0:00 My name is Dr Nick Simpson, I run the Electrical Energy Management Group at the University of Bristol in the southwest of the UK, and our research group is primarily involved in the management of electrical energy. So we design mechanisms and systems to convert electrical energy from one form to another, that could be from mechanical back into battery storage and vice versa. Now, it won’t have escaped your notice that most of the countries around the world have committed to carbon Net Zero by 2050, so we’re committed through the Paris Climate Agreement and various other agreements, and the aim of the game is to reduce our global CO2 emissions.

0:39 Now if you look at global CO2, you’ll notice that around a quarter of that is due to the transport sector. Now that’s only due to grow because of the population growth, but also because of new emerging markets, so things like short hop pure electric flights, so urban air mobility. Now in very, very mass sensitive applications like high-end automotive and aerospace, the the physical mass of the electrical systems is of critical importance. In particular the electric engine is the heaviest part of that component, and we need to reduce the mass as much as possible and improve its power density.

1:13 So institutions like the Aerospace Technology Institute and the Advanced Propulsion Center in the UK, and similar organizations around the world, have set industry targets of between 9 and 25 KW per kg by 2035. Now this has to be done done whilst maintaining an efficiency greater than 96%, so it’s no good creating a tiny motor and then cooling it to death, because you’ve got all of that cooling infrastructure. So you have to do a system level has to maintain more than 96% efficiency. Now at the moment that represents around a five-fold increase over the present state of the art.

1:50 Now the physical size of an electrical machine is ultimately dictated by the internally generated losses, so the energy that’s wasted as heat, and our ability to extract that heat, coupled with the temperature rating of the lowest rated material within that system. So typically at the moment that’s the electrical insulation system, and the upper limits are around 180 to 220° C continuous. And so if we want to create step changes in the power density that we can achieve, we need to do three things, and ideally we need to do them simultaneously.

2:22 We need to improve the efficiency of the systems, so at the moment the winding, which is the the mechanism that creates an electromagnetic field, is in the critical path. So we need to come up with ways of changing its topology and moving beyond conventional manufacturing methods to be able to improve its efficiency, so reduce the amount of heat that we’re generating. We need to be able to adopt new materials that have higher temperature ratings than conventional insulation systems, so go well beyond the 220° C limit that we’re at at the moment. And we need to improve our thermal management, so inevitably there are going to be losses in our system, it’s impossible to get a 100% efficient system, so how do we manage those losses? And critically, how do we manage them at source, because we can reduce the losses, but we might end up increasing the loss density, which will inevitably increase the peak temperature within that region.

3:12 And so a few years ago we started to look at this idea of adopting additive manufacturing to allow us to go well beyond the conventional manufacturing limits, which typically is still using drawn wires, so conventional cross-sections, rectangles, circles, cylinders, things like that, and we’ve been using those for a 100 years, it’s worked perfectly. But now we’re hitting those manufacturing limits and we need to try and go beyond them. However, the initial sizing of an electrical machine is still done using analytical equations, and it works really very well.

3:42 Now the problem with that, though, is that some of the the choices that you make are down to the designer, and they they’re typically based on experience, but they’re also based on bias. If you spent the last 20 years designing permanent magnet synchronous machines, every electrical machine problem, the solution is a permanent magnet synchronous machine, I can guarantee it. So how do we design our design tools to get away from those biases? But also the complexity of solving an electromagnetic problem is really rather difficult, because it’s transient by nature, you have to couple your electromagnetic simulation to your thermal simulation and your mechanical simulation, so it becomes an inherently multiphysics problem.

4:20 Now to get around that, in days gone by, because electrical machines typically have axial symmetry, we’ve essentially gone from analytical equation to two dimensional finer element analysis, and because that’s very rapid on on a modern computer we can cope with that kind of design and simulation loop. We only really go to three-dimensional final element analysis for validation and verification before we go to prototype. The major problem with all of this is that additive manufacturing allows you to go inherently 3D from the outset, so how do we constrain that computational problem?

4:53 The other issue is that the people designing these machines within this industry, a lot of the very well established industry-leading software, like Anis Motorad here, it’s all template based, and so how do you break free and actually design the geometry that you want to design rather than going based on templates that are pre-existing? Now if we want to actually drive adoption of this technology, we need new multiphysics design tools, we need to be able to cater for design fortive manufacturing, we need to understand the materials, not that are going into the process, but understand the behavior of those materials that are coming out of the other side, so what does your actual printed part behave like.

5:30 We need to understand post-processing, so how can we get rid of that surface R roughness and surface variability that we saw in previous presentations, how can we apply insulation coatings, and how can we ensure robust design. All of this needs to be done with a rigorous academic underpinning, and if we’re going to meet that 2035 target and to generate the impact that we desire, we have to have industry engagement, and that’s absolutely crucial to take the industry with us.

5:58 And so a few years ago I founded a thing called the Electrical Machine Works, so that was around 2 and a half years ago now, and it’s a group of multi multifaceted engineers, and we’re exploring the use of additive manufacturing for electrical machines and to improve power density. So one arm to this is looking at the geometric freedom, another is looking at thermal management, so how can we integrate cooling structures directly into our windings, another is to look at the functional grading opportunities that are in ad manufactur, so can we play around with the effective laser energy, for example, to play around with the the micro structure and the apparent equivalent electrical conductivity, for example.

6:40 We’re looking at postprocessing as well, so being able to take those relatively rough parts and being able to create those shiny surfaces that we’re used to seeing with conventional manufacturing. And the last thing is not taking AM as, you know, the the be all and end all, and it’s going to be the future of machines, but using the freedom that it gives us to start thinking about design in a very different way, to then retrospectively go back and find a much cheaper and much faster way of manufacturing something.

7:09 Now we’ve had quite a lot of success so far doing this, so we have a large aerospace company triing some of our technology, we have several automotive OEMs working with our technology at the moment, and we have an electrical machine OEM in the UK developing our technology. We also work with a company called Additive Drives, which are based in Dresden, south of here, and together the industry is slowly but surely starting to realize that this is a viable option.

7:33 The problem that we have is that the team around me are mechanical engineers, myself electrical engineers, electromagnetics engineers, material scientists, none of us are comp computational design engineers and none of us have a pure AM background. And so really, this this talk is a cry for help, to ask you guys how can we do this better, because although we’re getting results, we’re we’re looking at this in a very par, from a very particular viewpoint, and we just want to know how to do it better.

8:04 So the rest of the presentation are a few kind of examples of the the design choices and routs that we’ve taken, some of the results that we’ve achieved, and if you can spot where some of your technology might fit in and how we can improve things, then please come and speak to me afterwards.

8:24 So let’s take a first example. Conventional wires come in cylinders and rectangles, pretty much they they are only options, but we know from experience that if we can use the magnetic field, where these these conductors are operating, we can shape those conductors in such a way to minimize AC losses, so minimize loss generated by a time-changing magnetic field. So the theory of this is, if we take some of these conductors, we can see that the power loss within those individual conductors is related to the width to the cube, and only linearly to the height. So what that effectively boils down to is, if I can take my conductors and I can shape them such that they remain perpendicular to the impinging field, then I can reduce AC loss. So that’s a well-known design rule within electrical machines.

9:10 How do we actually apply it? So in this instance we’ve taken a two-dimensional time harmonic final element model, we’ve established from the very beginning we’ve assumed that all the conductors are going to be evenly disposed. We’ve then calculated the magnetic vector potential, and you can see the the curved lines there, so that’s effectively mag flux. We’ve sampled the field at certain points at the top and the bottom of each of those conductors, and we’ve extracted an equip potential. We’ve then joined the two equip potentials at the top and bottom of each conductor together to form a closed surface, and that is given as a conductor. And part of the reason why this is so computationally demanding is because as soon as I change the shape of the conductor, I change the shape of the field, and therefore we need to iterate until we get to a a steady state.

9:51 So we go through that process and we get something that looks actually quite aesthetically pleasing. So then we convert those conductors, which are spline based, and then we can convert it into a band representation of this model, so we’re not using meses at the moment, which is a good thing for a lot of reasons.

10:12 So if we compare the performance of our shape profile winding on the right hand side to two kind of well established and generic methods of of reducing losses, the left hand side is what you would do if you want lots and lots of torque, but you’re not operating at a particularly high speed spe. The middle one is what you would do if you need to operate at high speed, but you’re not too bothered about your DC losses, and the right hand side one, ideally, will give us the best of both worlds, and that’s exactly what happens.

10:40 So for the one on the left that you saw, you get very good performance at low frequency, but you have a huge gradient with frequency. For the other method, you’ve reduced the amount of conductive material that you have available, and so you pay a huge penalty at low frequency, but then it doesn’t have a particularly large gradient with frequency. Here, and then, our shaped version is the best of both worlds, and if we look at what that actually translates to in an electrical machine, is it allows us to push 20% more current into that machine for the same temperature rise, so that allows us to get 20% more performance effectively. There are some caveats to that, but let’s call it 15%, but all we’re doing is we are automatically applying a design rule that’s already well established within the industry.

11:24 So we’ll look at another example. So this was some work that we did with Additive Drives, so they designed this electrical machine, so these windings have been printed, this is a Formula Student vehicle, if you’re aware of that program between universities, and what they did is they designed some windings that have a slightly different physical size, but the cross-sections are still all exactly the same. So we decided that we would try and apply other methods to see if we can improve the performance.

11:47 So the first thing we did was say, okay, well, we want this to be as thermally performant as possible, so we’ll make those conductors conformal, so that we can make sure that the the conductors are really abutted to this boundary, because that’s our main source of cooling within this system. We then saw that we had an AC loss problem, so for various reasons, if we have these chunky conductors at the bottom of this slot here, we induce an awful lot of AC losses, so we need to segment them, so we split them into four different conductors on the side here.

12:17 We then said, okay, so we’ve got this segmented conductor, so at the top we’ve got solid conductors, at the bottom we’ve got segmented conductors, and we get this phenomena where we have circulating currents between the individual strands, which additionally leads to AC losses. So although we’ve reduced it, we can still reduce it further if we can get rid of these circulating currents. And so we turn to a thing called litzwire, or lit andr, which is effectively braided conductors, so the aim of the game here is to make sure that each of the conductors spatially lives in the same operating environment on average around a conductor. So what that effectively means is you want to twist your conductors as much as possible.

12:54 The downside to that is that when you do that twisting, not only do you create something that’s incredibly complex to model, incredibly complex to print, but you’re effectively increasing the effective path length, which means that you’re affecting your DC losses. So in an ideal world we would strategically transpose rather than transpose continuously. So what we did is we nestled this strategic transposition in the end winding. Now the reason it’s in the end winding is because it’s not within the active length of the machine, and so we can house it here without any penalty to the performance of the rest of the machine.

13:26 And so you can see from the graph that we’ve managed to reduce our circulating currents, so they’re not completely eliminated, because we haven’t been able to continuously transpose, but we’ve got significant performance improvement. And so this is what the whole structure looks like, and we’ve managed to to manufacture this this is obviously a rendering, but we have physical prototypes of this, and you can see that that transposition is fairly straightforward, so it goes up and then it comes back, and what that effectively does is twists those strands over and changes the position.

13:55 So if we look at this in terms of our performance, on the right hand side here, this is our baseline, so what we started with, so 100% loss, and its steady state operating temperature is 274. De, by by applying different strategies we’ve been able to reduce the loss by 40% and reduce the temperature significantly. So again, all we’re doing is taking well understood and well-known design rules in the industry and automatically applying them to our our systems.

14:22 Another example is our lit pins, so these are again transposed conduct, they’re very difficult to to physically print, but they have a huge loss performance impact. So for example, in the application that we designed those particular hairpins for, we reduced our loss by 47%. Now the challenge with those is that we need to be able to physically manufacture these things, and again I said that we we need to take industry with us to do that, we need to convince them that not only is AM viable, but it’s better than the alternative, even if it does cost more at the moment.

14:53 So you can see on the right hand side there, those cross-sections are from a CT analysis is, and in this, in the center here, you can see this is the cat geometry, it’s a nice lovely straight edges and things, and then this is what we ended up with in our first prototypes, those gap separations. We’ve now been able to get down to 100 microns, and the cross-section is much, much more, much more like the the cat.

15:17 Now one question that that we have is, we had to go back to our original modeling, we had to make lots of measurements in in our x-ray CT, we then had to update the effective cross-section from things with our card, then re simulate and match the data together, and it it worked pretty well, I mean we got very good close agreement, but could we take the XCT data directly and and model that in our CAD program instead, so using implicits, for example.

15:46 Now some of the major challenges are embedding thermal management into our structures, so I have a physical example of this here, so come and see it afterwards if you’d like, but what we’d like to be able to do is to reduce the torturous path of the heat to the cooling source. So here we’re bringing the liquid directly into the winding, and because we’ve got the freedom of additive manufacturing we can choose where we put that channel. Now in winding design, unfortunately, you end up with a temperature gradient, because you end up with more loss here toward the rotor than you do at the back iron here, and so being able to choose exactly where those cooling channels go is really valuable, because it means that you’re minimizing your head loss, so the amount of pressure drop that you’ve got in your system, and you’re targeting exactly way you’re putting your your heating structure.

16:32 Now, as you can see on the right hand side, these are X-ray CTs, and it faithfully represents pretty much the cad geometry, it’s really impressive, because those cross-sections of those channels are only 1.5 mm, and we were able to evacuate, and we’ve tested this and proven that it works. Now this was created using parametric boundary represented CAD, however you can see that when we start playing around with the geometry, some of these features start to fail, and that’s a serious problem for us. Nevertheless, it works extremely well, so even for very, very modest flow rates, so a small amount of fluid being pumped in there, you can significantly reduce the temperature.

17:10 Now an alternative is to look at using end winding heat exchanges, so the active length of a machine, where those those those channels would be, is actually a really valuable part of the machine, and ideally would be taken up by conductive material. And so if we can come up with schemes where we don’t have to take the active length region, and we can just work at the end winding, then we will take that option. And so this is where we can no longer use our band representation, we have to use other methods to represent things like gyroid infils.

17:40 Now some of our major challenges are articulating what can and can’t be built, and we need to understand that ourselves so that we can build it into our design software, but also our design tools have to produce multiple streams of CAD, so one that you would use for simulation that represents exactly how this device would be used within a larger system, but also because we need to electrically insulate these things, we produce how it would actually be built, which is different from how it would be used in reality. And we also produce automatically certain pieces of tooling as well, so that we can automatically insulate and we can post process, so if we’ve got critical surfaces and things that we need to be able to machine afterwards, we automatically generate the tooling to allow us to do that.

18:24 Now a significant challenge is the fact that if I go to, at the moment, if I go to 10 different vendors and say print me this copper piece of whatever, so this is a serpentine sample, or what we’ve dubbed to Serpentine sample, I will get 10 different components back, and they will all have different properties, same with the aluminium. And so we’ve had to come up with a measurement standard to allow us to measure the electrical conductivity of components, and to work very closely with vendors to understand exactly what the influencing factors are, and how we can remove those or minimize those in the process.

18:55 I mentioned earlier that we’ve started to use paracity to our advantage, so one reason why parity is important is because we can change the macroscale electrical conductivity of a material, so we can make a piece of copper electrically look like a piece of aluminium. The reason that’s important is because the skin depth, which dictates where current flows during alternating current operation, that effectively dictates how the losses are manifested within the device. And so if we can control the conductivity we can control that behavior, so we’ve shown that we we are able to do this repeatably and reliably.

19:31 The problem with proy, when you’re trying to design heat exchanges, however, is that if you don’t very accurately dictate exactly where that prosty is supposed to be or not, you end up with a heat exchanger that’s very, very leaky. So this is foam being blown out of the side of a of a heat exchanger, up until this point I had no idea that metal could be porous without visibly being porous. Surface roughness is a significant issue, and because of that we’ve put an awful lot of time and effort, or one of my material scientists has prala in developing chemical etching processes and electropolishing processes to be able to rem remove or eliminate that.

20:13 One of the advantages to the fact that we’re using chemical and liquid processes is that they are cheap, they are sustainable as far as possible, which means that we can we can have an economic compromise, so we can print faster, which tends to lead to much rougher surfaces, but it’s cheaper because you’re spending less time on the machine, but we can use a relatively cheap process to then postprocess and reduce that surface roughness, and so we can make some some economic gains there.

20:38 Electrical insulation is incredibly challenging, so, you, I mentioned earlier, you spend £4,000 on one of these prototypes and then ultimately you need to dip it in a bucket of gunk and hope for the best. We’ve now become a little bit more sophisticated than that, and we’re using electro deposition and various other methods to apply high performance insulation systems that can now operate at temperatures consistently at about 400° C.

21:02 So in conclusion, AM significantly increases the available design space for us in terms of shape, topology, material properties, integrated cooling, and other value add functions. All of the three design nams I mentioned at the beginning are facilitated directly by addtive manufacturing. Now we need to develop multi physics design tools and methodologies, we need to take design for additive manufacturing into account at the very beginning of process, the design tools need to take advantage of 3D design space inherently, which they don’t at the moment. We need to adopt perspectives from other disciplines, so you guys, and we need to understand how computational cost can be constrained, so there’s been lots of thermal solvers, CFD solvers, and things like that, I haven’t seen anything that can do electromagnetics, if you know of something please let me know.

21:50 So ultimately, digital design and digital manufacturing are a solution to step changes in power density for Net Zero, but in order to realize this we need to have many, many more overlaps in our ven diagram of designers. Thank you very much.