uncertainty of thermodynamic data: humans and …

UNCERTAINTY OF THERMODYNAMIC DATA: HUMANS AND MACHINES

drhgfdjhngngfmhgmghmghjmghfmf

MARIUS STANSenior Scientist and Group LeaderComplex Physical SystemsSystems Security CenterANL

ATPESC 2017St, Charles, ILAug. 5, 2017

THE PROBLEM

Temperature-composition phase diagrams

3

Humans can draw simple diagrams

4

Ternary diagram

Binary diagram

HUMANS ONLY

The Gibbs rule

J. W. Gibbs, On the Equilibrium of Heterogeneous Substances, in Transactions of the Connecticut Academy of Arts and Sciences (1875-1878).

6

Josiah Willard Gibbs (1839-1903)

),,(),,(

),,(),,(

xPTxPT

xPTxPT

BB

AA

At thermodynamic equilibrium, the chemical potential of

each component A, B, C … must be the same in all

phases α, β, γ …

Calculating equilibrium binary phase diagrams

M. I. Baskes, K. Muralidharan, M. Stan, S. M. Valone, and F. J. Cherne, JOM, 55 (2003) 41-50. 7

0 1

T = T1

,

A

,

B

)(xG

)(xG

A B

x

PT

Ax

xPTGxxPTGxPT

,

),,(),,(),,(

Phase diagram

Free energy of components A, B, C …

in phases α, β, γ …

Chemical potentials

PT

Bx

xPTGxxPTGxPT

,

),,()1(),,(),,(

Thermodynamic equilibrium

(Gibbs’ rule)

),,(),,(

),,(),,(

xPTxPT

xPTxPT

BB

AA

),,(...,,

...,, xPTG CBA

The Hf-O Phase Diagram

8

HUMANS AND MACHINES

Solid (A11)

M.I. Baskes et al., Phys. Rev. B, 66 (2002) 104

Liquid Liquid

Melting via Molecular Dynamics

10

),,(

),,(

xPTfH

xPTfH

S

L

Enthalpy

Predicting the equilibrium phase diagrams from atomistic simulations

1M. I. Baskes, K. Muralidharan, M. Stan, S. M. Valone, and F. J. Cherne, JOM, 55 (2003) 41-50.2M. I. Baskes and M. Stan, Metall. Mater. Trans. 34A (2003) 435-39

11

PT

Ax

xPTGxxPTGxPT

,

),,(),,(),,(

Phase diagram

Free energy of components A, B, C …

in phases α, β, γ …by thermal integration1,2

Chemical potentials by

composition derivation

PT

Bx

xPTGxxPTGxPT

,

),,()1(),,(),,(

Thermodynamic equilibrium

(Gibbs’ rule)

),,(),,(

),,(),,(

xPTxPT

xPTxPT

BB

AA

''

),,'(),,(),,(2

T

T

ref

refref

ref

dTT

xPTH

T

xPTG

T

xPTG

LEARNING MACHINES

The Bayesian Method1

1Philosophical Transactions of London, 53 (1763) 370 (free on Google Books).13

Take a guess Acquire info Improve model

N

i

ii

jj

j

MPMDP

MPMDPDM

1

|

||

Prior

Posterior

Reverend Thomas Bayes (1702-1761)That is how Humans and Machines learn

Bayesian Polynomial Regression

14

Bayes Rule:

𝑝 𝐰 𝐗, 𝐭 ∝ 𝑝 𝐭 𝐗,𝐰 𝑝 𝐰

predictive distribution:

𝑝 t 𝐱, 𝐗, 𝐭 = න𝑝 t 𝐱,𝐰 𝑝 𝐰 𝐗, 𝐭 𝑑𝐰

posterior likelihood prior

likelihood posterior

deg: 1

deg: 2

N. Paulson et al.

𝐰 𝑙 samples using MCMC (Metropolis-Hastings algorithm)

Uncertainty Quantification of Phase Diagrams

REFERENCES

3200

3100

3000

2900

2800

2700

Liquidus [2]

Liquidus [1]

Solidus [1]

Solidus [2]

3200

UO2 PuO2

3100

3000

2900

2800

2700

Te

mp

era

ture

(K

)

Mol % PuO2

20 40 60 80

UO2 + PuO2

Liquid

* M. Stan and B. J. Reardon, CALPHAD, 27 (2003) 319-323.

[1] M. G. Adamson, E. A. Aitken, and R. W. Caputi, J. Nucl. Mater., 130 (1985) 349-365.

[2] T. D. Chikalla, J. Am. Ceram. Soc., 47 (1964) 309-309.

• Uncertainty in fuel thermo-mechanical properties is often >10%

• Uncertainty of chemical properties (free energy) can be 10-15 %

Example:

• Uncertainty quantification the UO2-PuO2phase diagram*. ΔT = ±50K, δc = 5%

• Bayesian analysis of 15 data sets (melting temperatures , transformation enthalpies, …).

• Optimization via a genetic algorithm.

15

Bayesian Analysis

M. Stan and B. Reardon, CALPHAD 27 (2003) 319

TM

B

M

A

M

B

M

A TTHHM ,,,

TNDDDD ,...,, 21

Model parameter vector

Data vector

Diagram line model m

B

m

A

S

B

S

A

S

m

B

m

A

L

B

L

A

L

TTHHTfx

TTHHTfx

,,,,

,,,,

N

i

ii

jj

j

MPMDP

MPMDPDM

1

|

||Bayes’ Theorem

The mean model j

jj

i

i DMMM |

Standard deviation l

llCSTD

Correlation matrix mmll

lmlm

CC

CR

Posterior covariance matrix T

i

T

i

i

i MMDMMMC ||

16

Evolutionary (Genetic) Algorithms

17

TM

B

M

A

M

B

M

A TTHHM ,,,

TNDDDD ,...,, 21

TM

B

M

A

M

B

M

A TTHHM ,,,

m

B

m

A

L

B

L

A

S

m

B

m

A

L

B

L

A

L

TTHHTfx

TTHHTfx

,,,,

,,,,

18

Uncertainty of Hf-O Lattice Parameters

19

Thermodynamics of Hf-O1

1D. Shin, R. Arroyave, Z.K. Liu, CALPHAD, 30 (2006) 375-386

Uncertainty of the Hf-O Phase Diagram

20

THERMODYNAMIC DATA

NIST ThermoData Engine (TDE)

• Requires

– Knowledge Base: A trusted data archive with full, machine-interpretable metadata

– Inference Engine: software developed via systematic, test-driven analysis of real data systems with a team of subject experts

• Delivers

– A virtual data expert backed by a well-curated library for engineers

CMOS Gate Stack: it is very smallNIST/TRC SOURCE

Data Archival System

as of 2017-01-05

System Type

Chemical Systems

Data Points

Pure 24,000 1,600,000

Binary 52,000 3,300,000

Ternary 15,000 1,100,000

Reactions 7,200 17,000

TDE - NIST SRD 103a/b

Uncertainty Evaluation

Simulation software

Direct data dissemination

Property Prediction

Critically Evaluated Data

22

OUTLOOK

The role of computation is changing

24

19901950 1970 2010

com

ple

xity

calculations

document

processing

intelligent

assistance

simulations

Impact on Science and Technology

25

MATERIAL GENOME INITIATIVECreate a multi-dimensional Periodic Table of

multi-component, multi-phase materials

Personal assistance

Immersive visualization

26

Augmented reality

Impact on You