Highly Selective and Stable Multivariable Gas Sensors...Implement bio-inspired multivariable gas sensors for multi -gas quantitation at high temperatures. Perform laboratory optimization

R.A. Potyrailo 2018 1Public Copyright © 2018 General Electric Company - All rights reserved

Highly Selective and Stable Multivariable Gas Sensors for Enhanced Robustness and Reliability of SOFC Operation

GE: Radislav Potyrailo, Vijay Srivastava, Joleyn Brewer, Claire Henderson, Brian Scherer, Majid Nayeri, Andrew Shapiro

SUNY Poly: Michael Carpenter, Nicholas Karker, Nora Houlihan, Vitor Vulcano Rossi, Laila Banu

NETL Contract FE0027918

19th Annual SOFC Project Review Meeting, Washington, D.C., June 13-15, 2018



Project strategy:Develop in-line gas sensor for real-time SOFC diagnostics and enhancement of operation reliability

Project approach: Implement bio-inspired multivariable gas sensors for multi-gas quantitation at high temperaturesPerform laboratory optimization followed by field validation

Phase 1 outcomes:Fundamental understanding of performance and development of design rules of multivariable gas sensors at high temperatures

Project overview

This year top FIVE accomplishments

Initial

Improved

Time

Ref

lect

ance

PC1

H2CO

H2 + CO

Blank

PC2

-2

-1

0

1

2

3

4

5

6

7

8

9

10

11

0 1 2 3

Pred

icte

d H

2co

nc. (

AU)

Actual H2 conc. (AU)

H2 conc. = 5, 10, 15 %

Conventional technique

New technique

Benchmark

Sensor

Initial field validation

Sensitivity control

Selectivity control

Detection of gas mixtures

Sensor stability

1 µm


Mul

tivar

iate

resp

onse

#1

Gas 1

Gas 2

Gas 3

$ 250K instrument

Unmet need for real-time monitoring of H2 and CO gases

Status quo:Mature traditional detector

conceptsPerformance need:Multi-gas discrimination

Our approach:Multivariable photonic gas

sensors

Real-time knowledge of H2/CO ratio (3:1–2:1) of anode tail gases:• will allow control of efficiency of reforming process in the SOFC system• will deliver a lower operating cost for SOFC customers

In-line gas detection is not straightforward


Biomimicry –imitation of biological systems

Bioinspiration –new functionality, beyond Nature

High temperatureBiomimetics –recreation of observed functionality

Room temperature

Learning from Nature

1 µm

500 nm

2 mm

1 µm

Potyrailo, Carpenter, et al., J. Opt., 2018

Potyrailo et al., Nat. Commun. 2015

Potyrailo et al., Nat. Photon. 2007


Adsorbed gas molecules

Mul

tivar

iate

resp

onse

#1

Gas 1

Gas 2

Gas 3

Performance need:Multi-gas

discrimination

Our approach:Multivariable gas sensors

blankbottommiddletop

Free gas molecules

Potyrailo et al. Nature Photonics 2007Potyrailo et al., Proc. Natl. Acad. Sci. U.S.A. 2013Potyrailo et al. Nature Communications 2015

Design rules for gas-selectivity control• Spatial orientation of surface functionalization• Chemistry of surface functionalization• Extinction and scattering of nanostructure

Multi-gas detection using

bio-inspired multivariable photonic sensors

∆R

efle

ctan

ce

300 650Wavelength (nm)

Differential reflectance ΔR(λ) =100% R(λ)/R0(λ)

R(λ) = sensor with analyteR0(λ) = sensor with blank


Advancing design rules of nanostructures for high temperature gas-sensing applications

Selectivity control for vapors at room temp.

Selectivity control for gases at high temp.

Interference rejection control

•Polymeric nanostructure

•Absorption and adsorption of vapors

•Multi-material inorganic •nanostructure

•Catalytic reactions of gases

•Inorganic nanostructure

•Catalytic reactions of gases

500 nm500 nm1 µm

Potyrailo, Karker, Carpenter, Minnick, J. Opt., 2018

Potyrailo, Chem. Rev. 2016Potyrailo et al., Nat. Commun. 2015


400 450 500 550Wavelength (nm)

Spec

tral r

espo

nse

Needed diversity in spectral response for gas discrimination

600 700 800 900Wavelength (nm)

Spec

tral r

espo

nse

Diverse spectral response facilitates discrimination between different gases


Gas-discrimination control

100

Lamella spacing (nm)

150 200

75 100 125 150 Simulation data with SiO2 lamella

Diverse spectral response using different lamella spacing for discrimination between different gases

400 1400

Ref

lect

ance

Wavelength (nm)

600 1400

Ref

lect

ance

Wavelength (nm)

400 1400

Ref

lect

ance

Wavelength (nm)

600 1400

Ref

lect

ance

Wavelength (nm)


Sensitivity boosted by seven fold by• material selection • 3D structure design

Sensitivity control

Initial 3D

Improved 3D

Exposure to 8% H2


Example of H2 detection

0358

H2 conc. (%)

H2 spectral response H2 response: dynamics

Diverse spectra and dynamics uncover different response mechanisms in nanostructurePotyrailo, Karker, Carpenter, Minnick, J. Opt., 2018

500 600 700 800 900

Del

ta R

Wavelength (nm)

Del

ta R

Del

ta R

TimeTime

50 min50 min

860 nm660 nm


00.50.751

CO conc. (%)

CO spectral response CO response: dynamics

Diverse spectra and dynamics uncover different response mechanisms in nanostructure

Example of CO detection

Potyrailo, Karker, Carpenter, Minnick, J. Opt., 2018

500 600 700 800 900

Del

ta R

Wavelength (nm)

Del

ta R

Time

Del

ta R

Time

50 min 50 min

790 nm670 nm


670

675

680

685

690

695

0 5 10 15 20

Pea

k P

ositi

on (n

m)

H2 concentration (%)

Initial stability tests: univariate response

Peak position of one of spectral bands in a planar (control) sensor film

ExperimentH2, % = 5, 8, 12, 15, 20Number of cycles = 10 Test duration = 17 days replicates n = 10

Calibration curve over 17 days of testing

0

5

0

5

0

5

0 2 4 6 8 10 12 14 16

Time (days)

Peak

posit

ion

Peak

posit

ion

Tests with a planar sensor film and analysis of univariate (single output) response allows determination of sensor stability using classical methods


ROC curves with initial sensor stability at day 1

Receiver Operating Characteristic (ROC) curves

0 0.2 0.4 0.6 0.8 10

0.2

0.4

0.6

0.8

1

Prob

abilit

y of

det

ectio

n

Probability of false alarm0 0.2 0.4 0.6 0.8 1

Probability of false alarm

0

0.2

0.4

0.6

0.8

1

Prob

abilit

y of

det

ectio

n

ROC curves illustrate the diagnostic ability of the developed sensor to reliably detect gas concentrations

1.0000.5000.2500.1000.0750.050

H2 conc. (%)1.0000.5000.2500.1000.0750.050

H2 conc. (%)

ROC curves with sensor stability at days 1 - 17


Car driving modes

Software-driven boost of system performance

Smart phones: zoom, brightness control

Smart phones: face recognition

Consumer products

Electronics analytics

• Principal component analysis (PCA)• Discriminant Analysis (DA)• Artificial Neural Network (ANN)• Hierarchical cluster analysis (HCA) • Support Vector Machines (SVM)• Independent Component Analysis (ICA)• Partial least squares (PLS) regression• Principal Component Regression (PCR)

Potyrailo Chem. Rev. 2016


Multivariate data analysis e.g. PCA = Principal Components Analysis

Raw spectra PCA-processed

PCA

PCA – “classic” tool for reduction of data dimensionality and noise

PC = principal component

Poor sensitivityGood selectivity

Good sensitivityPoor selectivity

Good sensitivityGood selectivity

2-D response dispersion


0.985

0.99

0.995

1

1.005

1.01

1.015

1.02

400 480 560 640 720 800 880 960

Del

ta R

Wavelength (nm)

Initial discrimination between individual H2 and CO gases

Conditions:planar sensing filmTwo concentrations of H2Two concentrations of COSpectral replicates (n=2)

BlankH2 conc1H2 conc2CO conc1CO conc2

Different spectral regions of optical response of a single sensing to H2 and CO gases should allow discrimination of two gases

Spectral response to H2

Spectral response to CO


H2

CO

H2 + CO

Initial discrimination between individual and mixtures of H2 and CO gases

0.985

0.99

0.995

1

1.005

1.01

1.015

1.02

400 480 560 640 720 800 880 960

Del

ta R

Wavelength (nm)

Blank H2 conc1H2 conc2

CO conc1CO conc2

H2 + CO conc 1+1H2 + CO conc 1+2

Different spectral regions of optical response of a single sensing to H2 and CO gases should allow discrimination of two gases


-30 -20 -10 0 10 20 30 40PC1

H2CO

H2 + CO

Blank

-10

-8

-6

-4

-2

0

2

4

6

8

10PC

2H2

CO

H2 + CO

Initial discrimination between individual and mixtures of H2 and CO gases

Different spectral regions of optical response of a single sensing to H2 and CO gases allow discrimination of two gases


PCC / PFA = 1.0 / 0 PCC / PFA = 0.9 / 0.002 PCC / PFA = 0.7 / 0.03

No added noise Added 0.001 noise Added 0.002 noise

Wavelength Wavelength Wavelength

0.98

1

1.02D

elta

R

0.98

1

1.02

Del

ta R

0.98

1

1.02

Del

ta R

1

3

5

b

6

4

2

Predicted gas composition1 3 5b 6421 3 5b 6421 3 5b 642

1

3

5

b

6

4

2

1

3

5

b

6

4

2

Predicted gas composition Predicted gas composition

Actu

al g

as c

ompo

sitio

n

Actu

al g

as c

ompo

sitio

n

Actu

al g

as c

ompo

sitio

n

BlankH2 1H2 2CO 1CO 2H2 + CO 1H2 + CO 2

b123456

Probability

0

1

PCC = probability of correct classification; PFA = probability of false alarm

Confusion matrix analysis

Visualization of quality of prediction of classes

Gas compositions


Time

19 days with total of 75 cycles of H2 exposures

H20 1 2 3

H2 test cycle

Long-term response stability testing

H20 1 2 3

H2 test cycleH2 conc. = 5, 10, 15 %

50 min

Time

Spec

tral r

espo

nse

Planar sensor film and analysis of multivariate (many wavelengths) response allows determination of sensor stability using new machine learning methods


39500

42000

0 20

Co

unts

Time (days)

84000

89000

0 20

Co

unts

Time (days)

Raw sensor response to H2: effects of drift

650 nm

750 nm

Diverse temporal profiles of drift - possibility for drift correction ?



New technique

Predicted H2 concentrations

H2 conc. = 5, 10, 15 %

Drift correction became a reality using new machine learning tools


-2

-1

0

1

2

3

4

5

6

7

8

9

10

11

0 1 2 3

Pred

icte

d H

2co

ncen

tratio

n (n

orm

aliz

ed u

nits

)

Actual H2 concentration (normalized units)

Correlation of actual vs predicted H2 concentrations

H2 conc. = 5, 10, 15 %

Developed data analytics technique improved the prediction ability of the sensorin detecting and quantifying a single gas

by more than 10 fold


New technique


Benchmark instrumentsystem

Sensor under development

Initial tests of multivariable sensorat the SOFC factory at GE–Fuel Cells LLC

Benchmark system:Rosemount Analytical (Model X-STREAM Enhanced XEXF)


Spectrograph

Lightsource

Gas flow cell

Data acquisitionGas in

Light in/out

Fiber-optic reflection probe

Details of multivariable sensorat the SOFC factory at GE–Fuel Cells LLC

Gas flow cell

Laboratory components for testing of performance of 3D fabricated nanostructure


Sensing chip in gas cell with white light illumination

0

1.4 105

400 500 600 700 800 900 1000

Cou

nts

Wavelength (nm)

In-situ spectral collection

N2H2 = 4%CO = 22 %

Spectral features of 3D fabricated nanostructure are preserved in the gas cell design for field use


-1000

0

Del

ta C

ount

s

-3000

0

Del

ta C

ount

s

-2500

0

Del

ta C

ount

s

-5000

0

0 10 20 30 40 50

Del

ta C

ount

s

Time (min)

H2 H2 H2 CO COCO

460 nm

530 nm

600 nm

960 nm

Initial tests of developed 3D sensing nanostructure at the SOFC factory at GE–Fuel Cells LLC

H2 = 4%CO = 22 %

0.7

1

500 600 700 800 900

Del

ta R

efle

ctan

ce

Wavelength (nm)

Diversity of spectral features of 3D fabricated nanostructureFor independent quantitation of several gases with a single sensor


Summary of photonic multivariable gas sensors:developed capabilities

Next steps:• Summarize and document developed design rules for photonic multivariable

sensing at high temperature• Complete validation of developed multivariable sensor at the SOFC factory at

GE–Fuel Cells LLC

This year top FIVE accomplishments

Initial

Improved

Time

Ref

lect

ance

PC1

H2CO

H2 + CO

Blank

PC2

-2

-1

0

1

2

3

4

5

6

7

8

9

10

11

0 1 2 3

Pred

icte

d H

2co

nc. (

AU)

Actual H2 conc. (AU)

H2 conc. = 5, 10, 15 %


New technique

Benchmark

Sensor

Initial field validation

Sensitivity control

Selectivity control

Detection of gas mixtures

Sensor stability

1 µm


Acknowledgements

• GE Global Research Team• SUNY Polytechnic Institute team • US Department of Energy Cooperative Agreement FE0027918

This material is based upon work supported by the Department of Energy under Award Number DE-EE00027918. This presentation was prepared as an account of work sponsored by an agency of the United States Government. Neither the United States Government nor any agency thereof, nor any of their employees, makes any warranty, express or implied, or assumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or represents that its use would not infringe privately owned rights. Reference herein to any specific commercial product, process, or service by trade name, trademark, manufacturer, or otherwise does not necessarily constituted or imply its endorsement , recommendation, or favoring by the United States Government or any agency thereof. The views and opinions of authors expressed herein do not necessarily state or reflect those of the United States Government or any agency thereof.

Documents

Highly Selective and Stable Multivariable Gas Sensors...Implement bio-inspired multivariable gas sensors for multi -gas quantitation at high temperatures. Perform laboratory optimization