. And we know that although the voltage doesn't absolutely, accurately tell us the state of charge, it's in the right territory, right? I mean, a lithium-ion battery voltage, okay, let's take an MC type chemistry. It's very, you know, it's very between say 2.7, 4.2 volts, something like that. And so if the voltage is nearer to 4 volts, you'd expect the SSC to be higher and vice versa.

And so what we do is we build models, which basically tell us what the voltage drops inside the battery are. And then we can take the measure voltage and sort of run it through a model and work out what the sort of theoretical open circuit voltage would have been under no load conditions, everything equated. And then compare that to the state of charge that we get from counting up the current. We sort of bring those two things together in a clever way. And that gives us kind of the best blend of the two things. And that also allows you to reset the errors from the coolant counting. Does that make sense?

Yes. I won't dig any further because it sounds like there's a lot of depth and there's a lot of places we could go with that. But I think that's a good explanation for now. For me, it raises the question. Different batteries have different voltage curves. For instance, LFP has a very narrow voltage range and a high nickel chemistry has a very broad voltage range. So do each of those pose their own unique challenges or is one easier to deal with than the other?

They absolutely have unique challenges. And you've hit the nail on the head that LFP is challenging for SSC estimation. And the reason for that is because the voltage is quite flat as a function of SSC. So at higher or low voltage, you know, you get some variation. You have this kind of sigmoidal shape in voltage versus SSC. So if you go to the very top or the very bottom of the curve, great. You've got a good sort of indication of SSC in terms of the voltage. But if you're somewhere in the middle, say between 20 and 80%, that's a problem. You know, the voltage is super flat. It's great if you're a power electronics engineer and you're building a power converter because you've got a nice constant voltage power source. That's what you want.

But if you're trying to work out from the voltage, what the SSC is, there's a huge ambiguity. The voltage changes are really small. And there's an additional problem with NFP, which is something called hysteresis, which is similar if people have heard of magnetic materials, maybe you remember from school physics or whatever that you get, this BH curve for a magnetic material.

And what it comes down to is with lithium ion phosphate, you get a different voltage depending on whether you're charging or discharging. So there's a gap between the charge and discharge voltage at zero current. So this gap kind of doesn't go away, even if you take the current away and wait for a while. And you absolutely have to take this into account. It makes quite a big difference. And if you don't take it, if you just assume you have one voltage, it's going to be massively ambiguous. So you are in the mid range.

Yeah. All right. And you mentioned magnetism there. I was very weak in physics when I was in school. I was more biology. But does that have to do with because LFP is a magnetic material? Is that what? No, no, sorry. Yeah, apologies. That's not the situation. I was just, it's just analogous to the way that magnets behave. You have this thing called hysteresis. But so put magnets out of your mind. It's not related to magnetism. But what it really means, if I could draw a picture for you, what you would see is that I get a different relationship between voltage at very slow rates, open circuit rates, depending on whether I'm charging or whether I'm discharging. So I kind of go up one curve when I'm charging and then I come down to slightly different curve when I'm discharging. And there's a gap between those two curves. So if I gave you an LFP battery and I said, I didn't tell you anything about how to be used. But I said, it hasn't been used for the last few hours and you did a voltage measurement and I asked you what's the state of charge. It would be very difficult for you to tell me absolutely accurately what that was because you don't know whether it's been charging or discharging.

So you don't know which of these two curves you're on. And so this adds a complication because then the battery management system has to keep track of the history and basically use, in a clever way, some historical data about whether you come down or you're going up. And you can imagine if you don't do a full charge discharge, then interesting things happen. So say I go to 50% and then I discharge by 10%. I get like a small version of this gap in between.

So it makes a little sort of eye shape. You know what I mean? Yeah. You know, I understand people are not watching. You have to imagine basically I'm plotting with my hands on the vertical axis of function which is open server voltage on the horizontal axis, it's state of charge. And one curve which kind of goes up a bit and then along and then up again and I have another version of that curve and there's a gap in the flat region in the middle of between the two curves. And what I can do after when I publish the video, I can include an image.

So if you have a good chart of that, I get it when I put the video together, absolutely. Great. And you're saying that's something that's more of an issue with LFP than with a high nickel chemistry? Is that right? Yeah, yeah, absolutely. So there's really two issues with that. One is this hysteresis issue. So I need a different voltage curve depending on whether I'm charging or discharging, even it really very slow rates. So this gap remains even if I remove the current. And the second issue of LFP is just the fact that the change in voltage is very flat as a function of state of charge between sort of 20 and 80%.

So that means I don't get much information from the voltage about where the SSC is. All right, so if you were a hardware engineer, you'd like the fact that LFP has a very flat voltage curve. But if you're a software engineer, it's more of a challenge because you have all these other variables, all these kind of demons in the machine. Yeah, yeah, yeah. And then with a high nickel chemistry, the software is easier to look after because there's not these strange variables going on to where you have to do more work to estimate the state of charge. But the hardware engineers might not like it as much because there's this huge difference in voltage over the course of charge and discharge cycle. Is that right then? Yeah, yeah, that's a great summary, Jordan. Yeah, absolutely. So your power converter is going to have to be a little bit more carefully designed for a wider voltage range in a chemistry that has a wider voltage range. And actually going slightly off piece from this discussion, a good example of a challenging chemistry there is sodium ion because there's quite a wide voltage range.

Another good example is that there's a supercapacitor because if you really want to extract all the energy from a supercapacitor, you have to go down to zero volts. And so it's hard to make a DC or a DC to AC converter that has a huge voltage range. So at the end of the day, if I have a supercapacitor connected to a power converter, there'll be a lower voltage limit beyond which I don't want to go any lower. So I'm going to miss out on some energy at very low voltages. All right. And I actually had a question about sodium ion earlier and that answers part of it.

So on the hardware and sodium ion has challenges because there's a broad voltage range. Now somebody also mentioned to me that it besides the broad voltage range, it also has a high voltage hysteresis or a high hysteresis. Yeah. Correct? Yeah, that is right. So you're sort of getting the same problems you had with LFP with sodium ion. Yeah. Correct. So sodium ion, it's LFP and nickel have positives and negatives where sodium ion in both cases, hardware and software, it's more difficult to deal with, but not insurmountable, just extra challenges. Yeah. I think that's a good summary. So with sodium ion, there's definitely a voltage hysteresis and you need a fairly wide range of voltages for your power converter to really get the most out of the cells. I mean, these are sort of solvable problems if I can say that. They make our lives more interesting as researchers. Oh, sorry.

One of the things that's interesting from a sort of scientific point of view is why does this hysteresis happen in these different chemistries? And is it that we could sort of think of hysteresis in two different ways? Is it a thing that is always there? So if I remove the current and wait for forever, you know, does it stay or not forever, but you know, for a week, does the voltage gap stay there? Or if I wait for a day or two, does it gradually come back to the middle? And I think some materials like graphite, it could be more like the second one, right, where if you wait long enough, it will kind of equilibrate. Although on a practical level, that long could be quite long, right? And actually, so we've seen some data where we can get hysteresis from graphites and it gets worse at lower temperatures as well. So actually, I think some of these problems would manifest even if you just had like NMC graphite cell, there's a little bit of hysteresis from the graphite, at least on the time scales that we're interested in when we're driving our cars over a few hours, right? Yeah, we could wait two days, but I'm not sure we're going to do that just to get a BMS to be accurate. Yeah, we have practical real-world application. Yeah.

Now, when somebody charges their electric vehicle, so let's move on to the next aspect of this, we've just looked at how the state of charge is determined and the variables around that, which is we could do an hour on that. But the next factor is when somebody charges their vehicle, there's a charging curve that determines how fast the vehicle charges. Can you explain how that charging curve is created and how the vehicle determines the charge curve when it charges because that charging curve can change under different conditions. So how are those, I guess, that set of charging curves that manages battery charging, how does that make? Right.

Yeah, it's a great question. And I should start by saying, you know, lots of companies out there who are innovating in this space. So I'll try and give you a broad outline of what I think is happening, but there's lots of things going on that are quite exciting. And I'm sure some companies are doing stuff that I don't know about. So I might not say what is cutting edge. If that makes sense. But okay, the first thing we should appreciate is that there are two sort of major sources of constraint, right? So one is how powerful is the power electronics that's connected to the car? So if you drive up to a supercharger or a non-Tesla kind of charger, you know, that is 50 to 100 to 150 to 200 kilowatts, there's going to be ultimately a power limitation in the electronics in that charger. It can only deliver a maximum power of some number. And that means even if the battery could take a lot of power, at least if you start from zero SSC where the battery is, that's the kind of happy place where the battery won't state lots of power on a warmish day. The first limit is going to be just the maximum power limit of the charger, right?

And then the second set of limitations come from the battery pack itself, okay? The other thing I should say, and I'm sure you and everyone listening would know this, but you know, there are sort of two types of charger. There's onboard charger and offboard charger. So if you're charging a car at seven kilowatts at home, that's actually using a power converter that's on the car. Whereas if you're charging it while driving across country, then the charger basically plugged directly into the DC bus and the electronics for doing AC to DC is outside the car. Yeah, I think we all know that. So then how do we determine the charging curve?

Well, we need to think about what limits the power in the battery, okay? And so there's limits come from different sources. One might be a thermal limit. We don't want to get the battery too hot. So if it's a super hot day, we're in a hot country and the battery is generating a lot of heat, maybe because its internal resistance is a bit higher, whatever, which could be for lots of reasons, then the charging power might need to be limited to avoid overheating. The cooling system in the battery, which would normally keep a cooler on a hot day while you're charging, is struggling.

But having said that, in general, warming the battery up helps with getting more power in, okay? And the opposite is true. If it's really cold, it's hard to charge. And I think we know this. I mean, if any of us have been skiing or snowboarding and it's been super cold, your phone battery might have died or whatever. So that's one first rule is thermal. So batteries are like it hot, but not too hot. If it's below 5 degrees centigrade, the internal resistance of battery goes up a lot and this starts to cause a limitation with respect to charging. And actually on this thermal point, there's been some interesting innovation done where people have proposed putting heating plates directly inside the cells and then kind of heating up the batteries to 50 plus degrees C in order to get fast charging to go better. Basically there's a trade off there because if the battery is then really hot and you've charged it really quickly, you then need to wait for it to cool down before you drive away. So you have to think a bit more globally. But anyway, so thermal is a limit.

But the other limit is what we call lithium plating. Okay, so this is to do with driving the voltage on the anode, so on the negative electrode, the graphite with respect to, this is super technical, but with respect to a zero reference, which is defined by lithium electrode. Okay, I should say that for the science people in the room. Driving that voltage below zero results not in the intercalation reaction, which is the reaction we want, which is why we want to reversibly move lithium in an hour of the graphite, but instead it plates metallic lithium on the surface of graphite. And this is a problem.

This is not what we want. So in lithium ion battery, normally lithium ion battery, so I'm not talking solid state or anode free or any of that stuff, just normal lithium ion battery. We never want the lithium to be in a metal form. We always want it to be in a compound, right? So it's either going to be intercalated into the graphite or it's going to be in the metal oxide on the other side. But yeah, so if we get the situation where we drive the voltage in such a way that it goes below zero volts with respect to lithium on the graphite, then we get lithium plating and that's bad because we can build up, well, we can build up little things that can puncture the separation inside the battery. And there's a whole host of other ways that that can go wrong. Okay, so we want to avoid pushing the current too high or the temperature too low in such a way that it would cause this to happen. So on a practical level, what this means is that as we get to higher states of charge where the voltage of the graphite electrode is lower, we have to be more careful. And as we get to higher currents, we have to be more careful. As we get to lower temperatures, we have to be more careful.
这不是我们想要的。所以在锂离子电池中,通常情况下,我说的是普通的锂离子电池,不是固态或者无阳极之类的。我们永远不希望锂以金属形式存在。我们总是希望它以化合物的形式存在,对吧? 所以它要么会插入到石墨中,要么会在另一侧的金属氧化物中存在。但是,如果我们的电压以一种方式驱动,使得它相对于石墨降至零以下的情况出现,那么我们就会出现锂沉积,这是不好的,因为我们可能会堆积一些小东西,可能会刺穿电池内部的隔离层。这还有很多其他出错的方式。所以,我们要避免将电流推得太高或温度过低,以至于会导致这种情况发生。因此,在实际操作中,这意味着随着我们达到更高的充电状态,石墨电极的电压更低,我们必须更加小心。而且随着电流增大,我们也必须更加小心。当温度降低时,我们也必须更加小心。

So just to recap, so we can generally put lots of power into the battery if the temperature is ambient too high, but not too high. And if the current is, sorry, and if the state of charge is low, but if the temperature is low and the state of charge is high, we have to be a lot more careful about limiting the current. And you can solve this almost as an optimization problem, say like, what is the best path for the current to follow as I charge the battery up, as I follow an SSC?

And that's kind of going to dictate the power curve. And what that optimization problem will say is charge with very, very high power for the first few minutes, and then gradually drop it down in order to avoid this lithium plating happening. Now, that very, very high power might then be limited by the power of the AC to DC converter. So yeah, so you can kind of think of there being a region where we're not really limited by the batteries for the first 20, 30% SSC from zero, was just limited by the power converter.

And then after that, we start to ride along this constraint dictated by lithium plating. Does that make sense? It's kind of a drop to 80. That makes perfect sense to me, because just looking at a few of the variables here, graph height has a slightly different voltage than like pure lithium on the anode. If you push the charging too high, rather than jamming that into the graphite anode, which has a slightly higher voltage, then it starts plating pure lithium onto the anode. Is that correct? Yeah, yes. Yeah.

And you could say that there's a voltage where intercalation wants to happen, and there's a voltage where plating wants to happen. And when you get one or the other is dictated by what you do to the electrode in terms of voltage, it's not. All right. So there's, so we have those variables, the voltage and how hard you're pushing that to charge the battery.

And then you also have the temperature, which it's almost like a, it almost creates a speed limit or it's almost like the interest rate or something like that, where it's like it affects the fundamentals of the system, how it works. Yeah. Yeah. What's really going on there? The temperature is a slightly indirect effect because really the temperature limit is still a lithium plating voltage limit. But what's happening is because the battery's got internal voltage drops.

If you, so I'm an electrical engineer, so I think of it like the battery's like a voltage source and a resistor inside. And you can kind of think of that for each electrode for the graphite and for the NFC. Now, the resistor value is a function of temperature, and it's much higher resistance at low temperature and low resistance at high temperature. So what that does is it has the effective at low temperatures and high currents, causing a lot of internal voltage drop across the resistor, if I can put it like that, which drives the potential at the graphite surface in such a way that you get a little implating. Does that make sense? Yeah. Yeah. So this is, well, let's tie this together then.

So we have estimating the state of charge is one moving target and then estimating exactly when lithium plating is going to occur and at what temperature is another moving target. Is that correct? Those are, we have two moving targets there. Yeah. Okay. So how does a battery manufacturer determine when lithium plating is going to occur at one temperature? Do they just have to test hundreds and thousands of cells under different conditions to find exactly how those cells behave? Is that how they gather that data? Basically yes. Yeah. Lots of testing and quite a bit of increasing amounts of modeling. So yeah, basically get lots of cells from suppliers, test them at different voltages, currents, temperatures, and then often, or usually you do some destructive tear down analysis as well because you can kind of look at the surface and see whether there's been anything plating, expose it to lots of different diagnostic techniques and so on.

And then, you know, some iteration back to non-invasive measurements that you can do in the lab or on voltage and stuff like that as well. But yeah, there's no substitute for lots of test data unfortunately. All right. So what they do is they, my understanding is they gather all that test data and look at because each cell is actually slightly different because of manufacturing inconsistencies and inconsistencies and things like that. They take the best data they have, they form a table.

So when your battery is deciding to, when you decide to charge your battery, my understanding is there's a lookup table you go, the battery management system goes, all right, under these conditions, this is what should be a safe charge curve or a safe charge that charge the vehicle. So is that correct? Yeah. I think so. Yeah. And you, I mean, you could probably solve an optimization problem in real time if I could put it like that. But assuming that the battery's condition is not changing rapidly, which it shouldn't be if it's a nice well-designed commercial stable long-lit battery, there's no point in my opinion doing that. You've just solved that problem beforehand and then store it in a lookup table like you've just described with some safety buffers and so on.

I think where there's still quite a bit of uncertainty and kind of interesting research to be done is around, well, okay, what happens after five, ten? 15 years. You know, we know that batteries age, they degrade, it might be a slow process. Generally, I think it's a slow, slow process. But that's going to shift the goalposts, right? It's going to shift the behavior of the electrodes.

And another thing that happens when I haven't mentioned at all, but it's really important is that all of this lithium plating and stuff is not homogeneous. So it's not uniform. There could be variations from particles to particles and across the cell and maybe from cell to cell within the pack. And so it's really important that manufacturers are able to get a handle on that non-uniformity issue. And again, modeling testing helps with that.

But there's things like the position of the tabs in a pouch cell will have an impact on where the current is going and therefore some parts of the battery might warm up more and some will warm up less and the SSC will be slightly different in some regions compared to other regions and so on. So it's a really complex kind of spatially distributed problem within a cell. That makes sense.

And so the lookup table that you've described is going to be your sort of best guess at the average with some safety buffer. All right. And we'll look a little bit more at the hardware side of things in a second. Before we look at that, is the discharge rate, I know the charge rate is determined by these curves, is the discharge rate also governed by these things?

Right. So discharge, you're not worried about lithium plating. That only happens when you're charging. So that's good. So with discharge, we're more worried about things like thermal management, where the batteries can overheat, which to be honest, it's only an application where say you need to discharge our battery in five minutes is that a problem.

So maybe if you've got an un-introductable power supply specifically designed to do a five minute discharge from 100 to zero, then you might worry about overheating. But I think generally speaking, and someone will probably write in and say I'm wrong, fair enough. But generally speaking, we're not so worried about the discharges, we are about the charge.

I mean, another way of thinking about it is when you're charging the battery, you're kind of pushing everything in the direction it doesn't want to go. And so it's a bit more fragile and dangerous. But when you're discharging the battery, it wants to discharge. So it's a bit more well-behaved. That's a really hand-wavy explanation. But I hope you can make sense. That makes sense.

So let's tie this together with a few other different variables. When you're designing a battery cell, you take into account what type of characteristics you want out of that battery cell. And those include things like cost, cycle life, and energy density. There's this balance of things that you have to take into account. Does that also happen on the software side? For instance, could you discharge a battery, or could you set up the charging curve or the lookup tables for a battery in such a way that the battery would charge twice as fast? You would get about half the life out of it.

Right. Yeah, I see what you're saying. I mean, first of all, yeah, I completely agree that there's a kind of a trade-off between how hard you're pushing the battery, how long it lasts, and cost. And there's kind of good reasons for that. We don't have time to go into detail. There's some nice articles on the internet. But yeah, absolutely, you're right. So I think we see a good example of where this maybe plays out, if I can say, is in grid storage systems.

So we'll come back to electric cars in a minute. But in grid storage systems, I'm so this is stationary batteries basically plugged into the power grid. So I'm buying a large battery. In fact, we have one down the road from here. Well, a few miles down the road in Oxford, here where I'm based, which is 50 megawatts, so it's big. It's like a field full of containers full of batteries plugged into the transmission grid.

And what that battery does is it provides services to support the power grid. And those services range from just charging, discharging when the energy is cheap or expensive to make money or supporting the frequency response on the grid and so on. And if I'm an investor building one of these big batteries, you know, this is a tens of millions of dollars, tens of millions of pounds project, I need to know what's the return on my investment. And that's going to end a lot on two things.

The revenue that the battery makes and the lifetime. And so this trade off that you've just described, absolutely critical for that kind of application. And I have this difficult choice. Do I sweat the asset? Do I push the battery quite hard? Maybe cycl it three times a day or something, but degraded more quickly and then have to, you know, miss out on future revenue. I can put it like that. Or do I do it quite slowly? It lasts a long time, but I miss out on present revenue. And I think there's a sweet spot in the middle. There's a kind of octopus between those two extremes. We've done some work on this recently. So if anyone's interested, I can send you a paper. But I think this plays out to a similar extent with electric cars, although it's a little bit different with electric cars because you're not charging and discharging to make money per per say. You might want to control the charging so that you get this sweet spot correct. So you don't fast charge like four times a day. That would be a bad idea.

But most people I think are not doing that. But the discharge is not really, you don't have control. That just depends on driving. And so the battery just has to do what the driver needs it to do. Where I think it gets interesting is where we start to think about things like vehicle to grid because then your electric car is becoming much more likely grid battery and providing these kind of bidirectional services to the grid. The other thing just to finish the point is different chemistries may be slightly different in this regard. And also different variations of the same battery.

So I can actually optimize a battery for power or for energy. For example, I can have thin electrodes or thick electrodes. I can change the thermal management system. And it might cost me money to do that. Of course, more cooling, more expense, but maybe I can do more cycling without degradation. So when we talk about this three way trade off that you mentioned, we should definitely think about the system level as well. And how much do I want to throw out my HVAC system to keep a battery in the sweet spot firmly? Yeah. So you say, for instance, you might have the same battery and you could use it for grid storage or an EV and you might tweak the charge curve depending on which one of those use cases and what your priority is.

If you want that battery to last a lot longer, you might slightly change the charge curve as opposed to the way you'd manage that charge curve in the vehicle. Yeah, 100% definitely. And a really good example of that is I think if you were using your vehicle for vehicle to grid, you would limit the SSC window. So you'd probably bounce around 50%. I mean, it depends a little bit on the exact chemistry, but we know that if you hold a battery out, extremes of voltage, typically at high voltage, that's bad from a lifetime point of view. And so, for instance, I think vehicle to grid is actually a really good idea. I mean, I have an electric car. I also have a grid battery at my house because I'm a battery guy.

My home battery, it's much smaller than capacity compared to my electric car battery. It doesn't make sense, especially at the moment. I mean, we're in springtime here in the UK. I have solar panels on my house. Amazingly, we're still covering most of our energy from solar, even in the UK, right, where there's not always that great. And so we just need a few kilowatt hours here and there for the next six months to store our home energy. I mean, we're essentially off grid, right, which is pretty exciting. And it's sort of silly that I've got two batteries, one in my car and one in my house. I should just be using the mid 10, 15, 20% range of my car battery when it's plugged in and out.
我的家用电池,与我电动汽车电池相比要小得多。这在这个时刻并没有什么意义,尤其是在英国的春天。我的房子上安装了太阳能电池板。令人惊讶的是,即使在英国,我们仍然可以用太阳能覆盖大部分能量,对吧?这并不总是那么好。因此,在接下来的六个月里,我们只需要储存一点点千瓦时的家庭能量。我是说,我们基本上离网了,这真的很令人兴奋。而且我在车上和家里各有一个电池,有点傻。当我插上车时,我应该只使用我的车电池的中间 10、15、20% 范围。

Yeah, I know there's a lot of people looking forward to using the battery packs in the vehicle as well. Fully utilize it as an asset. Yeah. But in order to do that in such a way that you don't degrade it, you have to very carefully manage the SSC range and keep it quite limited. You can't just swing from zero to 100 and back to zero. I think that's probably a bad idea. Yeah. So moving along, Elon has said that LFP batteries want to be charged to 100%. But my understanding is that LFP batteries at it's a bit more complicated than that. And at 100% state of charge, just like other batteries, LFP batteries do degrade more quickly. They don't degrade as rapidly as say like a nickel battery minded 100%, but they still degrade more rapidly at 100% than like say 70%.

So from a cycle life perspective, you don't want to be charging to 100% all the time. However, it seems to be there's a tug of war because in order to accurately measure the state of charge of the battery, and for those battery cells to balance within the, to top them all up and for the system to get an accurate state of charge, you have to bring that battery pack up to 100% occasionally. Yeah. So my question is, where is the happy like because of that tug of war, where's like how often should people charge their battery to 100%? And if they don't charge their battery to 100% will that cause degradation by the cells getting out of balance? Does that make sense? Yeah, it does. So I think you're saying you need to go to 100% for SOC calibration based on all with everything we discussed earlier, hysteresis, flat voltage curves, but you don't want to go to, you don't want to stay at 100% all the time.

Yeah, because it will degrade the battery more quickly because there's a certain side reaction called solar electrolyte interface, which forms on the graphite. And this is accelerated at high voltage or high SOC. Yeah. And unfortunately, I don't have the numbers in front of me at the moment. I should have. So one question is how much more is this side reaction accelerated at 100% compared to 50% or 30% for LFE? And I guess I would say it is accelerated, but it's probably not as scary as we might think, especially when you bear in mind that most commercial battery applications have a buffer at the top anyway.

So you know, although you want to get beyond the flat voltage curve onto the sort of bit near the top. You would still leave a few percent up there. And so when the system says this is 100%, this is the case with my own battery. It's actually a 95%. And so this degradation issue, it's there. And if we could avoid it, that'd be great. But it's probably not worth worrying about too much because in general LFE is pretty good from a lifetime point of view. I think if we were holding the battery at true 100% at high temperature for days and days and days and days and days, that would be a problem.

But that's the kind of magnitude of where we need to go to make it a problem. That makes sense. That makes sense. So you just don't want to keep your battery plugged in at 100% and for instance high temperatures because over time, that would accumulate and would create more degradation. Yeah. So going, I keep talking about home batteries. It is a problem with, it could be a problem with home batteries. So right now, my home battery is bouncing between 100% and 60% every day because there isn't any option in the software.

It's not a Tesla system, I should say that. It's another one. But there isn't any option in the software for me to adjust the maximum SOC limit. And because we have a lot more solar than we need most days at the moment, which is amazing, the battery is just bouncing between 160% or whatever instead of what would be much better, 30% and 70%. And so, yeah, that's kind of nice in the sense that a simple software fix could actually extend the life slightly. So I'm trying to persuade the manufacturer to think about that. But of course, then what you said earlier might kick in that the SSC calibration goes out the window. And how much of a concern is that? Well, that is an interesting question.

And it depends a little bit on other choices that the manufacturer has made. How good is their current sensor, how accurate is their modeling and so on and so forth. I imagine that it could be the kind of thing where you could have a backstop. So you could say, look, yeah, fine, I'll let you have a maximum SOC of 60%. But then every week it goes back to 100 for one day or whatever, just to calibrate. So again, I think it's a sort of solvable problem. Yeah. So there's two issues there when the batteries go out of calibration. One or out of balance, that is where the battery management system can't read the state of charge as well anymore, because they haven't been topped up to 100%. One of those issues is, well, you can't get as much utility out of your battery because you don't know if you're actually at the accurate state of charge.
这取决于制造商做出的其他选择。他们目前的传感器有多好,他们的建模有多准确等等。我想这可能是一种情况,你可以设定一个安全限制。所以你可以说,好吧,我让你最多充电到60%。但是每周会有一天或者其他时间回到100%,只是为了校准。所以我认为这是一种可以解决的问题。是的。 所以当电池失去校准时有两个问题。一个是失去平衡,也就是电池管理系统不能读取电池的充电状态了,因为它们没有再次充满到100%。其中一个问题是,你无法充分利用你的电池,因为你不知道你的电池实际上处于准确的充电状态。

So 95% might be actually 100% state of charge or vice versa. You don't really know what 0% or 100% is anymore accurate. So you might lose some utility out of the battery cell. But my understanding is when those cells go out of balance, it starts creating a bit of strain on some of the battery cells because each of the battery cells isn't experiencing identical conditions anymore. Are those two variables correct? Yeah, I think we're talking about a few different things. So maybe just to clarify, so there's one issue related to the fact that the Coulomb counting that I mentioned at the beginning can just drift. So we can, and that would be the case, even if I had a single cell. So a single cell, if I've got some noise and some offsets in my current sensor, even if I'm just not sampling often enough, then the accumulation of SSC is going to gradually drift away from where it should be. And so one reason we need to reset is that reason. That is unrelated to having lots of cells connected together. That's just to do with the fact that I'm adding something up in accumulating error. The second issue that you've mentioned relates to what happens if I have lots of cells. Do they all behave the same? What happens when they start behaving differently? And the answer to that is it depends a little bit how they're connected. If they're connected in parallel, then they tend to self-balance a little bit. So if one of them gets to a higher voltage, then another one, some current will flow into the one with a lower voltage and they'll kind of regulate in that way. Again, that becomes a little bit more tricky when you've got flat voltages and hysteresis, just as a side point. But if I've got cells connected in series, then they all have the same current flowing through them. And so if one of them reaches the maximum or minimum voltage, sooner than the other ones, because we always measure all the voltages on every cell in series, from a safety point of view, we have to do that. I will then stop charging because I don't want to overcharge that one cell. It's got to maximum voltage. And then the balancing system would kick in and maybe bypass that cell and charge up all the other ones, which is typically a slow process because we can't put much current through the bypass circuit. That makes sense. But yeah, so I think in the case of the first situation where we have Coulomb-County-mitch drips, that's kind of a problem. We don't necessarily know where we are and we actually lost. In the case of the second situation, assuming one cell in the string reaches a voltage limit, we do know where we are and then we can make some decisions. Does that help? Yeah, what I was trying to put my finger on is whether it causes damage to the battery pack and then the cells are out of balance. In principle, it doesn't, although it depends. But in the simple example where I've got lots of cells in series, as long as I'm measuring the voltage of every cell and I don't overcharge or undervoltage any of those cells, it's not going to damage anything. I just might have less capacity than I think. That makes sense.
所以有可能95%实际上可能是100%的充电状态,反之亦然。你真的不知道0%或100%到底是多准确。所以你可能会因为电池单元失去一些效用。但我理解的是,当这些电池单元失去平衡时,会对其中一些电池单元造成一些压力,因为每个电池单元正在经历不同的条件。这两个变量是正确的吗?是的,我想我们谈到了一些不同的东西。 也许我可以澄清一下,起初提到的库仑计数可能会受到漂移的影响。即使只有一个单元的情况下也会发生这种情况。所以,如果我的电流传感器存在一些噪声和偏移,甚至我采样的频率不够高,那么累积的具体容量会逐渐偏离应该的数值。重置的原因之一就是因为这个原因。这与连接在一起的多个单元无关。这只是因为我在累积错误。 你提到的第二个问题与拥有多个单元的情况有关。它们是否都表现相同?当它们开始表现不同的时候会发生什么?答案取决于它们的连接方式。如果它们是并联连接的,那么它们会自我平衡一点。所以如果其中一个电池达到了更高的电压,电流就会流向电压较低的电池,它们会以这种方式调节。当电压平坦且存在迟滞时,这就会变得有点棘手。 但如果我的电池单元是串联连接的,那么它们都会有相同的电流流过。所以如果其中一个电池单元比其他的更早达到了最大或最小电压,因为我们总是测量串联电池中每个电池的电压,出于安全考虑,我们必须这样做。那么我会停止充电,因为我不想过度充电那一个电池。然后平衡系统会启动,也许会绕过该电池并为所有其他电池充电,这通常是一个缓慢的过程,因为我们无法通过旁路电路引入太多电流。这听起来有道理。但是,在第一个情况下,当库仑计数发生漂移时,这可能是一个问题。我们可能不知道我们在哪里,事实上我们已经迷失了。在第二种情况下,假设串联中的一个单元达到了电压极限,我们知道我们在哪里,然后我们可以做出一些决定。这有帮助吗? 是的,我试图确认的是,它是否会对电池组造成损坏,然后电池单元失去平衡。原则上,它不会造成损坏,尽管这要取决于具体情况。但在一个简单的例子中,如我有很多串联的电池单元,只要我测量每个电池单元的电压,不会对这些电池单元中的任何一个进行过充电或欠电压,就不会造成任何损坏。我可能只会比我想像中的容量少。这有意义。

And there's companies who, including a spin-out company from our research group, which are working on electronics to kind of get around this problem by giving you more degrees of control, more extreme balancing. That helps a lot because I have asked probably 20 people this question and I've never really gotten a clear answer. The answer is of course nuanced, but what I'll do is I'll take what I've learned from this video and I'll make kind of an explainer video on it because I don't know if everybody will be following exactly along with what we're talking about. We're getting into the weeds of it, which I really enjoy.

Yeah, no problem. And happy to give you a bit more info on that specific issue. Okay, so there's three questions left that we have. Sometimes people will see a rapid change in the state of charge of their battery cell. How does that occur? Is your battery, for instance, when it's using those lookup tables, does it use an average of information over the past five minutes or is it constantly in a real-time basis adjusting where it's looking on the lookup tables? Right, so I think I've seen something like what you described on an old laptop, for example, where, I mean, four years later, the SSC would kind of go down and down and then it gets like 15% and then it would just stop working sometimes. Is that kind of what you mean?

Yeah, that would be one of the instances, but I know some people who have parked their car sometimes and they come back to the car and there's an unexpectedly large change in the state of charge. Okay, so I think there's two different, potentially two different issues here. So the one issue which relates maybe to what you're talking about is actually not to do the lookup tables or anything with the BMS. It's just whether the car is using energy while it's parked and some cars use more than others. You know, if a car is running kind of power-hungry compute equipment, doing security stuff or whatever, that can actually suck more than you expect over a few hours. So that, if you're like, that's a sort of legit reason for the SSC to change. It's just actual use of energy.

But if it's changed and the car or the laptop or whatever hasn't really been using much current, then yeah, maybe what's happened is the lookup table has become invalid. So you do your test on your battery at the beginning of life. You say, I've got the voltage versus SSC curve. I know my model parameters brilliantly. Everything's amazing. Sensors are well calibrated, whatever. And then five, six, seven, eight years later, the battery's aged a bit. The cells have maybe damaged a bit. That lookup table, those values, not so good anymore. And so that would exactly cause what you've described. You know, the lookup table is saying one thing, reality is something, saying something else, the voltage measurement, saying something, and you suddenly cut off on the basis of a voltage measurement just dropping sooner than you expect.

Perfect. So now moving on to the second of the three questions. Are there any upcoming advancements in AI, battery modeling, or electronics that could make batteries last longer or fast charger? I've seen some people saying that, hey, we don't need to make any changes to the battery cells. We can extend the life considerably, or we can make them charge a lot faster simply by using a different or better system of monitoring them. Right. Yeah. So it's a good question. I mean, I think there's a lot of hype around this. So we have to be a little bit careful to kind of say that. Machine learning is exciting, but it's also, you know, everyone's on the bandwagon. It's very exciting.

So I guess one thing I would say is it depends a little bit on what your baseline is, right? So we could say, yeah, things are super exciting. We can make loads of improvements, especially if you have a badly built, badly designed, badly calibrated battery. But I would say in that case, well, you should probably just use the conventional techniques better, right? Get a proper current sensor, prioritize your circuit models, whatever. So I think on the other hand, AI is quite interesting. There's a few areas where I can see benefits. Lots of people are working on materials discovery. So for just for new battery materials, searching like massive such bases looking for new types of materials, that's one aspect.

I think you can use kind of AI type techniques to help speed up optimizations. So if you're doing design or something, you can use AI for that. If you have a lot of data, so if you're a battery manufacturer, building a gig factory or, you know, some kind of production system, then you could use machine learning techniques to look at anomalies in your data. You could improve your manufacturing processes. You could look at where the bottlenecks are and so on and so forth. I think with respect to all of this, a lot of those techniques are quite well established. The issue is actually more whether we have rich enough data. And it's particularly hard with lifetime modeling because we want to know how the battery is going to perform over five, six, seven, eight, nine, 10 years. But the battery technology in the chemistry keeps advancing all the time. So it's like a chicken and egg kind of situation, which makes it really tough. But yeah, if you were a manufacturer with lots and lots of data, I think you're in quite an exciting position to use AI to look for potential improvements. But I don't want to oversell those improvements. I mean, they might be five, 10%, maybe even 20% if you're lucky, but they're not going to be 200% in my opinion.

Yes, you discover a new battery material, a new battery configuration, which is like double the energy density. Yeah, like any other innovation in the battery space, you have to take a closer look at what they're doing and look at the market and go, all right, what's for instance, what's your benchmark? Because if you're using a terrible benchmark, of course, you're going to look better. Yeah, it's really easy to make things look great if you use a bad starting point. Yeah. So last question, is there anything you're working on that you'd like people to know about or what's the best way to follow you?

Oh, that's super kind. So I do have a Twitter account. I'm not on there very much these days. I'm trying to do more in real life. I'm on LinkedIn as well. So people can look me up there. We have research group website, which I must update. The kinds of things that we are working on, which I'm excited about, we've actually done a lot of work on grid storage. So it kind of moved away from electric vehicles a little bit. One of the reasons for that is because you have more control over the battery. So if you're a control engineer, you want to control batteries. Grid storage is great. And also, I think we're seeing rapid transition certainly in the UK, for example, over here in the power grid, lots of solar, lots of wind, which is great. But grid storage is kind of an unexplored territory in terms of the way of the markets and the machine learning and battery physics kind of coming together and evolving.

So doing quite a bit of work on that. We're also doing work on machine learning from lifetime prediction. We're doing work on equivalent circuit models. We're doing work on physics based models. So if you're interested in battery modeling in general, ranging from devices up to systems, then that's kind of where we're at. And I'm happy to tell you more. And sorry, what was the name of that team or that group or that you're working with on all this?

Yes. So the research group that I run here at Oxford is called Battery Intelligence Lab. I'll send a website later on. I appreciate your time. This has been very enlightening for me. As I said, a lot of these questions I've been trying to get better answers to for several years now. And now I finally feel like I have a better understanding of how these battery charging curves work and what their role is. And I'll use that and I'll package it up into a really tight video package at some point. So thanks for your time. I appreciate it.

Amazing. Thanks for having me. And hopefully some of what I said is actually correct. But I'm sure people can write in and tell us, and I'd love to hear from you if you disagree. Good. Good. Have a great weekend. Bye. Bye. If you enjoyed this video, please consider supporting the channel by using the links in the description. Also consider following me on X. I often use X as a testbed for sharing ideas and X subscribers like my Patreon supporters generally get access to my videos a week early. In that note, a special thanks to my YouTube members, X subscribers, and all the other patrons listed in the credits. I appreciate all of your support, and thanks for tuning in.