developing descriptors to predict mechanical properties of nanotubes dr. tammie l. borders may 16,...
TRANSCRIPT
Developing Descriptors to Predict Mechanical Properties of Nanotubes
Dr. Tammie L. BordersMay 16, 2013
2
Outline (1)
Problem description Methodologies
Informatics flow Selecting descriptors
CNT surfaces Derive critical descriptors for mechanical properties Modular extension of QSPR to two types of defects
Comparison to experimental results Investigating load transfer mechanisms in multi-walled CNTs Prediction of monolayer graphene mechanical properties
Future Work Conclusion
3
Problem description- Carbon nanotube (CNT) reinforced composites Goals
Improve load transfer at interfaces CNT-polymer CNT intra-wall
Maintain intrinsic properties of CNTs Consider realistic variations (i.e., defects, functionalization, etc)
Problem large parameter space Solution informatics methodologies
Informatics-molecular dynamics approach to explore large problem space; Refinement with experimental data provides quantitative accuracy
4
Problem description- Carbon nanotube (CNT) reinforced composites (1)
Goals Improve load transfer at interfaces
CNT-polymer CNT intra-wall
Maintain intrinsic properties of CNTs Consider realistic variations (i.e., defects, functionalization, etc)
Problem large parameter space Solution informatics methodologies
Informatics-molecular dynamics approach to explore large problem space; Refinement with experimental data provides quantitative accuracy
1) Surface morphology – stiffness / strength2) Intra-wall bonding – load transfer
5
Outline (2)
Problem description Methodologies
Informatics flow Selecting descriptors
CNT surfaces Derive critical descriptors for mechanical properties Modular extension of QSPR to two types of defects
Comparison to experimental results Investigating load transfer mechanisms in multi-walled CNTs Prediction of monolayer graphene mechanical properties
Future work Conclusion
6
Informatics flow
MEASUREMENTS• Process parameters• Characterization• Databases
THEORY• Atomic calculations• Mesoscale calculations• Continuum calculations
CORRELATIONS• Clustering• Feature selection• Data mining & analysis
KNOWLEDGE• Process-structure- property relationships• Materials discovery• Hidden data trends
7
Selecting descriptors- Borrowing for pharmaceutics
Most descriptors to date – for small drug like molecules
First approach – simple constitutional, topological, physicochemical
Properties of interest Pharma – activity, toxicity (ADME-T) Materials – mechanical, electrical, thermal,
lifecycle behavior / mechanisms
Scales of interest Pharma – molecular Materials – molecular macro
8
Selecting descriptors- The process
Data Collection & Pre-processing Determine Descriptors
Initial guess Clustering
Principal Component Analysis Loadings Plot Feature Selection
Principal Component Analysis Star Plots Partial Least Squares
Multivariate Analysis Partial Least Squares
Validation Y-scrambling
Can we improve model interpretability? Accuracy? Descriptor set next guess
Selecting Descriptors is a Human-in-the-Loop, Iterative Process with the Goal of Simultaneously Improving Model Interpretability and Accuracy
Initial Guess (20)
9
Selecting descriptors- Clustering & feature selection
Star PlotsIdentifies descriptor stable / robustness;
Provides measure of strength / correlation
PCA Loadings PlotIdentifies clusters, important descriptors
PC1 & PC2 account for 69% of variance
10
Selecting descriptors- Multivariate analysis and validation
PLS Regression ModelProvides interpretability, accuracy
Y-ScramblingMeasure of model robustness
11
Outline (3)
Problem description Methodologies
Informatics flow Selecting descriptors
CNT surfaces Derive critical descriptors for mechanical properties Modular extension of QSPR to two types of defects
Comparison to experimental results Investigating load transfer mechanisms in multi-walled CNTs Prediction of monolayer graphene mechanical properties
Future work Conclusion
12
CNT surfaces – descriptors Descriptors Definition
Theoretical Radius (Å) Theoretical radius of perfect nanotube
% Methyl Groups Percentage of surface functionalized methyl groups
# Missing C’s (CM) Number of missing due to a vacancy defect
# Methyl Groups (MN) Number of methyl functional groups
MN / CT Ratio of methyl groups to total number of carbons
CM / CT Ratio of missing carbons to total number of carbons
# Single Defects Number of single defect types
# Non-sp2 C’s (CN2) Number of non-sp2 hybridized carbons
CN2 / CT
Ratio of non-sp2 hybridized carbons to total number of carbons
Surface Area (SP) Total surface area of nanotube (uses average radius)
Defect Surface Area (SD) Surface area of defects
SD / SP Ratio of defect area to total surface area
Chiral Angle Chiral angle
Started with ~20 Descriptors, Massaged & Down-selected to 2 & 3
13
Young’s modulus – vacancy defect
Critical descriptors CN2 / CT – captures size and type of
defect Chiral angle
Accurate & interpretable QSPRs 20-descriptor & 2-descriptor model
have similar accuracy 2-descriptor interpretability improved
Test Set R2Train R2
Test Descriptors
3A 2-21 Å
0.95 0.76
Surface Area (SP)
Chiral Angle
CN2 / CT
% Double Defects
3B 2-21 Å
0.94 0.85
Theoretical radius
Chiral Angle
CN2 / CT
Test set R2Train Descriptors, Individual Coefficients
Chiral Angle Individual R2
Measure of Defects
Individual R2
3C 2-21 Å 0.92 Chiral Angle
0.25 CN2 / CT 0.66
Two Critical Descriptors - CN2 / CT & chiral angle
Young’s modulus – vacancy & methyl defects
14
‘Modularly extend’ to more than one type of defect (f(x,y) = ax + by +…)
Identify critical descriptors CN2 / CT – captures changes to CNT
surface structure Chiral angle MN / CT – captures intrinsic
properties of functional group Loadings plot clear indicate methyl
group is new cluster PLS R2 values ~ same for vacancy &
vacancy / methyl studies
Easy to add new groups - no need to start from scratchEasy to use higher fidelity calculations - for specific groups only
Model R2Train R2
Test Descriptors
20-descriptor QSPR 0.95 0.89
Surface Area (SP)
Chiral Angle
# Single Defects % Double Defects
CN2 / CT
% Methyl
3-descriptor QSPR 0.92 0.93
Chiral Angle
CN2 / CT
MN / CT
15
Outline (4)
Problem description Methodologies
Informatics Flow Selecting Descriptors
CNT surfaces Derive critical descriptors for mechanical properties Modular extension of QSPR to two types of defects
Comparison to experimental results Investigating load transfer mechanisms in multi-walled CNTs Prediction of monolayer graphene mechanical properties
Future Work Conclusion
Investigating load transfer mechanisms in multi-walled carbon nanotubes
Observation Load transfer increase of 2.4 (maximum is 3) Inner CNT ~95% area of outer CNT (assume area is equal)
Plausible explanations Assume all walls participate equally & load transfer is simultaneous
YM4 = 90% of YM0; 0.0002622 CN2 / CT predicts ~0% (off by 100x) Assume walls do not participate equally (1.0, 0.8, 0.6) & load transfer is not simultaneous
(Assume) no decrement to Young’s modulus (Observation / assumption) Inner walls do not participate until ~2% strain 1050 GPA (0-2% strain for 1 wall) + 832 GPA (2-4% strain for 3 walls) = 941 GPA
16
Sample Gauge Length (nm)
Inner Diameter
(nm)
Outer Diameter
(nm)
Broken shells
Total Shells
CN2 / CT (All single
vacancies)
CN2 / CT
(All methyl groups)
Young’s Modulus
(GPa)
Load transfer increase
1 1,852 - 14.72 1 12 - - 990
2 2,024 - 15.71 1 17 - - 1,049
3 2,105 - 25.97 1 2 or 3 - - 1,105
4 1,035 37.44 39.48 3 18 0.0002622 0.0000874 932 2.4 / 3
ollAF
Y
Monolayer graphene mechanical properties
Our equation: (Young’s Modulus) = -2,225 CN2 / CT + 1975 MN / CT
Elastic moduli (E2D) = volumetric Young’s modulus * thickness Experimental
Monolayer graphene: 342 N/m +- 30 N/m Monolayer graphene oxide: 145.3 +- 16.3 N/m
Computational Monolayer graphene: 382 N/m Monolayer graphene oxide: 212 N/m
Using our equation to predict graphene oxide 40% sp3 bonding with oxygen / carbon ratio of 1:5
CN2 / CT ~ 0.4 and MN / CT ~ 1/6 Use methyl descriptor as approximation for epoxide & hydroxyl groups
Predicts E2D of 150 N/m
17
18
Outline (5)
Problem description Methodologies
Informatics Flow Selecting Descriptors
CNT surfaces Derive critical descriptors for mechanical properties Modular extension of QSPR to two types of defects
Comparison to experimental results Investigating load transfer mechanisms in multi-walled CNTs Prediction of monolayer graphene mechanical properties
Future work Conclusion
Future work
New descriptors Experimental Length scales Higher fidelity, other types Other functional groups
Complex systems Quantify load transfer in multi-walled CNTs Optimize two interfacial transfers (CNT-polymer, inter-wall) Optimize interfacial stress transfer & dispersions
19
Conclusion
Critical descriptors for mechanical properties CN2 / CT – captures changes to CNT surface structure
MN / CT – captures intrinsic properties of functional group Chiral angle – effect at smaller radii; (not discussed) should be negligible for
larger radii New types of defects can be successfully modeled as new descriptors
(CURRENT) Evaluation of complex systems require large, complex simulations (NEW) Information in descriptors; easy to add experimental / higher fidelity ONLY
to descriptors that need it (NEW) Complex systems can ‘modularized’
Piece-wise simple systems or f(x,y) = ax + by +… Computational models
Good for qualitative explanations Potential direct link between computational-experimental (Raman spectroscopy)
Using descriptors – creates straightforward method to update with experimental data (quantitative accuracy)
20
Thank you!
Questions?
21