Below is the text version for the "2018 Cost Projections of PEM Fuel Cell Systems for Automobiles and Medium-Duty Vehicles" webinar, held on April 25, 2018.

Eric Parker, Fuel Cell Technologies Office

Good day, everyone, and welcome to the U.S. Department of Energy's Fuel Cell Technologies Office webinar series. We've got a great presentation this month from Brian James on 2018 Cost Projections of PEM Fuel Cell Systems for Automobiles and Medium-Duty Vehicles. My name's Eric Parker. I provide program support within the Fuel Cell Technologies Office, and I'm the organizer for the meeting. We'll begin in just a moment, but first, I have a few housekeeping items to tell you about.

Today's webinar is being recorded, and the full recording, along with the slides you'll see today, will be eventually posted, and we'll be sure to let everyone know. All attendees will be on mute throughout the webinar, so please, submit your questions at any point during the webinar via this chat box on WebEx, and we'll be sure to relay them. And, if we don't cover all your questions by the end of today's webinar, we'll be sure to address them at a later date. With that, I'd like to introduce today's DOE webinar host Greg Kleen, who is joining us remotely from Colorado. Hi, Gregg.

Gregg Kleen, U.S. Department of Energy

Hi, Eric. Thank you very much. It's my pleasure to introduce Brian James today. So, Brian James is vice president of energy analysis services at Strategic Analysis Incorporated, and he has over 20 years' experience working on fuel cell related techno-economic analysis. He leads Strategic Analysis' team, conducting DFMA cost analysis of transportation fuel cells systems.

Such work involves consideration of a wide range of components and manufacturing processes, and combines cost estimation, system performance modeling, conceptual design, and manufacturing process definition. Mr. James holds a master's in aerospace engineering from Virginia Tech and a bachelor's in aerospace engineering from the University of Virginia. He's a recipient of three DOE Hydrogen Program awards and has been issued six U.S. patents. With that, Brian, I thank you very much for being here today.

Brian James, Strategic Analysis Incorporated

My pleasure. Thank you, Gregg. I'd like to begin my acknowledging my coauthors on the work—Cassidy Houchins, Jennie Huya-Kouadio, and Dan DeSantis. There we go. Very good.

So, the motivation for the overall presentation today and the overall project is to identify fuel cell cost system drivers. This informs the Fuel Cell Technologies Office in their early stage R&D plans. We do this by updating the PEM fuel cell system cost projections for 80-kilowatt light-duty vehicles—specifically, cars—and also 160-kilowatt trucks, which are medium-duty vehicles. We examine three different levels of technology, and this covers both component technology, system configuration, and performance changes among those three levels. And the three levels are the current year—2018—a 2020 year technology, and a 2025 year technology.

We do the cost analysis based on a DFMA methodology. DFMA stands for Designed For Manufacture and Assembly, and it is a process based, bottoms up, cost methodology. At its core, it's simply combining the material cost plus the processing cost and the assembly cost, and sometimes, marking it up for the business realities of GNA and overhead. It's power is that it's processed based, so it mimics each of the steps used in the overall fabrication process, and so, it allows us to change each of those minute details to assess the cost impact. And because we're able to do that, we're able to gauge the impact of cost—or the impact of manufacturing rate—upon the cost by varying the material cost, selecting the manufacturing method, looking at the machine rate—which is really another way of saying the amortization of the capital equipment used for making the product—and also looking at tooling amortization—so, dividing up the overall tooling costs among the number of units produced.

Beginning with the light-duty vehicle systems—the cars—I'm gonna go over the system schematics for all three of the time frames. So, I'll talk about the system definitions, then I'll go into the individual components. These are basically the neat things that we've studied over the last year that merit discussion. And I'll conclude the automotive section by showing you the overall costs. After that, I'll go into the medium-duty vehicle trucks.

So, it turns out that we use the same system configuration for all three technology years. It has the fuel cell stack at its core—a single stack producing around 88 kilowatts gross, 80 kilowatts net. It's supported by a hydrogen system—that is a supply and a hydrogen recirculation loop in there. Then, there's the high-temperature coolant loop, which is the main cooling loop for the stack. And then, there is a low-temperature cooling loop over here, which cools the motor as well as the air pre-cooler.

Finally, there is the reactant air loop, which enters over here on the right, goes into the compressor, gets cooled down, goes into the membrane humidifier to be moisturized into the stack, out of the stack, back into the membrane humidifier, and then, it's looped around the exhaust gas expander to capture some of the energy of pressurization. Compared to last year, we've added a couple of components in the balance of plants. Once is a stack shutoff valve at the end. This is after consultation with the fuel cell tech team and looking at the Mirai system, too. It is to isolate the stack to avoid back air infiltration during stack shutdown.

We've also added a stack bypass valve in here, which allows us to divert air from the air compressor either the stack or just to do a bypass that allows us to independently control the mass flow through the compressor, and the mass flow into the stack. And this is now—this system diagram is in full coordination with the Argonne modeling system diagram, as we strive for every year, but both systems have changed, and we're all on the same page. And lastly, we have a new pulse ejector as part of the hydrogen recirculation system. An ejector works on the Bernoulli principle. So, it takes the motive input gas, speeds it up in a conversion diverging nozzle, and, at the low-pressure point at the throat, it draws in low pressure recirculated gas.

The new part is adding an injector. One's an ejector—this is an injector, which pulses the primary flow to allow the ejector to operate at a lower flow rate than it otherwise would be, because it starts the ejector, but then, it accumulates the hydrogen and then increases, temporarily, the mass flow which allows the ejector to work at a different operating point. And finally, there's the second injector, which is basically a bypass around the ejector, and it's used for purge of the stack to allow a mass flow of hydrogen that is greater than that which would be allowed through the ejector itself.

I next have a couple of pages that define the light-duty vehicle system. It's a bit of an eye chart, so I'm only gonna hit a few points. We have three columns—2018, 2020, and 2025—and the numbers in green are those values which differ from the 2017 analysis last year. So, we've slightly increased power density. We've held platinum loading the same. The gross power and the cell voltage are slightly different, but largely the same. And in 2020, we've switched in terms of membrane material or support specifically, we switched from being supported on ePTFE to an electrospun PPSU support—and I'll talk about that.

In 2025, we further add in a lower equivalent weight membrane to allow more presumably correlating with increased performance that is assumed up here to allow more acid groups within the membrane. This is the second page of system definition. I'll let you read it on your own at a later date if you're so interested. Under bipolar plate forming, we switched over to hydroforming. There was a similarity in cost between the progressive die stamping and the hydroforming, but in consultation with the fuel cell tech team, we went with which one was slightly cheaper and that was the hydroforming, so that's a switch from last year.

And, in terms of bipolar plates, there was an alteration from last year's analysis for 2025 where we previously assumed there was no coating required on the bipolar plates, however in further consultation with the fuel cell tech team, there was a consensus that coatings would be needed in the foreseeable future, so we added the TreadStone TIOX coating application back into the analysis for 2025. The third page is system definition and under the air compression system, we increased the motor and motor controller combined efficiency from the previous value of 80 percent up to 90 percent for all three technology years. And lastly, as I mentioned, the anode recirculation is now using a pulse ejector with bypass for all three technology years.

So, a key aspect of the annual updates and the progress in the fuel cell systems, is the polarization performance. Back in 2016, for the 2016 analysis, we were basing it on platinum nickel catalysts on—or de-alloyed platinum-nickel catalysts—on a carbon substrate. And then, at GM and elsewhere, there were developments of high surface area carbon support, and that led to an increase—a substantial increase—last year, in projected polarization performance, resulting in, as you can see here graphically, the 2017 operating point in terms of power, current density bumped up substantially over 2016. We further increased the power density by six percent based on additional Argonne modeling of the systems, but it's a much smaller increase compared to the big chunk last year when switching over to the high surface area carbon. Now, it's understood that the high surface area carbon was responsible for the performance increase, but also, between 2016 and 2017, we switched from a de-alloyed platinum nickel catalyst to a de-alloyed platinum cobalt catalyst. However, from modeling and experiment, the metal selection was largely—yielded similar performance. It was really the difference between the carbon support versus the high surface area carbon support that yielded the performance difference.

However, last year, we didn't do a full analysis of platinum cobalt. We basically used the platinum nickel synthesis method with platinum cobalt materials. So, this year, we went ahead and did a full up DFMA on the full de-alloyed platinum cobalt HSC catalyst, as I'll talk about on the next page. This last column shows some of the operating conditions for this new operating point. The stack pressure of 2.5 atmosphere at 94 degrees remain unchanged from last year.

So, this slide represents the new synthesis route that we did with platinum nickel on carbon shown from last year, and platinum cobalt, over on the right. Previously, we used a chloroplatinic acid precursors—no, not precursor, but, if you will, platinum donor material, into a precipitation reactor where we were able to introduce nickel and platinum directly on to the carbon substrate. And, after that, it went through a filter and wash, dry, crush, and annealing for 2 hours at 1,000 degrees, a 24-hour de-alloy to cut down on the nickel, more filter and wash, dry, and crush. So, all of these steps over here on the right are duplicated in the new synthesis method for platinum cobalt, except we added processing steps for the high surface area carbon. So, instead of putting in like, a Ketjen black carbon previously, we now put in a cheaper carbon black, but then, put it into a fluidized bed at 950 degrees C for multiple hours to come up with a model for the high surface area substrate.

Secondly, where previously the two metal species—previously nickel and platinum—had been added in one reactor, now, they are added in two reactors—one through a platinum nitrate formulation to deposit the platinum, and then the other, through a cobalt formulation to deposit cobalt. Overall, though, when we do the numbers, there's a net cost decrease resulting from switching over to platinum cobalt, partly because of the—this is compared to last year—partly because of the increase in power density, which means less RAMs of a cathode catalyst. It's $0.13 per kilowatt cost to decrease in switching from the nickel to the cobalt. This is shown in a tabular form over here—a bit of an eye chart, and much to see—but the top three steps are the new processing steps that were added for the new platinum cobalt synthesis method. And, as you see here, at 500,000 systems, the overall price delta is only around $11 a system. So, it's a quite small impact in the switch.

This is the DFMA cost estimate of the high surface area carbon. We show here in blue, the cost starts off at a very high—over $3,000 a kilogram for the carbon and goes down to around $100 a kilogram at high production rates. However, this high price is perhaps artificially high due to the very small batch size and the poor utilization of the equipment. When we spoke to industry, we're able to get rough price quotes, approximations of $1,000 a kilogram for small orders.

This means that's the market price right now, so in terms of the value used in the model, we capped it at $1,000. And then, when $1,000 was over, the projected price at volume we switched over to the other cost curve. And the lower price here is because the production rate that it's actually being produced is presumably greater than the single application with demand. So, at 1,000 vehicles a year, you really don't need many grams of HSC. But it's currently more than that is being produced.

Next, I'll talk about some electrospinning materials that we looked at. Electrospinning can be used for a variety of applications. Specifically here, we examined its application for the membrane support, which would be a direct substitution for the ePTFE in the Gore-like membranes. We also looked at it for making the entire supported membrane, and this is done via dual-fiber or co-spinning on both the support and the ionomer material. And then lastly, one can directly electrospin the electrodes, both anode and cathode. So, I'll speak about each of these individually.

First up is the membrane support. We went out to examine electrospinners, and there are a variety of units on the market. The one that was highest volume and used as the basis for the modeling is the Inovenso Nanospinner 416, which is a series of nozzles putting forth a certain mass flow rate. And when we worked up what exactly it would be spinning—which would be PVDF or preferably, a PPSU polymer material for the support in a solvent, you put this mixture together, you put it through an individual nozzle, and you can lay down a certain grams per hour.

When you combine a certain number of nozzles together at a line rate and a support thickness, then you could come up with an overall area, and lay down, and then convert that over to a projected price after going through the DFMA process. Over here on the right, we have the red curve is the ePTFE baseline supports is what we previously were using for the support. The green is the Giner directionally stable material—excuse me, directionally stable membrane support material previously analyzed by us. And then, these three curves are three variations of electrospun PPSU supports. It's somewhat of a mini cost sensitivity here because we have three different lay down rates.

And that impact, though, is that you project the electrospun support could be potentially substantially less expensive than the ePTFE supports that we have. That's very encouraging. And this is an interesting result, because one gets the seemingly, cognitively, dissident sentiment that electrospinning is cheap, but how can it be cheap when the flow rate that it is depositing is so low? You do that by not needing much material and by going massively parallel in terms of the number of nozzles. And the numbers bear out that the cost can be lower than the overall processing prices that we project or that we model for ePTFE.

Next, we move on to the dual fiber approaches. This is making the membrane itself support in the ionomer material by having some of the nozzles be for the support, and some of the nozzles being for the ionomer material. So, you get this nest of alternating fibers in here. PEO is used to facilitate the electrospinning, but it would poison the system, so it has to be removed, and that is modeled as a five-minute hot water bath for the removal process. And then, these fibers get compacted and melted, per se, and there are various techniques that you can use to do this. We've modeled conditions where the Nafion is being melted around the PPS new fibers.

So, we end up with structurally, something very similar to the ePTFE fibers surrounded by the ionomer. It's possible to do it the opposite way where we melt the support material around Nafion fibers. When we do the cost estimates on $1 for square meter basis, we have the baseline membrane with ePTFE supports at the top, and then two cross curves for two different variations showing that there are potential cost savings available with the electrospun combination material. Now, of course, our conclusion is that the electrospun membrane is projected to be less than the ePTFE supported membrane—that is, of course, only valid if the electrochemical performance using the new membrane is equivalent or superior. We think that's the case. We'll leave it to others to more fully demonstrate that that is, indeed, true.

And then, lastly, there is—using the electrospinning process to make the electrodes itself. So, we looked at the cathode catalyst. In this case, the polymer, if you will, is a mixture of the platinum—in this case, we modeled platinum nickel on carbon because we did this earlier in the year—in Nafion combination along with a solvent. That's what—it's slurry is being put through the individual nozzles. On a cost basis, they're gonna be close because of the platinum costs are—the platinum nickel catalyst costs—are quite similar.

When we put them into a comparison of the different MEAs using various combinations on a total dollars per square meter basis, what we find is that because of the high cost of the platinum, the cost curves appear very close together. So, they have the differences associated with the previous and the material processing excluding platinum; excuse my earlier statement—but because of the high cost of platinum, they come together visually, at least. And so, what we conclude is that the prices are nearly identical with a slight nod towards the electrospun elements, but really, performance and durability will be the deciding factor in selection of which components, if any, are electrospun.

Next, we took a look at an alternative carbon—an alternative catalyst synthesis method, and this is directly as an alternative to the aqueous synthesis that I spoke of earlier for the platinum cobalt and also, in that sense, for the earlier examined NSTF process, which is a PVD. So, basically, the process is using physical vapor deposition to coat onto a carbon powder. And this is piggybacking on the investigation of Jim Waldecker at Ford Motor Company in association with Oak Ridge and Exothermics to demonstrate the basic synthesis using platinum niobium oxide as the base materials for the testing.

We took a look at the public literature associated with this kind of deposition, and we built up a design for a dual barrel deposition system. The inner barrel contains the powder carbon substrate, which is down at the bottom, shown here visually, and inside that inner barrel are the magnetron sputtering units. So, they can create a vapor that goes down and coats the top layer of the carbon powder, and this inner barrel is rotated to tumble the carbon powder to constantly present a new face of the carbon powder so that the materials are distributed evenly across all surfaces of the individual carbon elements. That's the inner barrel holding the carbon materials, and then, there's an outer barrel to contain the vacuum.

There are also a lot of individual components that come into play in the overall bill of materials. There are motors to turn the inner drum vacuum pumps. There's a heating system to drive off impurities of the system and a variety of other platinum recycle elements that need to be taken into consideration. We took a look at all this, created a bill of materials, created a cost estimate, and then, we took a look at the processing parameters that are summarized here looking at both 30 to 70-kilogram batch sizes. These are quite large batches.

They have a projected cycle time of 4 days—just under 96 hours. As such, you only get 63 batches a year, but because you don't need very much material, this is good for in excess of 100,000 vehicles a year. So, it is a high-volume production method. We worked through the elements of the cycle, including powers and the material systems, and then went and did the cost analysis. This slide here just shows an image of the Exothermic facility and gives a hint as to the very large diameter chambers that we're talking about: 6-foot diameter, 4-foot in length, holds reasonably30-70 kilograms of catalyst per batch.

Here are the catalyst synthesis cost results. The top blue curve is our baseline platinum cobalt on HSC aqueous method cost in terms of dollars per system versus systems per year, and then, the red curve is the PVD platinum niobium oxide system. So, we are showing a modest cost reduction for the new system, and so, that's very encouraging. Of course, as with all of them, the proof is in the pudding. We have to show that the performance of the new deposited system is equal to or better than the alternatives. So, it's very encouraging. It shows—at least on first pass—the feasibility from a cost perspective of the PVD powder coating system.

Next, we looked back at the compressor motor expander unit. This is an update of the design that we worked with Honeywell on back in 2008–2009 period. It was based on a three atmosphere 165,000 RPM centrifugal compressor and expander units floating on air bearings, which we worked with them in proprietary drawings to do a very elaborate, very detailed cost estimation, and then, we're able to use scaling parameters to adjust pressure ratio and flow rate and motor power. But, we hadn't really looked back at the design for quite a few years, so, in association with Argonne and Honeywell, we talked about what, if anything, changed over the years and needed to be updated in design. The long and short of it was that the basic CEM design was still appropriate and no significant design changes were needed.

However, we did update a variety of things. We updated the airflow rate for the air bearings, which only had a minimal effect, but we were able to capture that air. We, of course, resized it for the 2018 mass flows and the motor power. As mentioned earlier, we updated the motor and motor controller efficiency from the previous 80 percent to 90 percent value based on fuel cell tech team inputs and discussions with what's possible with Honeywell. And, we also took a look at inflation. And, unfortunately, over the years, because of the complexity of the previous analysis, we hadn't been updating the material costs and unit costs of the CEM every year. So, we basically let the old 2009 time frame price ride on the CEM with only adjustments for scaling—so, once again, the airflow and the power when we changed compression level and gross power of the unit.

So, it was scaled and adjusted every year, but the core base that was being scaled was not adjusted for inflation for quite a while. So, we went back and said, "Gosh, that's not right. What should we do to remedy that?" And we ultimately ended up just applying the PPI to it to get a cost estimate in 2018 dollars, which is not exactly right, because it's a gross inflation adjustment that may not be appropriate for high-speed turbo machinery and electronics. However, it was deemed to be a better representation than just going with the old non-inflation adjusted price. This, unfortunately, had the impact of raising the cost over $1 per kilowatt net for the 2018 system.

We also took a look at the laser welding system. We had done an elaborate study previously. We had further discussions with Lincoln Electric and other component vendors to re-evaluate and dive into even further detail. In general, the capital costs increased as we added more detail and sort of backed off our perhaps too aggressive parameters. One thing we didn't appreciate was the substantial weld fixture costs. These are custom weld fixtures to hold the pieces together and they can be much more pricey than we were anticipating.

We also got feedback about the index and rotation times between stations and increased that substantially. We had—basing it on a turntable design—this is for the laser welding of the two halves of the bipolar plates. We have to weld the periphery to seal in the gases and liquids and we have to seal at least over a portion of the active area to ensure good electrical connectivity. But we envision robotic loading of multiple pieces into fixtures and rotation of the turntable into a welding station with a certain number of individual lasers and a certain number of bipolar plates simultaneously being welded.

It could be multiple welding stations to allow some cooling time in between and refixturing to allow line of site access of lasers and then, an automated unloading process for a robot. So, we broke that down in different configurations depending on low volume and high volume, and there was actually an intermediate step in between. We took a look at the breakdown of the different cost components, and we got some price ranges from the vendors which we spanned. We took the midpoint of these, and what was the end result? Well, that the low production capital cost went up about 16 percent and the high-volume production capital cost went up close to 70 percent from $1 million to $1.7 million.

So, we factored that into that into the overall DFMA analysis, and some parameters are shown up here, and graphically, the 2018 cost is around $0.50 higher than the 2017 projected systems, due to increases in capital cost and slowing down of the cycle time due to the index time to rotate the wheel. So, by way of summary on the automobile, we have the total fuel cell system cost—dollars per kilowatt versus production rate. The shape of the curves is similar to what we had seen before, within either curve occurring around 10,000 units, and the costs are now roughly $46, $43, and $38 for each of the three time frames. This represents a cost increase for all three, other than middle year is largely the same. It's well within the margin.

The only reason we add the pennies on this is just to show you how we're rounding it, so you know whether we're—in case we're rounding up $0.49 or rounding down $0.49. And the cost is going the wrong way. We'd like it to go down rather than going up, but it is based on cost realism and multiple factors simultaneously factored in. I'd also like to do waterfall charts that show more graphically the impact of each of the costs. So, we started with last year's $45 per kilowatt and we had a series of changes.

The power density increased by six percent. Increase in power density was in $0.80 per kilowatt reduction, and unfortunately, pretty much all the other changes increased costs. So, we grouped them into the bipolar plate changes—which was the capital cost of the laser welding that I talked about before; it's $0.50. The MEA changes, which was the catalyst cost, decreased with some other internal changes, which was a net $0.50 increase. The balance of planned components—those are the new shutoff valves, etc.—37, and then, the CEM costs, which was only a net 61 because the increase in cost due to inflation was offset by the decrease in cost due to the motor control or efficiency improvement. And excuse me—this is a typo—it's 80 to 90 percent improved here rather than 95.

Next, I'll go into the medium-duty vehicle system. I'll show you the results of our scoping study that we looked at. Once again, we do show some schematics for the three time frames that we had—as similar to the automotive, do system definitions, operating conditions, and conclude with medium-duty vehicle pricing. We try to leverage our work as much as possible from what others have done in the past, and so, we've heavily leveraged the work of Ram Vijayagopal out at Argonne who looked at 12 different medium and heavy-duty truck applications. We also combined in studying the work by the group 21st Century Truck, which although it is not fuel cell power system focused, it does include advanced powertrains and impacts the power levels and aerodynamics of future trucks, which come into play for our system sizing.

In general, there are two different powertrain architectures that could be considered. The first is the fuel cell range extender, which uses a typically small fuel cell to extend the range of a battery truck, in this case, and the other is a fuel cell dominant system, which is what we have for the automobile, which uses a typically large fuel cell for the size for the peak power acceleration or top sustained load speed for the vehicle. And that's what we are using/selecting for this analysis. For those of you familiar with vehicles, both cars and trucks are defined—Department of Transportation as well as other groups, FTA—by gross vehicle weight. So, we are looking at the medium-duty class in terms of power levels, as I'll show you on the next page.

Our selection of power level services many of these medium-duty vehicles, but the 21st Century Truck Group has adopted a medium-duty vehicle baseline, which is a class 6 truck, and they've also adopted a heavy-duty vehicle baseline, which is a class 8 line haul truck. So, we sort of line up with theirs, although they're more specific in terms of what kind of vehicle they are modeling. The reason that we are general in the vehicle is that when one graphs the fuel cell stack rated power against the test weight, we see that many different truck applications have similar fuel cell power requirements—and this is using the Argonne study from Ram directly. So, he looked at the different requirements for these different types of vehicles, and a lot of them would require fuel cell power around 160 kilowatts. So, we picked 160 kilowatts as the medium-duty vehicle power level that we are modeling, and we rounded down to 160 because it's 2 times the 80 kilowatts that the auto uses.

And therefore, one of our conclusions and observations is that the stacks can be built up from 80 kilowatt modules—which makes it right in line with the architecture of the automotive stacks. The other observation is that two main power levels capture most of the medium-duty and heavy-duty applications. The second clustering tends to be around 260 kilowatts. This outlier here is the long haul, over the road, very heavy 70,000-pound line haul truck application, which requires yet another 100 kilowatts from the second grouping. So, once again, we are looking at 160 kilowatts for the medium-duty, and that power level is applicable to a wide range of vehicle applications.

This is the system schematic for the 2018 MDV system. It is very similar to the automotive system with two exceptions. One is that it has two stacks instead of one stack to measure it with the 160-kilowatt net power versus 80 kilowatts with the car. And secondly, it does not include an expander over here. So, it's just a compressor and then an exhaust without capturing exhaust pressure from the stack.

But it does include the stack shutoff valves and the bypass valve as well as the pulse ejector that is used in the automotive system. When we go to 2020 and 2025 systems, the system schematic is identical, with the exception of the new expander that is added on here. So, then, it brings it in line conceptually with the automotive system. I have the system definition charts for these MDV systems, but there are quite a few rows to go over, so I'm just gonna skim over them. I would note that these three red columns are for the three years of the truck, and this first column is for the bus analysis that we did year before last.

The bus is a 40-foot transit bus, so it is a medium-duty vehicle. The expectation was that the parameters would be very similar to this year's MDV analysis. However, in green, we show the various parameters that vary from the column to the left. So, you see that a lot of things are different from the previous bus analysis, and that's because we have chosen to model the truck as influenced by the recent Mirai fuel cell vehicle, which is a car, but it has two good things going for it. It has a very modern fuel cell stack compared to the bus systems that typically are on the road that architecturally wise have not changed for the last 10 years and secondly—the Mirai system has been put into a demonstration truck application using just two Mirai stacks into the truck, which is directly analogous with what we're talking about doing. So, it forms a credible basis for projecting that future truck systems will be constructed car-like in that regard.

I'll go over the individual parameters on the next slide, but these graphs—this page, this second page, and this third page—are there to fully specify what the truck system looks like in terms of material and efficiency selections. So, this next table is a comparison of operating parameters. It has the similar columns before with an added automotive column over here on the left—the 2018 LDV. Interestingly, the number of systems produced per year for trucks is quite high. It can be 100,000 systems a year for medium-duty vehicles. This is substantially higher than the bus production rates, which are closer to 1,000 or 4,000 nationally and less, by a factor of 5 or so, than the auto.

So, the trucks, medium-duty, are a bigger market than the bus is, but smaller than the cars. Secondly, we had to adopt some type of compression methodology, whereas the bus system previously and buses on the road still use a roots type blower compressor with no expander, and that's in contrast with the auto systems which use a high-speed, centrifugal compressor and expander. We have adopted three different techniques for the truck. We start off with roots blower only and we add a roots expander to go with the roots blower in 2020, and then, in the far term—or farther term—we switch over to an automotive style centrifugal compressor. It is our considered opinion that there will be—at least in a configuration basis—a convergence of the truck and the auto because it will result in lower costs.

An important parameter in distinguishing the truck is its durability of the stack—nominally, 25,000-hour lifetime, which has been demonstrated in the bus systems. But it's well in excess of the 5,000-hour targets for the automotive system. To obtain this 25,000-hour lifetime, two main approaches are used. One is to increase the platinum loading up to roughly .35 milligrams total platinum—so, roughly three times the loading of the car—and also, to reduce the stacked temperature. Operating at a lower temperature has, apparently, very impactful effects on the lifetime.

So, going along with these, there are also other parameters. The pressure—which we adopted the Mirai 2.35 and one and a half stoke, and this, overall, results in a polarization performance level that is approximately equal to the car in the near term, due to the higher platinum loading, but we strained in the far term as not ever achieving as high a performance as the automotive system in the future. So, the car was up to 1,500 milliwatts per square centimeter projected in the future, and we've limited 2025 to 1,350.

This all gets wrapped up in the MDV system cost results. They have a similar downward slope for the three different years—roughly $100, $90, and $80 at 100,000 systems a year. When we compare this to the automotive system for the 2018 curve—so, comparing the two blue curves—there's roughly a 2 to 1 ratio—$100 versus $50. And note that this is not the $46 as previously discussed, because the $46 a kilowatt was at 500,000 systems, and this is backed off to the peak rate of the truck system of 100,000. And also note that the trucks are roughly twice the capacity—160 kilowatts versus 80 kilowatts. Also, we're comparing on a constant system basis. There's a difference in installed power.

Overall, the large cost differences between the auto and the truck are due to the platinum loading, the lack of a CEM—which impacts the gross power, increase of parasitic loads—and then also, a business impact of non-vertical integration structure so that the business entities are more spread out with additional layers of business markup, which tends to boost the cost of the truck power plan. So, in summary—I won't go over each of these in the interest of time, but we have the auto cost results, which are slightly up from last year, but are $46, $43, and $38 a kilowatt for each of the three years examined, and it factors in the 6 percent increase in polarization performance made available by ANL updated modeling of the platinum cobalt HSC catalyst. A couple of these bullets are left over from our previous observations, where the bipolar plate material is a substantial fraction of overall bipolar plate costs. And so, if you were gonna reduce the cost of the overall system, you really have to do something about reducing the bipolar plate material cost. One thing you can do is to go to high speed 2D or roll-to-roll processing for bipolar plate fabrication. That also gets around the problem of a massively parallel production line.

We switched to pulse ejector, and we'd like to get down to the 2020 DOE targets. We're not quite there yet. And on the 160 medium-duty vehicle truck systems, we're finding that the stacks are very similar to the auto system, except for the higher platinum loadings longer lifetime running cooler. And they're roughly $97, $90, and $80 per kilowatt that we just spoke of.

There are a few remaining barriers and challenges that I'll just briefly touch on. There's some uncertainty in the ionomer cost, so, we'd like to do an analysis there. There's the bipolar plate material cost issue that I just mentioned. There's an issue of ammonia contamination—particularly within the air humidifier, so, we'd like to get rid of the air humidifier, but then, of course, you have to worry about ammonia just from atmospheric ammonia from other cars running down the road, poisoning the membrane either in the air humidifier or actually in the stack. So, there's that issue hanging out. The $40 a kilowatt DOE target for 2020 is difficult to achieve. We're within spitting distance of it, but it's not quite there yet. And the $30 per kilowatt DOE target's obviously even harder to achieve.

We have this massively parallel bipolar plate forming line issue that I alluded to earlier—that transition to 2D production might be a methodology for reducing the number of parallel lines needed. And under the MDV truck category, we think that there is room for improvement in our understanding—the community's understanding—of the trucks for operating mode—whether that be range extender or fuel cell dominance or running the fuel cell at its sweet spot and then relying on battery augmentation to really do the ups and downs so that you get the long life of the fuel cell by just running in the optimal condition pretty much all the time. It's either on running optimal conditions or off.

So, we are interested in better exploring that. Our proposed future work piggybacks on a lot of these things we want to continue on the automotive side to examine durability into the cost modeling, look at the cost of the ionomers, look at 2D manufacturing. In terms of bipolar plates, there's a new company called Precors, which has a carbon coating process, a non-vacuum based, that looks to be low cost and is getting very promising performance results. We'd like to model that. I'd like to continue doing cost sensitivity studies, because those give insight into which areas we really need to screw down on the values and which ones are insensitive. On the truck side, there is an upcoming fuel cell truck workshop sponsored by DOE, so, we're looking to that to get an important feedback from the community and to start a broader community discussion, specifically about fuel cells and trucks. And, of course, all of the above work is documented in a final report that will be coming out—or submitted to DOE at least—at the end of September.

I'd like to thank our sponsors. Gregg is our immediate sponsor at the Department of Energy in the Fuel Cell Technologies Office. Our partners on the project are primarily Rajesh Ahluwalia at ANL and Mike Ulsh at NREL. And we've collaborated with a variety of FCTO-supported groups and a much longer list of industrial partners who have provided us very valuable insight and information in the component technology. With that, I'd be happy to take any questions.

Gregg Kleen

Okay. Thank you very much, Brian. Yeah. We do have a few questions. If you do have a question for Brian, please enter it into the chat box and we will try to get to as many questions as we can here.

The first question was—is there a need for a particle filter in the hydrogen loop?

Brian James

Interesting question. So, there are particle filters coming out of the tank. There's certainly one associated with the hydrogen system here, but there is not currently one inside the recirculation loop. If any pieces were freed up somehow in this loop right now, they would not be captured. So, I'll make a note of that, I guess, suggestion, and talk to my colleagues just to see whether that needs to be added as, unfortunately, yet another balance of plant component, which would increase cost.

Gregg Kleen

Okay. Thank you. The next question is—couldn't it be more cost efficient to take a compressor without air bearings, allow some oil droplets to be admitted, and then, add a separator for oil droplets downstream of the compressor?

Brian James

I don't think so. I think the air bearings are a very simple and inexpensive component. They're literally foil bearings, and they allow the system to operate at very high RPM, higher speeds that would be available with a roller bearing rotational system. And then, we completely obviate the oil issue, so there's no oil in the system at all. And I think some have argued that you can try to remove oil, but you'll never get all of it, and over time, it will accumulate in places that you don't want it to, thereby, eliminating durability of the overall system.

I think it loses—I think we ended up where we did—both from a performance basis and from a cost basis—because I think the unit would be more efficient the way we have modeled it with high speed turbo machinery floating on air than with lower speed components.

Gregg Kleen

Okay. The next one is—on slide 36, what percentage of the price difference is a result of market layering?

Brian James

I'll have to get to slide 36. That is a good question and I don't have an answer for you right now. But because we have, one would say, a complicated model, we can go back in and take out the multi-tiered structure of markups to assess that. So, we will include that in a future sensitivity.

Gregg Kleen

Okay. And then, there's a question—is the ammonia contamination forcing the use of cathode air filters absorbing this gas?

Brian James

Yeah. So, yes, and no. Right now, we do not have a—say, like a carbon filter in there to absorb impurities in the air system. We just assume that there isn't a problem, per se. We have spoken to manufactures as to what it would take to beef up the cost of the air inlet to remove gaseous particles—so, maybe another way of saying it is it is just a particular filter system now, but there are commercial products out there that add a gas filter onto the air filter specifically for that purpose.

The downside of that—in addition to the cost—is it's another part that potentially needs to be replaced, and it also adds pressure drop, which means the gross power would go up to some extent. It might be a minor impact, but possibly some.

Gregg Kleen

Okay. And then, have there been any new improvements on GDL manufacturing process and cost?

Brian James

We did a big GDL ground up DFMA last year and—the question was, "Have there been any big improvements in costs?" A lot of the improvements in cost that we were projecting is attainable has more to do with the engineering side than with the technology breakthrough side. In other words, going to wide [inaudible] DL coating and heat treatments to increase the throughput is the biggest cost reduction element of the system. But, we'd like to get away from carbon fiber itself and the inherent costs associated with the carbon fiber—if you want to call it "precursor"—material that forms the GDL, but we have not modeled, nor do I, at this time, know of any big breakthroughs or real possibilities to break that—I don't want to call it a "rut" but the track that we're currently on. There are desires to reduce the thickness of the GDL, and therefore, reduce its costs, and there are some schemes to try to do away with it entirely, but that's a bit of a misnomer, because it serves a functional purpose.

And so, the function will still remain, but it's possible, perhaps, to not have the GDL be a standalone layer, but rather to have it embedded or be the surface of the bipolar plate itself, and therefore, that's more of a transfer of cost from the standalone GDL fiber mat to a coating or an additive layer onto the bipolar plate itself. Hopefully, that raises some issues that are useful to the questioner.

Gregg Kleen

I guess along those lines a little bit here, do you have any comments on a comparison of metal versus carbon composite bipolar plates in terms of addressing the bipolar plate material cost?

Brian James

Yeah. So, we've looked at composite plates in the past, both compression molded/injection molded composites, which are mixtures of largely conductive carbon fillers and a non-conductive resin, as well as the old Ballard approach for the composite plates which are embossed plates, and frankly the—I'm forgetting Graphtec, I guess, is the technology I'm talking about—that has come a long ways in the last number of years. The killer with those systems was the resin cost, and they've found ways to get by with much less resin than they previously used. And so, the costs of that can approach the down—similar to the cost of the metal bipolar plates. There's not a big cost disparity as existed a number of years ago.

But, I don't know that there is a huge breakthrough. I wouldn't say that there's a huge price disparity between the two things on your sets of assumptions of how it cuts. But it's not a panacea.

Gregg Kleen

Okay. And I think—and here's another question. Does lowering the operating temperature increase the cost of thermal management?

Brian James

Yes, it does, by increasing the size of the radiator because the delta theta ambient is lower. And so, for the automotive system, that is a packaging issue because you have to put sufficient frontal area on the vehicle to reject the appropriate amount of [inaudible] power. With that lower delta T, it is less the concern for the bus and truck application, which generally have more real estate. You could also make it a thicker core—radiator core—for a fixed frontal area, but then, you have pressure drop and mass flow issues associated with that. So, there is a negative cost impact due to running at a lower temperature that is separate from just the cost impact that results from a different power density at a lower temperature.

Eric Parker

All right. Well, thanks, Brian, and Gregg for fielding those questions. If we didn't get to your question, please be sure to reach out directly and we'll try to get to them and add them as addendums to the presentation. But, I'd like to thank everyone for joining today and thank our presenter for an informative presentation. That does conclude the webinar for today.

If we—like I said, if we didn't get to your question, feel free to email the FCTO webinar mailbox. We'll also be sure to let you know when the full webinar and recording are available online and to sign up for our monthly newsletter for information on future webinars. Have a great rest of your week, everyone, and goodbye.

[End of Audio]