Simulation and Modeling of Bioenergy Conversion from the Atomic to Process Scales Webinar – Video Text Version
This is the text version for the Simulation and Modeling of Bioenergy Conversion from the Atomic to Process Scales webinar.
>> Moderator: Hello everyone and welcome to today's webinar, Simulation and Modeling of Bioenergy Conversion from the Atomic to Process Scales. Before we get started, I'd like to go over a few items so you know how to participate in today's event. You're listening in using your computer speaker system by default, if you would prefer to join over the phone, just select telephone in the audio pane of your control panel and the dial in information will be displayed. You will have the opportunity to submit text questions to today's presenters by typing your questions into the question pane of the control panel. You may send your questions in at any time during the presentation. We will collect these and follow up with you after today's presentation. Our presenters today are Seonah Kim, Peter Ciesielski and Jim Parks.
Now I will turn it over to Jim.
>>Jim Parks: Hello everybody. I'm Jim Parks from Oak Ridge National Lab and I'm happy to present to you today activities we have in computational science related to ChemCatBio and the Bioenergy Technologies Office. We will be giving a tag team presentation and very much a collaborative presentation and also a collaborative effort in modeling of this program.
So as you can see from the title slide there's quite a few people involved in this and there are five national labs representing this part of the program.
So first I would like to talk about the Consortium for Computational Physics and Chemistry from a high level view. We are five national labs, Oak Ridge National Lab, Pacific Northwest National Lab, the National Energy Technology Laboratory, the National Renewable Energy Laboratory and Argonne National Lab, and we work together to perform modeling and simulations to support the many activities in the Bioenergy Technologies Office at the Department of Energy, and today we're going to be specifically talking about our support of the ChemCatBio program as we're sharing this info for this ChemCatBio webinar.
We also support other activities in the program such as biopower, biochemical process modeling and simulation, the Feedstock-Conversion Interface Consortium and the Bioprocessing Separations Consortium.
Today we'll be talking to you about three different areas. This is kind of how we organize our team. We have modeling activities going on at the atomic scale where we're doing modeling of catalysis at the very molecular level. We also have modeling at the meso-scale where we're actually modeling catalyst particles and the different processes that go on at that scale. And finally we do modeling at the process scale, full reactor systems.
And so today we're going to give you some examples of the modeling work that is going on at these three different levels and we'll have Seonah Kim talk about the atomic scale, Peter Ciesielski will talk about the mesoscale, and then I'll follow up with process scale modeling. In addition to this webinar, if you'd like to know more about our consortium, you can check out our website, which is www.cpcbiomass.org.
So now I'm going to turn it over to Seonah who will talk about atomic scale modeling.
>>Seonah Kim: Thanks so much, Jim. I'm Seonah Kim from NREL and I will present atomic scale catalyst modeling that we are doing in this part of the CCPC activities. So when—we've tried to guide compositional catalyst design using multiple atomistic toolkits. So starting from—using very high level theories of, for example [inaudible] to the atomic active sites of that study, and also we add some kind of the confinement to reactions and the study of them. And also we expanded to include the surrounding medium when using the MD simulation.
So when we start we have some catalyst substrate system and through the mechanism study we do the thermodynamic and kinetics study, and including all of these kind of information, we do the screening of the catalyst and read the activity and selectivity catalyst to info, to test in the experiment. This is one like as the group of the work and then we do the—we use a lot of this background experimental group and vice versa. So we do like a grouping game for that inside of that CPCC activity.
So our ultimate goal for that atomistic simulation team is to support the ChemCatBio, first we want to understand catalyst activity and selectivity as a function of the composition of the catalyst, and also we want to guide new catalyst synthesis and the work with ACSC team inside of the ChemCatBio. And finally, not only the design in the catalyst active drive, we also want to optimize the operating conditions like temperature, impression, and etc.
How we do the atomistic modeling to achieve these goals. We do the modeling for both homogeneous and heterogeneous catalyst to understand the reaction chemistry. Also we also explore the catalytic cycle with the nature of the active sites to understand the structure and function relationship. We also expand to exploring the molecular-scale transport, for example using the molecular dynamics simulation. And finally, we also screen the catalyst to guide new synthesis using the machine learning, linear regression and [inaudible].
So this is what we want to achieve the goal to design the performance-advantaged catalysts. Here I want to share a few examples that we are doing right now. The first example that I want to share with you is develop the bimetallic catalysts for selective carboxylic acid reduction—reaction. So along with a lot of people like the experimental group collaboration, we have tested three different surfaces, tin, ruthenium, and tin oxide on top of the ruthenium. For the tin, ruthenium and—ruthenium surface, you see a lot of side effects along with acid and carboxyl acid reduction with high energy barrier.
However when we use the tin oxide surface, this reaction is becoming very selective for only the carboxylic acid reduction with very low activation energy barrier. So this is what we did to rationalize the catalyst disparity and also probing the mechanism for this reduction reaction. And then this detail, we try to evaluate using experimental tin and also we try to tune a little bit of the design parameters to include that catalyst activity and reactivity and then we tested in our [inaudible] and experimental group.
Our second example that I want to share is to understand water deactivation of the metal oxide catalyst. We chose magnesium oxide because this is very well known crystal structure they have and also it's a relatively a lot simpler than other metal oxides. So our goal is to understand the water deactivation and also reactivate the activity of the catalyst.
So first we started with the core deduction [inaudible] diagram and shown in the left hand side here and then to see the reaction between the surface and reactants here. Using this information, we studied the reaction energy differences of the metal oxide. Our target reaction is aldol condensation here. At the same time, not only the reactivity we studied, we also studied the water tolerance and also catalyst deactivation.
And the third example that I want to share is ethanol upgrading over [inaudible] for the zeolite here. We studied not only the internal of pore, we also studied the external of the pore and then think about what is fact, a critical fact, for the enthalpy and entropy contribution to understand the Gibbs free energy here.
So you see here in the left hand side graph, you see there is some difference between the external and internal branch of active sites for the enthalpy and also free energy. And also they can be a [inaudible] to reactive rates [inaudible] some also.
Here in the right hand side, we tested a single branch of active sites versus double branch of active sites showing the improvement both internal and external pore, and also we try to decrease the temperature and then you see a lot more enhancement for both internal and external pore site pore. So we concluded like the thermodynamic profile is tuned by increasing the branch of active sites and also lowering the operating temperature. So that is what is a good kind of a point to the experimental group.
So we also, as I mentioned earlier, we also did some MD simulation for this kind of work and I want to share microporous versus mesoporous zeolite catalysis work used for that diffusion because diffusion is very important in biomass upgrading, which impacted the coke formation and of the process of selectivity and yields and separations like that.
So the first—the top image I'm showing you here, this is microporous in ZSM-5, and then we compare p-xylene versus toluene here in this figure. When you see the toluene case, they reorient a lot more and they pass through the pore to the next adjacent pore site pore. So it diffused a lot slower than p-xylene here. We also calculate the diffusability of the coke precursor. Here we use the benzene, naphthalene, and anthracene as the example in both microporous and mesoporous here. In mesopore we use the two different mesopore sites here.
And then this is one of the examples in benzene diffusion in mesoporous, how they work from mesoporous to the micropore here. And then this is another motivation we studied for the coke formation work. We think the coke formation is also very important inside of zeolite and also external surface. So the first example is this—is we ran a lot of experiments here inside of the micropore to see how they formed and/or chain growth here.
So they cannot make a lot of [inaudible] inside of the coke formation here because of the pore size, so we are thinking to move onto the external surface here. This is very preliminary, so I'm just showing you the [inaudible] to how they can form from chain growth and then ring formation and then PAH formation here like this.
So now we want to incorporate this work in [inaudible] external surface of the zeolite catalyst and then see how the coke formation formed. That is what we have here. So this is where we are in the atomic scale modeling work and then I will hand over to Peter Ciesielski for the mesoscale catalyst modeling.
>>Peter Ciesielski: Thanks Seonah, that was really good and she set up—you did a good job setting up a lot of the stuff that I want to talk about. So I guess I'll just motivate mesoscale modeling is kind of abstract at first if you're not used to hearing about it.
So whenever you go and measure experimentally the kinetics by a bench scale or small reactor scale experiment, really the observed reaction rate is the function of a lot of different things. So there's the intrinsic reaction kinetics that are really how quickly a chemical reaction forms once the reactant has reached inactive site, how quickly that reaction can proceed. So that's what we'll call the intrinsic reaction kinetics. And then also that's delayed by the time at which it takes for a reactive molecule to diffuse into a catalyst particle and reach an inactive site, so that's a delay that also manifests in the observed reaction kinetics.
And then finally—so that's really a diffusion process and Seonah did a good job of talking about the importance of intra-particle diffusion. There's also extra-particle mass transport limitations. For example, the reactants and products diffusing through a boundary layer that sets up a round of catalyst when it's a velocity field, and we're doing a lot of calculations recently that highlight the importance of that process.
So all of these things together play into what you're measuring experimentally. And so mesoscale modeling is one way that we can handle these things explicitly and it's a great tool we've found for interpreting experimental results.
So I'll start by showing you what a zeolite catalyst particle looks like. This is an XET scan and you can see kind of the porosity of this particle. So that's one factor we can consider and then if you take a cross-section of this in the TEM, you can see the porosity where the catalytic sites are kind of bound together with binder. If you zoom way in on that eventually you can see the microporous structure of the zeolite where the active sites actually reside. And so transport through each one of these blank scales is significant.
And then the other thing that it's important to remember is that we've got down here an optical micrograph of a catalyst particle that's partially deactivated from a reactor here at NREL. So you can see how it's sort of deactivates from the outside in, and so at any given time a fraction of the catalyst sites are active and a fraction of them are deactivated and it's really a function of transport throughout the catalyst particle and the residence time of a catalyst particle within a reactor. And so all these things provide opportunities to optimize not only the catalyst design, but also the operating conditions to provide the most optimal interaction between catalyst particles and reactants.
So our approach to doing this is we use very detailed characterization of catalyst particles to measure directly the porosity, and this is, again, with image analysis that we perform on XET reconstructions, and then this particular example of the catalyst particles that we were working with have two pretty distinct porosity domains where the outer region has much more void space and the inner region is much more dense.
And so, as you might imagine, transport through these different regimes, is quite different and we can model that explicitly and also we can use approximations for transport and porous media, which actually reduce the computational load substantially. And so there is kind of a finite element domain or finite element model here showing how we model this and then there's a few conservation statements there are the bottom that we use to solve this in a finite element type format. I won't belabor the math today.
So this is for a zeolite catalyst particle and I think Jim will talk more about later how we use these types of simulations to extract kinetics that are independent of transport phenomena, but I just want to show you what one of these simulations looks like. So when I start this movie you can see the experimental data that are plotted on the left, and then we'll plot the line of what the simulation predicts through those data. And then on the right is just visualizing the coke formation and product evolution that's inside one of these particles.
I should say that these data were actually optimized—the simulation was in fact optimized for the simulation results, so this is an example of where we used the mesoscale simulation to extract intrinsic kinetics. So as this movie plays you can see we're able to predict, spatially and temporally coke formation within the catalyst particle. And I guess the other interesting thing that we've learned from this simulation is that if you look at the very last time step, which is where the movie stopped here, you still have a lot of product that's trapped within the catalyst particle.
So in a reactor scenario this catalyst particle would then go on to regeneration cycle where all the coke has burned off, and so you can also estimate how much of your desired product you're essentially forfeiting at that point, which is again another opportunity to optimize your operating conditions at the reactor scale.
So we also, like I mentioned, this work is tightly coupled to advanced catalyst characterization and we're working with the advanced catalyst synthesis and characterization team, particularly Kinga Unocic from Oak Ridge National Lab who uses advanced microscopy and chemical imaging to help us validate, for example, our predictions of coke formation in these catalyst particles.
And so finally I just want to cover another example of this where we are looking at supported catalysts. And so this is in collaboration with our experimental partners at PNNL, particularly the group of Rob Dagle. They're using SBA mesoporous silica to support a catalytic system that converts ethanol to butadiene. And this is a nice platform in terms of modeling and also in terms of catalyst design because it's quite a tunable system.
And so the mesoscale really covers from the atomic scale and it builds a bridge up to the reactor scale that Jim's going to talk about later. We're able to handle all the important transport phenomena that occurs essentially in that regime. So in this project we started at the atomic scale and we did molecular diffusion simulations, very similar to what Seonah Kim just described to you in the previous section where we able to compute diffusion coefficients or reactants and products within the mesoporous structure. We then take that and tessellate it into a larger domain and then we apply the effective diffusion coefficients for the different locations and then we're able to do finite element, again, finite element diffusion calculations over a larger scale and then we can incorporate that again into a particle scale and finally into a reactor scale.
I will walk you through sort of each one of those scales in the next few slides. So these are the diffusion calculations where we are able to calculate diffusion coefficients for reactants and products in the different kind of regimes within these catalysts. So within the necks that connect these larger cages in the SBA 16 structure. I guess the reason why that's significant is you see a lot more surface effects where the molecules are confined within those narrow regions. The diffusion within the larger cages actually begins to look more like bulk diffusion, but we're able to compute directly what transport through those different regimes would look like.
So then what we do is we take those diffusion coefficients and we apply them in a finite element simulation. So in the cage domain we use the cage diffusion coefficient and in the pore domain we use the pore diffusion coefficient, and then from there we are able to apply concentration gradient across this and now we're doing finite element calculations, and from there we're able to compute effective diffusion coefficients that take into account both regimes and then apply that, which will allow us to do reaction diffusion simulations at much more larger length scales.
That's great, that's what we need because the experimental data that we're using was collected in a packed bed, so now that we have effective diffusion coefficients that essentially account for molecular level and mesoscale transport, we're able to overlay that with chemical reaction kinetics and fit directly—compare directly against experimental data and we can get very nice agreement and we can also extract chemical reaction kinetics for different systems.
So we're solving conservation equations again for, in this case, just math and momentum. So we assume this to be isothermal, although it's easy to incorporate heat transfer if that's significant; in many cases it is. And then we use, again, approximations for transporting porous media to account for transport not only through the gaps between particles in the packed bed, but also, as I mentioned, through all the different porosity regimes within these catalyst particles.
Okay and then what's nice about this is that once we have a parameterized simulation, we can begin to do sensitivity analysis for parameters that are easy to control across different length scales. So once nice thing about SBA 16 is you can tune the cage size, you can tune the pore size that connects the cages; that's sort of at the nanoscale. And so we can do parameter sweeps over that to see how that affects the performance of the catalyst.
So essentially we asked our experimental partners, "What catalyst attributes are easy for you guys to tune?" And they told us particle size and they told us changing the pore size. And so we do parametric sweeps over that and we're able to provide them predictions of how changing those things is going to impact the performance of their system. And this is kind of a nice tidy analysis because at the end of the day we can provide them very actionable predictions.
And so what shown there in the bottom right, it's a simple as reducing the catalyst particle size and increasing the pore size, our models predict, won't result in improved catalyst activity lifetimes. And so those are experiments that are going to be performed hopefully next year. So, yeah, I think that's all I want to say. And now I'll pass it over to Jim.
>> Jim Parks: Thank you, Peter. So we'll finish by talking about the process scale model. Of course, as you can see, there's been a lot of research in supporting the catalyst innovations in ChemCatBio, and then understanding the phenomenon at particle scale as well. And we also need to do this across the scale. Our ultimate goal of course at ChemCatBio is to support the bioenergy industry and they have reactors in the private sectors, and we're also working towards that goal to support those efforts, and we have reactors in the national lab system as well and we're modeling those reactors to support the scale-up of catalyst from the innovations that occur in ChemCatBio all the way to demonstration and process scale.
So a lot of these process scale modeling is done with computational fluid dynamics with multiphase flow and what you'll see here is a model on the left of a reactor called the R-Cubed Catalytic Upgrading Reactor. This is an entrained flow reactor where the catalyst particles enter the bottom and mix with pyrolysis vapors, and then as you go up the column, those catalysts then help to convert those pyrolysis vapors into hydrocarbons that are more appropriate for feeding into our fuel infrastructure in the United States.
So you can see on the right the actual reactor, which is at the National Renewable Energy Laboratory. And so we work very closely with those experimentalists on that team where they're taking validation data for our models, and then ultimately we want to have this computational and experimental data to compare and frame the complete picture of the catalytic upgrading that's occurring.
And, by the way, this type of model is done with MFiX, Multiphase Flow with Interphase eXchange, which is a CFD type code that was developed by NETL as part of our program. So this is an open source code and so whatever tools we develop and models we develop are directly portable to anybody.
So in terms of the reactors that we're modeling in the program, we actually have three different size reactors. This one is a two-inch fluidized bed reactor, and we have a Davison Circulating Riser, and then finally the R-Cubed upgrading system which you saw on a previous slide.
An important thing to note about these reactors is they operate at very different scales in terms of the biomass going into the reactors and so you can see the scales in terms of kilograms per hour demonstrated here. They also have different relevance relative to the purpose. The small reactors, the bench-top reactor, which is a two-inch fluidized bed reactor, is really one of the first catalysts that you'll see the new catalyst innovations tested on, and so it's important to understand those early results so we're modeling that type of a system. And the larger one is actually getting closer to something that is more relevant to industry and relevant to the verification program that goes on in our research program.
The Davison Circulating Riser is also relevant to the industry because industry, in particular WR Grace, uses this system to actually provide performance information for their catalyst. So there are a lot of different reasons to study these different reactors at different sizes.
An important aspect that the process scale models that these reactors give us is the actual resonance time distributions for the catalyst and the reactor. And so here we show on the left we show MFiX simulations of a bubbling bed reactor. This is the two-inch fluidized bed reactor, and this—we've inserted a tracer to setting the flow of new catalyst going into the reactor and then the bubbling bed does catalyst mix and then the catalyst exit out of the reactor. On the right you see the R-Cubed reactor which is that entrained flow system. So these two very different types of reactors have very different resonance times. So the two-inch fluidized bed reactor we have over 800 seconds of average resonance time, whereas in the riser we're looking at only about 139 seconds. So very different resonance times, which will lead to very different performance, and then also the catalyst particle sizes vary as well.
So these are important aspects in terms of the process that we need to capture with these models and are able to do so. And then Peter talked to you about a lot of the work at the mesoscale and one exciting thing that we've been doing here recently in the program is developing kinetics specifics to these complex bioenergy processes, and in particular the catalyst and the process that we've been studying is the catalytic fast pyrolysis and looking at the upgrading of those pyrolysis vapors with the ZSM-5 catalyst, and the approach we're using is to use a lump sum type approach to characterize the kinetics of the different reactions going on.
It's important to note that there's a lot of activity out there studying kinetics, particularly at the universities, for instance at Northwestern University and the University of Delaware, Professors Broadbelt and Klein, respectively, at those institutions are studying micro-kinetics. We're very much in support of that effort and hopefully would like to include those connections or models someday, but for the time being we're using a simpler scheme that doesn't capture fully the products that are coming out that we can measure and then we can actually get these kinetics rates.
So this has been done in close collaboration with experimentalists in the ChemCatBio program and you can see some of the kinetics rates here this slide. Peter mentioned to you a little bit about this process and this is a little bit more detail on this slide in the collaboration with the experimentalists to get these kinetic informations. And so we've been working closely with experimentalists who have been running the catalyst in a spouted bed reactor and then detecting the products with a molecular beam mass spec detector, and then with feedback of our models that Peter developed, we've been able to come up with these kinetic schemes.
So at this point we're actually using those kinetics in our models, and so this shows a model of the R-Cubed reactor that's showing the reactions to pyrolysis vapors along with the products groups that we've determined for this kinetics scheme, and you can see that we get a different distribution of those products as you go up the reactors.
But we're at the point now where we can actually run these models and predict the performance for these different products and we're now working on validating the models together with our experimentalists. Some of the validation of the models is shown here. This is actually just the hydrodynamics and in the next coming months we expect to get our initial results on the actual upgrading itself, but for now we've been looking at pressure differences in the reactor and comparing that to what's predicted in our models based on the hydrodynamics, and have come up with pretty good agreement there. So we're really excited about the upcoming validation experiments that will be done with the actual pyrolysis wheel.
So just to summarize, I'd like to point out that the modeling that we do in this program is very much in support of ChemCatBio. We see ourselves as an enabler to enable success in the innovation of new catalysts and then also the understanding of that phenomenon that can lead to getting those catalyst working in true processes that would be of interest to the bioenergy industry.
As we mentioned, and as you saw, we've got activities that actually cover a very wide range of scale. Seonah talked about the atomic scale modeling, Peter talked about the mesoscale, and then I followed up with our activities at the process scale. Again, we thank the experimentalists in the program, and also our industry advisors that have been helping us to get this work done, and we're excited about the kinetics that are coming out of this and our application of those, and we look forward to more experimental validation of that as we go forward in the program.
And finally I would just like to leave the last slide up with the many names of the researchers at the five national labs that have been contributing to the smaller program. I would also like to thank the Department of Energy Bioenergy Technologies Office, in particular Jeremy Leong, Trevor Smith, and Kevin Craig. And then I'd also like to give a shout out to our industry advisory panel. We utilized their input to help steer us into catalyst and processes of the studies that are very relevant to the bioenergy community, so we appreciate their help in doing that.
And I guess now we'll just take any questions.
>> Moderator: Great, thank you, Jim, and thank you, Seonah and Peter. It looks like we do have a bit of time for questions today. As a reminder, you can still submit questions through the question pane in your attendee control panel. Our first question here says "Peter, are you using dual porosity porous flow model to account for inter-particle porosity and inter-particle packing porosity at the same time?
>>Peter Ciesielski: Yeah, so that's a great question. I guess effectively that's what we're doing. So essentially we build up the transport model like an onion, so to speak. So first we compute effective porosities in each of the nanoscale domains of the catalyst and then we apply that to find that element model where we can account for effective transport throughout a catalyst particle. So that's intra-particle diffusion.
And so then we essentially couple that to a—it's basically a theory modulus sort of Darcy's law type approximation that accounts for it. So now we have an effective diffusivity within a catalyst particle; now we have to account for transport between catalyst particles in a packed bed, and that's done with another type of approximation. So, yeah, that's a good question, whoever asked that. We are accounting for those different types of transport phenomenon that arise at different length and time scales.
>> Moderator: Great, thank you! And that person did say thanks. So we do have another question here. It says "A more general question. Does CCPC consider the particle level modeling for coupled transport generation expulsion of biofluids within biomass particles? This could be critical for estimations of effective conversion kinetics that goes into the macroscopic reactive transport model."
>>Peter Ciesielski: Yeah, so again, that question I think is towards actually conversion of the biomass particles and the initial deconstruction step, that's how I'm interpreting it at least, and so I think that's kind of—can you repeat that one more time?
>> Moderator: Yep. It's right here.
>>Peter Ciesielski: Yeah, so we have very, very detailed biomass particle, thermochemical conversion models and we didn't cover those today. We've been working on those for actually longer than we have the catalysis models and we just published a paper in Energy and Fuels—actually we've got the cover—and I'd be happy to send it to whoever asked that question, where we do detailed transport phenomenon coupled through thermochemical reactions within biomass particles and we account for the microstructure of those particles. I think this question was kind of focused on the ejection of vapors and products as those biomass particles convert.
We handle that pretty well. We make assumptions where we can, but I should say we were able to predict the conversion yields coming out of a reactor at NREL within .1 percent of the experimental measurement. So that's probably better than the experimental error, but we can make very, very accurate predictions.
>> Moderator: Great! So to follow up on that, it says, "Peter, it's Hi from INL, please send me the paper you just mentioned."
>>Peter Ciesielski: Sure.
>> Moderator: Great! So it looks like we don't have any other questions at this time. I'll just give everyone another minute to go ahead and submit any questions through the question pane of your control panel. We can also follow up on any other questions you might have after the webinar today.
>> Jim Parks: While you're doing that I'll just add to what Peter mentioned about the thermochemical models of biomass particles—you can read more about that in the paper and we have a list of our publications on our website. So feel free to go check out the website and see that. You'll probably also see some examples of the models of biomass particles that Peter was talking about.
>> Moderator: Great, thanks! It looks like we don't have any other questions at this time, so thank you to all of our presenters and thank you everyone for attending today's webinar Simulation and Modeling of Bioenergy Conversion from the Atomic to Process Scale. You will receive a follow up email within the next couple of weeks with a link to view a recording of today's webinar. Thank you for joining us today and have a great rest of your day.
[End of Audio]