U.S. Department of Energy - Energy Efficiency and Renewable Energy

Fuel Cell Technologies Office

Automotive and MHE Fuel Cell System Cost Analysis (Text Version)

Below is the text version of the webinar titled "Automotive and MHE Fuel Cell System Cost Analysis," originally presented on April 16, 2013. In addition to this text version of the audio, you can access the presentation slides and a recording of the webinar (WMV 130 MB).

Alli Aman:
Thanks for joining today's call. Just a few housekeeping items before we get started. Today's webinar is being recorded, so the recording along with slides will be posted to our website in about ten days. You'll get an email from myself, Alli Aman, once those are posted. I also encourage you all to visit our website and sign up for our monthly newsletter. These webinars are held monthly, and we have a lot of other valuable information on those newsletters, so I encourage you to visit our website and sign up for those if you aren't already receiving those.

Also, I just wanted to let everyone know you are on mute or listen-only mode. So if you have questions, which we encourage that you ask, please submit those via the question function, and hopefully at the end of the webinar we'll have a few minutes to go over most of those questions.

Since today we do have two speakers, I just ask that you indicate whether your question is for speaker number one, Vince, or speaker number two, Brian. So if you can just indicate that, that'll help the Q&A go a little bit more smoothly at the end. And then I just want to thank you again for joining today's call. Just next month we will not be hosting a webinar, because we are having our annual AMR event. So in May there will be no webinar, but we'll be back in full swing in June, so I encourage you to check back to our website.

On that note, I'm going turn this over to Reg Tyler to introduce today's speakers, and Reg Tyler is a fuel cell technologies project manager with the U.S. Department of Energy Fuel Cell Technologies Office, and he is located in Golden, Colorado. Reg?

Reg Tyler:
Thank you, Alli. This morning, we have, as indicated, two presentations on cost analysis. The first presentation will be presented by Vince Contini. Mr. Contini has more than 10 years of experience at Battelle in mechanical and systems engineering for energy and healthcare products industry. His emphasis has been on designing, implementing, and testing in the area of product development. Mr. Contini has served as task leader and project manager for multidisciplinary teams in a role that is expanded from strictly technical to logistics, risk management, and strategy.

He has been issued 15 patents and several other patents pending. Vince has an MBA from Ohio State University, as well as a Bachelor of Science in mechanical engineering from Ohio State University as well. With that, I would turn the program over to Vince.

Vince Contini:
Thanks, Reg. I'm here to talk about a manufacturing cost analysis that we did here at Battelle on fuel cell systems for material handling applications.

[Next slide]

Very briefly, what we'll talk about today is we'll go over the background of why we did this study, the approach we took. We'll talk about the system design and what went into that, and more specifically, we'll talk about the fuel cell stack design. We'll then look at the stack, the balance of plant, and the system cost models, and go over a system cost summary, and then a summary of the results.

[Next slide]

So this is part of a five year program to provide feedback to the Department of Energy on evaluating fuel cell systems for stationary and emerging markets by developing independent models and cost estimates. This year, we looked at material handling equipment. The overall study is going to look at primary power, including combined heat and power, backup power, auxiliary power units, and material handling. The fuel cell types that will be included in the study include both low temperature PEM and high temperature PEM as well as solid oxide fuel cell technologies.

We'll look at annual production volumes of 100, 1,000, 10,000, and then also 50,000 but that will only be for the primary production systems. And the sizes that we will consider for the study include 1 kilowatt, 5, 10, 25, 100, and 250 kilowatts. The—what we did this year was look at 10 and 25 kilowatt PEM fuel cell systems for material handling applications, and that's what I'll be talking about today.

[Next slide]

Battelle applied an established methodology used successfully on previous fuel cell cost analysis studies that we've done for the Department of Energy, beginning with a market analysis, and then completing a system design. The system design is a very iterative process. Beginning with a literature search, and blending with past experience, we developed the entire system that would be needed. This is then vetted with industry, keeping in mind the necessary manufacturing processes, and modifying accordingly. We then moved on to the cost modeling, and finally perform a sensitivity analysis.

[Next slide]

So this is a schematic of the fuel cell system designed for the material handling applications. The system design was based on Battelle's experience with fuel cell system integration as well as industry feedback. The annual production volumes for this, as I mentioned before, are 100, 1,000, and 10,000 units, so that was part of the consideration that went into the design. The same basic design is used for the 10 kilowatt and 25 kilowatt systems, with the stack and balance of plant component sizes differing.

Part of the design—we had a decision to hybridize the system with a lithium ion battery to provide the peak power for the lifting. There were different energy storage devices that we considered, but ultimately from industry feedback, lithium ion was the one that we chose to use. Industry feedback also indicated that cathode humidification should be included. That was not part of our original design, based on some literature searches we performed. However, we got enough feedback from different companies suggesting that that should be included.

[Next slide]

So moving on to the system specification, you can look through here for the various parameters. The power density was .65, with the current density being one amp per centimeter squared. Catalyst loading was chosen at six milligrams of platinum per centimeter squared, with the cathode having a two to one relative to the anode. The active area and cell count chosen was to permit the use of a buck only DC to DC converter, and that was per industry feedback. The gas diffusion layer and membrane base material were purchased parts as part of our study, and were—and DFMA was not used on those.

[Next slide]

So when looking at the PEM fuel cell stack manufacturing process overview, the run size was based on a catalyst batch size, kind of the minimum that made sense to perform, and just to give you a sense, when you were doing 100 units, that only included one run per year. For 1,000 units, that included 4 runs per year, and for 10,000 units, it would include 12 runs per year.

The major difference in this compared to past studies that we've done was that we employed decal transfer versus using a dual slot die coater for the MEA process.

[Next slide]

So the methodology for calculating manufacturing costs, we used Boothroyd-Dewhurst estimating software, and employed standard processes where they existed, and developed custom models as needed. The custom models that we used were for the bipolar plate compression molding, the platinum catalyst manufacturing and application, the MEA hot press and die cut, and the silicone gasket die cut. We used standard models for the cell machining of the end plates and the stack assembly.

[Next slide]

So moving on to the major stack material process assumptions, some of the key things on this slide were the platinum costs of almost $1,400 per troy ounce. This is a difficult one, and this is a higher number that we've used in past studies, and what you've probably seen in other studies. This is a pretty volatile price, and the $1,100 number that has been used in the past just seemed too low for what's common today. And actually, over the past year since we picked this number, it's continued to go up, and just recently come back down a little bit.

The Nafion, the membrane, and the gas diffusion layer volume costs were based on DTI's 2010 report, as well as discussions with industry that we had. The catalyst loading we used was .6 milligrams per centimeter squared. This is pretty common for what you'd find out there today, but advances are continuing to drive this down, and the value used is on the conservative side. And we address this in our sensitivity analysis, so we'll talk more about that later.

[Next slide]

The capital cost assumptions are shown here. We've essentially used the same as we did for our 2007 report for backup power. There was no reason to believe that the cost had changed in any significant way.

[Next slide]

So looking at our 10 kilowatt stack manufacturing cost summary, you can see the breakdown here. One of the things that jumps out on this slide is that the material cost is really driving the cost of the fuel cell stack. This would be affected by reducing the catalyst loading. For example, if the catalyst loading was reduced from .6 to .3 milligrams per centimeter squared, that would reduce the material costs by about $500 per stack, or 13%, at 10,000 units per year. But the material cost would still make up almost 70% of the total cost.

[Next slide]

The balance of plant was surprisingly a big part of the cost of the system. Past studies have somewhat neglected some of these items, and they show up here in a pretty big way. You can see that over 50% of the balance of plant is in the top three components, which include the battery, the hydrogen tank, and the DC to DC power converter. We'll talk a lot more about this in later slides. The graph to the right is showing the distribution of the balance of plant costs when you eliminate those top three items. And one of the reasons for doing that is to compare to past studies, because those items have largely been left out on studies that have been done in the past.

[Next slide]

A summary of the fuel cell system costs. You can see for both—for all three of the volumes that we chose, the balance of plant dominates. One other comparison, when you look at the ARRA material handling equipment costs, they list $33,000 for a system in a class one or two material handling application. The one thing about this number, and I'm still trying to get more information on it, it's not clear exactly if that $33,000 is for what volume and what power level. The best I can tell, it's for a 5 kilowatt system. So if you look at our 1,000 units after markup, the system cost is about $38,000, so it's reasonable to believe that for slightly less than 1,000 units, a 5 kilowatt system would be about $33,000. So that feels about right.

[Next slide]

So another thing that has come up in conversations with the Department of Energy and others is a comparison of our study to automotive studies. There are a few things that make that difficult, because it's not exactly apples to apples, but this slide attempts to try to validate our study based on those past studies and look at how things are different.

So the top left, it shows DTI's 2010 automotive update and the key characteristics. One thing that jumps out at you is the number of cells is a lot larger. That's due to both the cell size and the fact that their system is a lot higher power. That's compared to 66 cells for ours.

Two of the other differences are—well, membrane power density, but that's actually—if you look at the current density, current density that we used was 1.0 amps per centimeter squared. That seems to be pretty consistent with what's available today. That is, however, being driven down or driven up, actually, to 1.2 or 1.5 amps per centimeter squared. And I believe—so the DTI study at 8—or .833 watts per centimeter squared is based on something closer to 1.2 amps per centimeter squared.

The other fact is the volume that we're looking at, we're really not getting into truly high volume manufacturing. With the 10,000 units that we tapped out at, we really are just getting into the lower part of the automotive study. If you look at the automotive study, manufacturing 1,000 systems, which was the smallest that they looked at, that would include 369,000 cells. And if you try to make that equivalent to the Battelle system, that would equal almost 5,600 systems.

Another—this—oh, I'm sorry. The other difference that I've already mentioned is the platinum costs. So if we change our assumptions to match closer to what DTI had in their analysis, you can see on the right there, on the bottom right, our stack cost ended up being $174 per kilowatt net compared to $159 per kilowatt net for the DTI automotive, so that's within 10%.

[Next slide]

The sensitivity analysis, this is where we took a look at if we made some modifications to our assumptions, how that would affect things. As I've already mentioned, current density is continuing to be pushed up by industry, and you can see that the top and the third biggest effect is with that current density.

The other thing was the platinum loading that we've already talked about. Going down to .3 from the .6 that we used makes a fairly significant difference, and that shows up as the second, and you can see how the other ones stacked up.

[Next slide]

Getting back to the balance of plant, the three dominant cost drivers, as we already discussed, are the energy storage, the hydrogen fuel storage, and the electronics and control. You can see what the breakdowns are for both our 10 kilowatt system that we looked at and our 25 kilowatt system. All three of these things are areas that are probably ripe for invention and cost reduction. I'll just make a mention on each one of them.

For the energy storage, one thing that's been happening in the material handling applications is they've taken existing lead acid batteries and they've used that space to put in a power—a fuel cell power system. So reusing the lead acid batteries for your energy storage is not really something that is feasible, because of the power density. So that's what forces you into the lithium ion batteries.

If you did design a fuel cell forklift from the ground up, one potential would be to use lead acid batteries, because they really don't hurt you on weight. In fact, they might be beneficial from a ballast standpoint. And that would allow you to save probably two-thirds of the cost for the energy storage.

Hydrogen storage, we use composite tanks. The—some of industry is beginning to use metal tanks. This is something that's probably not feasible in the automotive world because of the weight of the tanks, but as mentioned, the weight's not as big of a concern for a material handling application. The one thing that's holding that back is just getting acceptance and the testing that is necessary, since the composite tanks are kind of industry standard now.

Electronics and controls, as I mentioned, the DC to DC converter is a very large line item. The feedback that we've gotten from industry is they're actually even trying to eliminate that DC to DC converter, so that's where some invention on how to manage that could reduce the balance of plant costs in a significant way.

[Next slide]

Okay. I kind of already prompted this slide. I should have moved ahead. Sorry. This is taking a look at one of those examples, and when you look at composite hydrogen tanks compared to the all-steel hydrogen tanks, you can see that at the 10,000 unit volume, you can save over $2,600.

[Next slide]

Kind of just staying on the balance of plant here, the system cost comparison, the largest expense, as has already been mentioned, is the balance of plant hardware, and that makes up over 80% of the costs. Let's see.

[Next slide]

Kind of trying to wrap up here, so the—it seems like I'm beating a drum here, but the balance of plant costs are driving the system—the total system costs, and the potential for eliminating the DC to DC converter, eliminating stack humidification potentially, as well as using the steel hydrogen tanks, are all areas that could have a big effect.

The production volume, from what we've found on the volumes we've looked at, has a negligible effect on stack cost. This is largely due to the precious metal and the graphite composite and the commodity costs for—the commodity cost for cost and over the volumes we're looking at. One thing that would have an effect on this cost is the lower catalyst loading and higher current density. Both could lower the amounts of platinum necessary, and therefore lower the overall material costs.

[Next slide]

The proposed future work, this year we're going to complete the assessment of a one and five kilowatt solid oxide fuel cell system for auxiliary power unit applications. This is something that we started in fiscal year 2012, and did not finish, but we're going to be finishing that up this year, as well as updating our past assessment of backup power applications.

In fiscal year '14, '15, and '16, we'll be completing additional new analysis, and that'll be something that we coordinate with the Department of Energy, as well as revisiting and updating any of the previous analyses that we've done. So with that, I will turn it over to Brian.

[Next slide]

Reg Tyler:
Thank you very much, Vince. As we mentioned, we have a second presentation.

[Next slide]

The second presentation will be presented by Mr. Brian James, who is the director of energy programs at Strategic Analysis. Mr. James has 25 years of professional experience working on high technology projects and alternate energy analyses. Particular specialties include fuel cell power systems, hydrogen reformer systems, systems performance, and cost effectiveness analyses.

He has conducted numerous technical economic analyses and manufacturing cost analyses for the Department of Energy, NREL, and private industry. Mr. James has—was the 2005 and 2007 recipient of the DOE Hydrogen Program R&D Award, and holds five U.S. patents. He has a BS and a master's of science degree in aerospace engineering from University of Virginia and Virginia Tech, respectively. With that, I will turn it over to Brian for the second presentation. Thank you.

Brian James:
Thank you. So today I'll be talking about some manufacturing cost analysis methodology and its application to automotive fuel cell systems, and this is part of our five year contract with the Department of Energy to annually update the automotive fuel cell system cost analysis. Here we go.

[Next slide]

So in terms of an outline, I'll be talking about our purposes and goals, and then getting into a little bit about our cost analysis philosophy associated with the DFMA analysis. And I'll go over the general steps and then get into the basics of—or the details, I should say, of the 80 kilowatt automotive system, including cost results and applications to other systems.

[Next slide]

So for purposes and goals, they're perhaps on one level self-explanatory. It's an estimate of the total cost of the system when produced in quantity, paying special attention to the changes in cost with manufacturing rates. And when we're looking at automotive, we're looking all the way up to 500,000 systems a year, which is substantially more than the stationary systems that Vince just spoke of.

We also want to identify the key parameters that drive system cost and determine cost differences between designs and manufacturing processes. Importantly, we also want to force identification of the changes between what's commonly called a lab design, which would be a low production rate design, versus a true mass production design. And sometimes they're substantially different.

[Next slide]

We also employ a DFMA methodology. It comes from Boothroyd-Dewhurst, Incorporated. It's used by hundreds of companies worldwide, particularly in the automotive industry. We practice a blend of textbook DFMA, which is to say literally the textbook that BDI puts out. We license their software, as was referenced in Vince's presentation, and we also use extensive custom models of individual processes, because standard processes in the BDI software are not available there. Also, we have more—when one does your own—one's own internal model, then you have more control over exactly which parameters to change and more transparency into exactly what's going on.

DFMA is a process-based analysis, so it mimics the actual manufacturing across the steps. We can use that to reflect the impact of the rate of manufacturing, so we can put in different material costs at different rates, select different manufacturing methods, pay attention to the machinery, which I'll talk about in a few minutes, and of course amortize the tooling over the expected lifetime of the tools.

In its essence, the costs that we're predicting here are merely the summation of the material costs, the manufacturing costs, and the assembly costs with an appropriate markup.

[Next slide]

Per DOE directive, when we do the automotive designs, we include what's shown in blue down at the bottom, which is a combination of fixed costs and the variable costs, including factory expenses and the direct material and direct labor. That goes into the costs. What's not included in the analysis are these terms up here in the lighter brown color—profit, one-time costs, which is non-recurring engineering, general expenses, such as warranty, advertising, taxes, and G&A. So these are typically not incorporated in our analysis, and they different from the numbers that Vince just showed you. So what we're reporting today is cost rather than OEM price, and there is of course a substantial difference between cost and price.

[Next slide]

The basic cost modeling workflow is similar to what Vince described. You have to create a system design for the overall system to ensure full functionality. One develops a bill of materials, which is the full listing of the components used in the system, as well as an image of their physical embodiment of each component, so you have to understand the dimensions, the materials, and how it all fits together.

Then you go into the—specifying the manufacturing and assembly process for each of the components. You tabulate the cost results and conduct sensitivity analysis. Typically, we do tornado analysis and also Monte Carlo analysis, and I'll speak more about those in a few minutes. And very importantly, and reinforcing the iterative nature of the design that Vince spoke of earlier, one needs to vet the results with experts and then incorporate the feedback that one receives.

[Next slide]

So let me get a little philosophical here on you and tell you what mindset we use here at Strategic Analysis when we're doing DFMA cost analysis. It's a process-based cost analysis, so one needs to break down complex systems in understandable, simple steps. Basically, you can gather quotes for commodities, which are going to be multiple buyers and sellers, and therefore like a rock bottom price anyway. And you use process-based analysis for everything else.

It's all about specifying the details and knowing which details to specify. So you can do a very in-depth analysis, which takes longer, but it gets down into the very detailed level of analysis. Or you can do a more conceptual top level analysis with top order assumptions. But in general, one wants to make many small assumptions rather than a few big ones.

We're not afraid to estimate values, but then when we do that, we have to do sensitivity studies to see whether the values that you're estimating are really critical components or critical impactors on the bottom line, or really it doesn't matter, so you can spend your time on the important parameters, getting those exactly right. There's a heavy dose of innovation in all of this, because you're in most cases, particularly with fuel cells, envisioning processes that are not currently in use or not—or they're extrapolations of current processes.

One should always apply the concept of continuous improvement. So you try something, and then you try something else, and you compare the cost results. Factory robots are very commonplace in industry today. They're surprisingly inexpensive, and very fast, and I have an example in a minute on that.

[Next slide]

Another philosophy is to only pay workers for the time they work. So if you only need a worker for four hours a day, you only pay him for four hours a day, so he becomes a part-time employee, or you can job—cobble jobs together to fully employ him. But you're really only paying him for the time that he actually spends working. When you're unsure of what to do, analyze all the pathways, and then let the cost results guide you. Consult vendors. They're a wealth of information.

We base everything on the automotive industry, which typically has a seven-hour productive shift. So full-time is defined as 2 shifts of 14 hours productive per day and 240 days a year. I've already addressed the difference between cost and price.

And then the three most important costing parameters are cycle time, which is merely the time it takes for a machine to produce an individual part: for instance, a stamping machine, it would be how many seconds it takes for a part to come off the line. Capital costs, which is self-explanatory. And then this machine/line utilization, which is the fraction of time that the machine is actually productively employed. And by way of explaining that utilization as well as the machine rate, I'm going to give you an example here for a modern single-armed robot.

[Next slide]

Processing cost is merely the product of the machine rate in dollars per minute times the cycle time, the time it takes to perform the operation. And this machine rate is defined by this equation, which is the annual capital repayment, based on the installed capital costs. So you have a base cost say for a robot, which is one of these two on the right, plus some customization cost, plus a cost factor for installation. And then you hit it with this capital recovery factor, which is the percentage of the capital cost that must be repaid each year to repay the cost of the capital expenditure over the lifetime of the device.

So I have some typical parameters here in terms of including the discount rate, taxes because you need to make—be making this repayment as an after tax payment. And importantly, the equipment lifetime factors into this. So this is basically a fixed cost. Then we have these annual operating payments, which largely are dependent on how much you actually use the equipment, which is based on utilities and plus the labor cost associated with it, and you can use fractional people, which would be one person minding multiple machines, typically.

Plus, you have a percentage of capital costs typically for maintenance and spares, as well as miscellaneous expenses, and these are historically derived numbers based on other machinery. They can be customized to the machine or just used as standard value as representative for a value that is perhaps unknowable.

So down here on the right I've graphed machine rate in dollars per minute versus the average operating hours per day for the maximum 14, which is the automotive standard, 14 hours per day, 240 days a year. Obviously, if you don't use the machine very much, then its operation time in terms of dollars per minute is going to be high. The more you use it, the cheaper it gets per minute, okay—which is not necessarily—fixed costs don't go down, but per minute times do.

And I've plotted it against the worker-only effective machine rate. This is $45 an hour, same as what Vince spoke of before, what a worker would cost, assuming he had the same cycle time as the robot in question here. So we see that at 4.5 hours, if you're going to use the worker or the robot less than 4.5 hours a day on average, then it's less expensive to employ the worker, and at greater than 4.5 hours, it is less expensive to employ the robot.

[Next slide]

Getting into the automotive system, we have an 80 kilowatt electric automotive system diagram shown here. It's a single PEM stack with a cooling loop shown in purple, and in orange a hydrogen recycle loop shown at the top. Contrary to the stationary analysis that Vince spoke of, we do not include the hydrogen storage tank, or subsystem I should say, in the cost analysis per DOE directive.

On the right hand side over here, we have the reactant air coming in. It goes through a compressor, a pre-cooler to cool it off after compression up to 2.5 atmospheres. It goes through a membrane humidifier, which we do have in our system, and then through the stack, and then out through an expander. So this unit over here is the CMEU, the compressor motor expander unit, which is an integral part of the overall system, and turns out to be a significant cost factor.

[Next slide]

I have a similar diagram here for a 160 kilowatt fuel cell bus diagram. It's very similar. This is for a 40-foot-long transit passenger bus. It has a common cooling loop and a common airstream, but we have dedicated hydrogen recirculation loops for each of the two fuel cell stacks.

Now there are lots of details here, and I won't go into them, but the key differences between the systems are the bus typically operates at a lower pressure, so we don't have an exhaust path expander here, and of course, it's at higher power, so we have two stacks instead of one.

[Next slide]

This next chart details the—some of the parameter differences between the automotive and the bus system. Once again, I won't go into the details, but I'll just point out the chief differences between the two. Because it's at a lower pressure, nominally 1.8 atmospheres versus 2.5, we don't use an expander. We have higher power, so it's bigger stacks, as I indicated before. And we typically have higher catalyst loading, because we want longer lifetime out of the fuel cell stacks than we need out of the automotive stacks. This together results in a lower power density in terms of milliwatts per square centimeter of the bus versus the automobile. This is—these are largely design choices. They're not intrinsic values associated with the bus versus the automobile, but they represent typical values for those applications.

[Next slide]

The fuel cell stack is schematically shown here. Part of it is the membrane electrode assembly, the MEA, shown in the blue with a frame gasket around the outside, and then with the various flow plates to channel the hydrogen and oxygen. This has a one to one ratio of cooling cells to active cells.  

[Next slide]

By way of example of how we do the DFMA, I've shown the MEA assembly, the membrane electrode assembly, here from a 3M presentation drawing. In the orange is the membrane itself, and I draw your attention to the blue gaskets that are on the outside. Obviously, you need to seal around the edges of the membrane itself to prevent gas from leaking out, and the bipolar plates would press up against the subgasket.

This particular design—I'm showing the physical embodiment that we modeled in 2012—has two layers of subgasket with an adhesive layer before, and they sandwich the membrane on the edges. So this is an example of the—defining the physical embodiment of it. And next you have to envision how it would be produced.

[Next slide]

So what we did is we went to the patent literature and found a 3M patent that represents a roll-to-roll process for putting those subgaskets around the active area of the membrane. So then we went through this, and we estimated the cost of the process line as well as its energy use and its machine rate—or not machine rate, but line speed—to come up with an overall machine rate, and then use that in our cost analysis.

[Next slide]

Now two years ago, in 2011, we used a frame gasket approach to bracket the MEA assembly, and this was an injection molded plastic piece, basically, around the MEA, and it resulted in a projected cost at 500,000 systems per year of roughly $3 a kilowatt. So in 2012 we came in with this new roll-to-roll process, which had the roll-to-roll processing addition I just showed you on the previous page plus a screen printing step for that adhesive between the two layers of subgaskets, and it resulted in an overall cost of roughly $1 a kilowatt, saving $2.30 a kilowatt.

So this shows both the physical embodiment and how we—each year of the analysis we can go in and reexamine it and potentially come up with a lower cost approach, either through refinement of the analysis or through advancement in the technology.

[Next slide]

Jumping into the cost results, once again, at 500,000 systems a year, we have the stack cost down the left, which is dominated by the catalyst ink and application, and indeed, the majority of this cost in the reddish color is catalyst material cost itself. So the membrane cost and GDLs are actually a relatively small fraction of the overall stack cost, which is roughly $1,600, and adds up to about $20 a kilowatt. Now this obviously isn't the only cost in the system. There's a substantial balance of plant cost, which is dominated by the air loop, which in turn is dominated by the compressor motor expander unit that I spoke of before. So the sensors, the fuel loop, and the coolant loop are also substantial cost contributors to BOP.

And the BOP is actually more expensive than the stack, contributing a little over $2,000 to the 80 kilowatt system, which works out to be $27 a kilowatt. So the grand total for the system, once again, at 500,000 systems a year, is $47 a kilowatt.

[Next slide]

We did single variable sensitivity via tornado chart and went through each of the suspected important parameters and identified each of them, what their low value and their high value range would be for each of the parameters, and then independently assessed them. And we see here that variations in power density lead to the largest cost decrease as well as cost increase, followed by platinum loading variations. Now these are done individually, single variable, one at a time, and obviously power density and platinum loading often could influence one another.

[Next slide]

We also did multi-variable cost analysis via Monte Carlo, which is a probabilistic technique. We went through each of the important parameters, assessed a minimum and a maximum value, and then assigned a probability distribution function, PDF, typically a triangular distribution for each parameter. Then we probabilistically simultaneously varied each of these in about 50,000 runs to come up with an overall probability distribution function for the overall system cost. And we see here that the highest probability is the cost resulting in around $47 a kilowatt, which matches the previous values we showed.

But more importantly, what we can do is we can look at the tail ends, the low end and the high end, and chop off the bottom 5% of the possible outcomes and the top 5% to establish a middle 90%. So this becomes our middle 90% confidence range, where we can say with 90% confidence that the cost will be—of the system will be between $43 a kilowatt and $52 a kilowatt. That's how we use the Monte Carlo analysis.

[Next slide]

We can also plot the curves as a function of annual production rates. We spoke earlier—I spoke earlier of the 500,000 values, and here's the $20 for the stack in the red curve, and the $47 per kilowatt for the system values. And as you go down in manufacturing rate, you go up in system cost. But the knee of the curve is surprisingly low, which is promising, meaning that you can get most of the cost reduction of economies of scale at relatively low manufacturing rates.

We've been doing this analysis for many years now, so going back to 2006, we had an estimate of the overall system cost, and we can plot each year's analysis. Back then, it was around $100 a kilowatt at 500,000 systems a year, and it got better every year, going down to $47 a kilowatt in 2012. So some of this cost projection reduction is due to advances in the state of technology, which is good. We definitely want to be tracking that. And a portion of it is due to refinement of the design and refinement of the analysis. But the end result is our projection improving every year and going down to $47 a kilowatt.

[Next slide]

Now I mentioned the buses earlier, the 40-foot transit bus. We did the cost analysis in a very similar fashion as just described, except the single annual production rate that we looked at was at 1,000 systems a year, because buses are sold in the U.S. at only roughly around 4,000 a year. So 1,000 represents a representative production rate for a single line, and in fact, that's somewhat at the high end itself.

The cost result was roughly $200 a kilowatt at this 1,000 systems a year for the bus, and we want to compare that to the DOE targets. However, the DOE targets include the battery and control system, whereas ours do not include any batteries, they’re just the fuel cell power system alone. So it's not really fair to compare the DOE targets with our $200 a kilowatt, plus the DOE targets are at 400 systems a year versus our 1,000 over here.

So what we did is we looked back at the literature and we saw that UTC Power back in 2010 had projected $200 to $300 a kilowatt in thousands per year of production, so we graphed that in purple, and we see that the Strategic Analysis 2012 projection for the bus is not out of line with the UTC 2010 projection. We've also looked at a variety of other systems that I'll just give you a snapshot today.

[Next slide]

We also looked, similar to Battelle, at the stationary fuel cell power systems with three different technologies in line, low temperature PEM, high temperature PEM, and solid oxide, at 1, 5, 25, and 100 kilowatt systems, and we were able to graph the results there.

[Next slide]

We also looked at high pressure hydrogen storage vessels. These are the automotive variety that would go on the vehicle to hold the pure hydrogen gas. They are—this particular—these particular results are for 70 megapascals, which is 10,000 psi, and it's to hold 5.6 kilograms of hydrogen in a type 4, which is a plastic lined carbon fiber wrapped vessel.

So we immediately see that at 500,000 systems a year, that this green color is the composite material, and it's dominated by the actual cost of the carbon fiber itself. Manufacturing of the fiber winding, it's this orange color, and it's actually surprisingly small.

The other big factor is this purple color, which is the balance of plant component. So these are the valves and pressure reducers contained in the system. And while they're roughly 25% of the system cost at 500,000 systems a year, down at 10,000 systems a year they're more than half. So they really are a major cost factor in the overall system.

[Next slide]

So in summary, we used process-based cost analysis to do the cost estimation at multiple manufacturing rates and assessed key cost drivers, and to iterate on various designs, both cost—system design as well as processing, which is to say manufacturing process, and see which is the lowest. And you can apply it at a high level of approximation or a very high level of detail. And the 80 kilowatt systems were nominally $47 at 500,000 systems, and the bus size system was nominally $190 or $200 a kilowatt at 1,000 systems.

And finally, in summary, the standard outputs from this class of analysis is system schematic, component designs, cost variation with manufacturing rates, tornado sensitivities for single variable sensitivities, and then Monte Carlo analysis for multi-variable sensitivity, showing the 90%, or really any other percentage you want, confidence cost range.

So with that, I'd be happy to take any questions. Back to you, Reg.

[Next slide]

Reg Tyler:
Okay. We do have some questions, and we do have some time for some questions. So if you have a question, be sure to type them in, and we will try to get to them. The first question is for Vince, and the question is why use spray coating? It doesn't seem like any of the major MEA suppliers are using spray coating. Vince? Did you hear the question?

Alli Aman:
Oh, you know what? It looks like he just got disconnected.

Reg Tyler:
Looks like Vince may have gotten disconnected.

[Crosstalk]

Brian James:
I can field that.

Reg Tyler:
Let's go to the second question, and maybe you can answer this, Brian. What are the reasons for declining cost as you produce more of the item?

Brian James:
They—it's several factors. One, you get economies of scale in terms of purchasing commodities, so the actual raw materials often are less expensive, and that's particularly the case with the membrane ionomer, as well as the—I guess the purchased membrane, if you were purchasing the membrane.

But more importantly, you get better utilization of equipment. So when we do a range of manufacturing rates, we will postulate different sizes, different pieces of equipment, and in general, the bigger the equipment, the faster the through rate, as long as it's operated at high utilization rates, will be lower in cost to operate, so it'll be more productive. So in general, you are better utilizing the capital. You get better capital utilization, which leads to lower price—cost.

Reg Tyler:
Okay. Thank you. Vince, are you back on?

Vince Contini:
I am. Sorry about that. The mute button's a little too close to the speaker button.

Reg Tyler:
Okay. Let me repeat the first question for you. Why use spray coating? It doesn't seem like any of the major PEM MEA suppliers are using it.

Vince Contini:
Okay. Actually, yeah, Fritz Eubank is with me, and he's going to try to address that.

Fritz Eubank:
The—we had originally set up the model for spray coating, and the more we started looking into other direct application methods, it became—for example, on slot die coating used to go onto the MEA directly, apparently, there were some problems associated with the fact that the MEA is very fragile. And so consequently, it seemed at the time that spray coating might be a good alternative, just because it preserved the integrity of the MEA a little bit better. And there were some—there was some information coming out of Europe that seemed to indicate that spray coating was a method that was used over there with high success. In particular, the spray coating method we modeled up was an oversonic spray, which is very accurate at laying down patterns and thicknesses.

Reg Tyler:
Okay. I have another question, then. It's actually addressed to both of you, but I'll address it first to James—to Brian. In either analysis, has there been any consideration of the cost impact of designing in cold start capability? Brian, do you want to comment on that?

Brian James:
Right. To—the short answer to that is no. There is no right now dedicated element of our cost analysis specifically for cold start capability. And whether you're—I guess the expectation is that the startup time would be very fast, and that these systems are going to be hybridized, so there would be a battery associated with them, and therefore they'd be able to boot strap up on the battery for the short duration it takes to get them warmed up.

Reg Tyler:
Thanks. Vince, another question for you. Vince, please comment on the commercial viability of fuel cell-based forklifts in light of the recent financial challenges that Plug Power has faced.

Vince Contini:
I can't specifically comment on what Plug Power has faced. The—I think the challenge will continue to be reducing the cost of the systems to make the value proposition even stronger than it is today. I mean, there is a value proposition out there that makes sense in some applications, particularly when it's high usage. And that's another thing that we've done, is a life cycle cost analysis, but I didn't have time really to go much into that today.

I guess the other comment would be—well, I think in addressing those costs, you know, attacking the areas that are going to have the biggest impact on getting that cost down. Reliability and getting people to be more familiar with the technology and, you know, hydrogen infrastructure, are all things that would weigh in.

Reg Tyler:
Okay. Thanks, Vince. A question for you, Brian. Do you plan on—plan to pursue the simplified auto system design in which a humidifier is eliminated altogether? What is the industry trend with respect to elimination of the cathode humidifier?

Brian James:
We, as will be shown in the AMR next month, we just updated our humidifier analysis, which is based on a plate frame humidifier as opposed to a tubular humidifier that we used before. And the projection at 500,000 systems a year is $80 for the—for the entire humidification—humidifier, excuse me. Not $80 a kilowatt. So it's be $1 a kilowatt, $80 overall.

So with the cost of the humidifier being relatively low, I'm not sure there's a huge clamoring to eliminate it, though of course if one can show longevity and good performance without the humidifier, then it—I'm all for system simplification. But you're not going to save a huge amount of money by eliminating it.

Reg Tyler:
Thank you, Vince and Brian. I think we'll wrap up the questions with that question, and I'll turn it over to Alli to wrap up the webinar.

Alli Aman:
All right. Thanks, everyone, for joining us. A special thanks to Vince and Brian for taking time out of your schedule to host the webinar today. We greatly appreciate it. Just a reminder to everyone on the call—this webinar was recorded, so we will have the recording as well as slides up on the website in about 10 days. I'll send an email out as soon as those post to our website. And again, I encourage you all to check back to our website and sign up for the newsletter that comes out monthly, and then we will also be posting future webinars there, so I encourage you to look back—or check out those as well.

So thank you again, Brian and Vince, and thank you, everyone, for attending.

Brian James:
Thank you, Alli.

Alli Aman:
All right. Thanks, guys. Bye bye.

Reg Tyler:
Goodbye.