DOE Bioenergy Technologies Office (BETO)

2019 Project Peer Review

Rapid Construction of Validated Chemistry

Models for Advanced Biofuels (EE0007982)

March 7, 2019

Co-Optimization of Fuels and Engines

Principal Investigator: William Green

Massachusetts Institute of Technology (MIT)

Co-PI: Subith Vasu, University of Central Florida (UCF)

This presentation does not contain any proprietary, confidential, or otherwise restricted information

Goal Statement

• Goal: Demonstrate capability to rapidly

generate accurate combustion chemistry

computer models for advanced biofuels

• Outcomes:

– Rapid simulation of proposed biofuels in future

engines, including fuel effects on performance.

– Faster, more reliable selection of proposed fuels.

– Computer models for each promising biofuel

available to engine & fuel designers, accelerating

tuning of engines to new fuels and vice-versa.

2

Quad Chart Overview

Timeline• Official Start 1/15/17 (funded 8/1/17)

• Project End Date 3/30/20

• 30% Percent complete

3

FY 17 Costs

FY 18 Costs

Total Planned

Funding

(FY 19-Project End Date)

DOE Funded

4k 263k 626k

Project

Cost

Share*

3k 40k 56k

•Partners:

MIT (60%)

Univ. Central Florida (40%)

Lab Contact: Bill Pitz (LLNL)

Barriers addressed

ADO‐E. Co‐Development of Fuels and Engines

ObjectiveDevelop computer methods for generating computer models for advanced biofuel combustion. Test accuracy of models with laser/shock tube experiments.

End of Project Goal

Demonstrate capability to rapidly generate accurate combustion chemistry computer models for advanced biofuels

1 - Project Overview

• Many biofuels proposed, but few have been successful.

– Significant performance risk as well as economic risk.

• Various biofuel + future engine combinations have significantly

different performance, costs, and societal benefits

– Which biofuel + engine combination is best?

– Much less expensive to explore on computer than by experiment

– ….but need fuel chemistry computer model for each biofuel

• Building biofuel models by hand is tedious and challenging

– PI expert in computer-construction of chemistry models

• How to Test Accuracy? A few data on ignition delays, little else

– Need measurements at conditions relevant to future engines

– Pre-ignition species time-profiles are most useful

– Co-PI expert in high-sensitivity species measurements

4

2 – Approach (Management)

• Prof. Green (MIT) is PI, Prof. Vasu (UCF) is the co-PI– Green responsible for all the modeling, Vasu for all the experiments

• Project only involves ~6 people total, and we are

modeling and measuring the exact same systems, so it

is easy to keep in good communication and collaborate

• Broader Co-Optima team involves many people at many

institutions, is challenging to stay well-connected with

everyone– We participate in the Co-Optima meetings and teleconferences

• We have received excellent advice from and collaborate

with Bill Pitz (the lead chemistry modeler on Co-Optima

team), Mark Nimlos and others at NREL, and Scott

Goldsborough (ANL).

5

6

2 – Approach (Technical)

• Computer-constructed models: Need computer to correctly

identify the important species and reactions as model is being

constructed

– If computer misses an important reaction the model will not be accurate

– If computer includes many unimportant reactions, process is not rapid

– Often mis-identifications are due to poor thermo or rate estimates

• Experimental Test of Model Accuracy: Reliably measure species

time profiles

– Extremely low concentrations pre-ignition, need high sensitivity!

– Measure absolute T-dependent line-strengths and pressure broadening

• Some biofuels and intermediates have intramolecular H-bonds:

Develop methods to compute their thermo & rates

– Developing high-accuracy method and also fast estimation method

• Some fuel mixtures too complicated for computer modeling

alone: Correctly combine modeling with a few lab experiments

3 – Technical Progress

Methods for Constructing Biofuel

Models

7

• Milestone 4.5: Implement and document an improved model construction workflow.

• Software developed for automating quantum calculations for thermodynamic properties to reduce time spent on calculations

• Machine learning based higher accuracy rate estimator used to reduce time wasted on unimportant reactions

3 – Technical ProgressIntegrated Accurate Thermo from Automated Quantum Calcs into Model-Building Workflow

Chemical Knowledge– Thermochemistry– Reaction kinetics

Fuel Chemistry

Model

Mechanism Generation Algorithm

Exptl Data

Literature,ManualCalcs,

Experiment

Important Species and Reactions

Improved Parameters

Sensitivity Analysis

Thermo sets Keq’s.Automation greatly speeds development of accurate chemistry models

Validation

8

3 – Technical Progress

Improved k(T) Estimation reduces time wasted on unimportant reactions

C-H + R2 O-H + R2

R1-H + R2 (-> R2-H + R1)

O-H + OO-H + CO-CH2-H + R2 C-CH2-H + R2

Example decision tree:

Importance of reaction depends on k(T).

Initial k(T) estimates from classifying

new reaction as similar to known

reactions. New Machine Learning Decision

Tree method improves accuracy of

classifications, cuts k(T) error bar 18x

9

3 – Technical Progress

Measurements of Species

Time-Profiles

10

• Milestone 2.1: Measure CO time-histories during biofuel combustion for pressures up to 10 atm.• Measurements have been taken during cyclopentanone

oxidation and pyrolysis, methyl propyl ether oxidation and pyrolysis, and 2,4,4-trimethyl-1-pentene oxidation

• Milestone 2.2: Measure CO time-histories during biofuel combustion for pressures up to 30 atm.• Diagnostics have been configured and tested. Measurements

ongoing for ethanol oxidation

• Milestone 3.1: Measure HCHO time-histories during biofuel combustion for pressures up to 10 atm.• A sensor has been developed and measurements have been

taken during the pyrolysis of methyl propyl ether

3 – Technical ProgressMeasured absolute CO cross-sections

at high T and PMeasured CO absorption cross-section behind reflected shockwaves at 10 and

20 atm over a temperature range of 900-1500 K. This data is used to extract

concentrations of CO during combustion of the Co-Optima biofuels

11

-0.20 -0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 0.20

0

1

2

3

4

5

6

Mea

sure

d A

bso

rban

ce

Time (ms)

Measured Absorbance

Measured Pressure

T=1434K

P=8.77atm

Absorbance

Pressure

CO=1%

He=12%

Ar=87%

0

2

4

6

8

10

Pre

ssure

(atm

)

1000

1100

1200

13001400

1500

6

7

8

9

1 0

1 1

1 2

2.00E-019

2.20E-019

2.40E-019

2.60E-019

2.80E-019

Pressu

re (at

m)

Ab

so

rptio

n C

ross S

ectio

n (c

m2/m

ole

cu

le)

Temperature (K)

CO Cross-Section

n=2046.30 cm-1

3 – Technical ProgressMeasured CO time profiles up to 20 atm

12

As one example, time-histories of CO have been measured up to 20 atm in the

Co-Optima fuel candidate, ethanol.

3 – Technical Progress

Tests of Models with Experiment

13

• Milestone 4.1.1: Develop models for first 2 biofuels• Combustion models developed for methyl-

propyl ether and cyclopentanone. Cyclopentanone model was developed in collaboration with Bill Pitz and other Co-Optima team members.

• Milestone 5.1: Validate year one models against experiment• Both models validated against our shock tube

data and additional data from experimental groups, some in Co-Optima team.

Time-histories of CO have been measured up to 10 atm in three Co-Optima

fuel candidates: cyclopentanone, methyl propyl ether, and 2,4,4-trimethyl-1-

pentene. Here we show our measurements for cyclopentanone vs. models

3 – Technical Results

Cyclopentanone Data vs. Models

14

Co-Optima Team Model: Zhang et al. (2018)

includes rates, thermo computed by

Green’s group at MIT

LLNL = Co-Optima team model

Zhang et al. (2018)

Thion = model of Thion et al. (2017)

0.0 0.5 1.0 1.5 2.0 2.5 3.0

0

1000

2000

3000

4000

5000

6000

CO

Co

nce

ntr

atio

n (

pp

m)

Time (ms)

1.61% dP5/dt

0

2

4

6

8

10

12

14

16

T5= 1250 K

P5= 8.51 atm

C5H8O=0.001

O2=0.007

Ar=0.992

Measured Pressure

Model Pressure

Measured CO

Model CO

P5

(atm

)

0.74 0.76 0.78 0.80 0.82 0.84 0.86

1

10

LLNL

Ignitio

n D

ela

y T

ime (

ms)

1000/T (K-1)

0.1%C5H8O\O2\Ar

f=1

1 atm

6 atm

8.5 atm

Thion

3 – Technical Results

Methyl Propyl Ether Model vs. Data

15

CH4/C2H4

C3H6/C2H4

(Conditions: ~8% O2, ~0.5% MPE, ~91.5% He, 1 bar, 2 s)

Experimental flow reactor data from

G. Fiorina & R. McCormick (NREL).

Computer generated model by MIT.

Some important k(T) computed by

M. Nimlos, B. Lintao, S. Kim (NREL)

Model predicts Fuel conversion

and major product ratios

accurately for T>750 K.

3 – Technical ResultsMethyl Propyl Ether Combustion: UCF CO

Time Profiles vs. Computer-Generated Model

(0.1% MPE, 0.6% O2, 99.3% Ar)

16

Peak CO yields predicted

within 20%, MPE decay time

and CO rise & fall times all within

factor of 2. This is close to exptl

uncertainties in all quantities.

Without any tweaking to force a fit,

computer-generated model

predicts biofuel chemistry with

fairly high fidelity –

Our method appears to work!

3 – Technical ProgressHCHO measurement scheme developed,

first time-profiles measured

17

• Sensor has been designed and setup to measure HCHO time-histories up

to 10 atm during combustion of Co-Optima fuel candidates.

– Absorption cross-sections of HCHO have been measured behind

reflected shockwaves to accurately extract HCHO time-histories

– First demonstration: Methyl Propyl Ether pyrolysis

1100 1200 1300 1400 1500 1600

5.00E-020

1.00E-019

1.50E-019

2.00E-019

2.50E-019

Measured Cross-Section

Best Fit

Absorp

tion C

ross-S

ection (

cm

2/m

ole

cule

)

Temperature (K)

n=1745.735cm-1

0 1 2 3 4

0

2500

5000

7500

10000

12500

15000

17500

0

1

2

3

4

Measured Pressure

Model Pressure

Measued HCHO

Model HCHO

Pre

ssure

(atm

)

Specie

s C

oncentr

ation (

ppm

)

Time (ms)

T5=1287 K

P5=2.10atm

C4H10O=1.25%

Ar=98.75%

Pressure

HCHO

Model: MIT MPE model

18

4 – Relevance

Rapid Construction of Validated Accurate Biofuel

Combustion Models

• Part of BETO’s mission: “develop… technologies to enable… biofuels”

• Directly addresses BETO’s Advanced Development and Optimization

Challenge E: Co-Development of Fuels and Engines

• Too many possible future fuel + future engine combinations to test

them all experimentally.

• Pre-screen on computer using Biofuel Combustion Models.

• Focus experiments on biofuels most likely to succeed.

• Computer-aided design is used to develop future engines; to pick up

fuel effects it needs fuel-specific combustion chemistry sub-models

• Automakers are big consumers of biofuel combustion models – they

want their engines to work with whatever fuels will be commercialized

• Current methods for building fuel combustion models are slow and

unreliable – our new approach can change that.

19

5 – Future Work

More fuels and new methods

• Milestones 4.2.1, 4.3.1: Model six more biofuels.

– Check robustness, efficiency of modeling methods.

• Task 5.0: Test new fuel models vs. experiment

• Milestone 4.4.1: Method for modeling

intramolecular H-bonding (for oxygenated fuels)

• Task 6.0: Time-histories of other combustion

species (e.g. CH4, C2H4, CO2, H2O, H2O2)

– More comprehensive test of fuel models

– Requires development of probe methods for each.

• Task 7.0: Develop method for modeling fuels

whose composition is known imperfectly

– Some biofuels are complex mixtures, hard to analyze

20

Summary

Need fast reliable methods for assessing biofuels

Ideally, evaluate biofuels on computer.

Requires computer models for each fuel’s chemistry…

…also useful for co-optimization of fuel with engines

Need models for many fuels: computer builds the models!

Can this really work? How accurate? Test with experiments!

Developed new experiments measuring CO, HCHO vs. time

Created models for 2 fuels and tested with experiments.

++++Looks promising so far!++++

Next: model more (and more complicated) fuels & measure

more species, to see how accurate & robust new method is.

21

Additional Slides

22

Publications, Patents, Presentations,

Awards, and Commercialization• Zhang, K., et al., An experimental, theoretical, and modeling study of the ignition behavior of

cyclopentanone. Proc. Combust. Inst. https://doi.org/10.1016/j.proci.2018.06.097

• Ninnemann et al. Shock tube and CO laser-absorption measurements during cyclopentanone

oxidation. Eastern States Section of the Combustion Institute, 2018. Penn State University.

• Ninnemann, E., et al. Pyrolysis of cyclopentanone: A shock tube and laser absorption study. Joint

Propulsion Conference, 2018. Cincinnati.

• Johnson, M.S. & Green, W.H. Machine Learning Approach to Rate Estimation. American

Chemical Society National Meeting. 2018 Boston.

• Khanniche, S. & Green, W.H. Chemical mechanism and kinetics of cylclopentanone combustion:

A theoretical and RMG approach. American Chemical Society National Meeting. 2018 Boston.

• Laich et al. A shock tube and laser absorption study of CO time-histories during bio ether

oxidation. AIAA SciTech, 2019. San Diego, CA.

• Ninnemann et al. Shock tube ignition study of prenol- a “hyperboosting” fuel relevant to the Co-

Optima initiative. 11th US National Combustion Meeting. 2019 Los Angeles, CA. (accepted)

• Nimlos, M.R. et al. Low Temperature Oxidation of Methyl Propyl Ether. 11th US National

Combustion Meeting. 2019 Los Angeles, CA. (accepted)

• Green, W.H. Automated Construction of High-Fidelity Fuel Chemistry Models: Status &

Challenges. 17th International Conference on Numerical Combustion. 2019. Aachen, Germany.

(invited talk)