UNIT-V FMM.HYDRAULIC TURBINE - Construction and working
How to Leverage Artificial Intelligence to Accelerate Data Collection and Analysis of Diffusion Multiples
1. 1
How to Leverage Artificial Intelligence
to Accelerate Data Collection and
Analysis of Diffusion Multiples
Ji-Cheng (JC) Zhao
The Ohio State University
NIST Workshop on Artificial Intelligence for Materials Science (AIMS)
August 8, 2018
2. 2
• An example of alloy design using
CALPHAD & regression analysis
• Diffusion multiples for phase diagram
mapping
• Precipitation kinetics & microstructure
• Property mapping
• Summary
Outline
3. 3
• An example of alloy design using
CALPHAD & regression analysis
• Diffusion multiples for phase diagram
mapping
• Precipitation kinetics & microstructure
• Property mapping
• Summary
Outline
5. 5
0.1
1
10
100
0.1 1 10 100
PredictedTimetoFail(h)
Measured Time to Fail (h)
Regression done using
only alloy chemistry 0.1
1
10
100
0.1 1 10 100
PredictedTimetoFail(h)
Measured Time to Fail (h)
Regression done using
alloy chemistry and
predicted γ' volume fraction
γ’ volume fraction included
Computational thermodynamics for composition optimization
Regression-based prediction of creep strength in Ni-base superalloys
γ’ volume fraction not included
Crude machine learning of legacy data with model input
Zhao & Henry: Adv. Eng. Mater., 4, 501, 2002.
760 ºC rupture life 760 ºC rupture life
6. 6The National Academies Press, Washington, D.C., 2012
GTD222 GTD262
4 years from concept to production
• 2X creep strength ↑
• Ta replacement by Nb: cost ↓
• Computational design: validated with only one set of 4 alloys.
Now widely
used in GE
gas turbines
A Successful Example of Accelerated Alloy Design: GTD262
“The rapid development of GTD262 is the first
successful landmark that has helped establish
within GE the credibility of computational alloy
design and its associated methodologies,
models, and databases”
Inventors:
L. Jiang
J.-C. Zhao
G. Feng
7. 7
• An example of alloy design using
CALPHAD & regression analysis
• Diffusion multiples for phase diagram
mapping
• Precipitation kinetics & microstructure
• Property mapping
• Summary
Outline
8. 8
• Local equilibrium at phase
interfaces defines the tie-lines
• Interdiffusion creates all
single-phase compositions
0
10
20
30
40
50
60
70
80
90
100
0 100 200 300 400 500 600
Distance, microns
Composition,at.%
Co, at.%
Cr, at.%
fcc
σ
bcc
Co Cr
fcc σ bcc
Co Cr
100 µm1100°C, 1000h
Diffusion-Multiple Approach: the idea
10. 10
Mo Fe
Ni
0
5 12
14
100 µm
Ni
Mo Fe
δ γ
P
αFe
1 2
3
4
5
0
6 7 8 9 10 11 12
13
14
µ
15
16
1100°C
1500 hrs
Mo
FeNi
γ
αFe
µ
δ
P
0
1
2
3
4
5
6
7
8
9
10
11
121314
15
16
1100 °C
Diffusion-Multiple Approach: phase diagram mapping
Mo Mo
Mo
Cr
FeCr
FeCo
Co
Ni
Ni
26. 26
High-Throughput Property Mapping: thermal conductivity
A general model for thermal and electrical conductivity
Wei, Antolin, Restrepo, Windl, Zhao: Acta Mater., 126 (2017) 272
Now thermal & electrical resistivity can be incorporated into CALPHAD
27. 27
Wei, Zheng, Cahill, Zhao:
Rev. Sci. Instr. 2013
Pump
( f = 10 MHz )
Probe
Sample (Λ, C)
Al
8 µm
d ≈ 600 nm
fC
d
π2
Λ
=
Pump
( f = 100 KHz )
Probe
Sample (Λ, C)
Al
8 µm
d ≈ 6 µm
t (ns)
0 1 2 3 4
-Vin/Vout
0
5
10
15
Ni
Gd
MgO Si
f = 9.8 MHz
t (ns)
0.0 0.2 0.4 0.6 0.8 1.0
-Vin/Vout
0
5
10
15
Si
Ni
MgO
Gd
f = 123 KHz
0
20
40
60
80
100
120
140
0 20 40 60 80 100 120 140
Accepted CP J mol-1 K-1
MeasuredCPJmol-1K-1
B Si
MgAl2O4
Al2O3
Gd
MgO
Pd
Ni
Pt
Heat Capacity
Wei, Zheng, Cahill, Zhao, Rev. Sci. Instr. 84 (2013) 071301.
High-Throughput Property Mapping: heat capacity
32. 32
Artificial Intelligence & automation are desperately need for acceleration
Mo Mo
Mo
Cr
FeCr
FeCo
Co
Ni
Ni
Summary
• Enormous amounts of data can be extracted from
diffusion multiples
• Active learning & autonomy/automation are the future
• Interactive use of computed & prior data/knowledge
• Data fusion & autonomous fault detection