AI for Materials Science: Tuning Laser-Induced Graphene Production - - PowerPoint PPT Presentation
AI for Materials Science: Tuning Laser-Induced Graphene Production and Beyond Lars Kotthofg and others who did the actual work Artifjcially Intelligent Manufacturing Center larsko@uwyo.edu Leiden, 29 August 2019 AI for Materials Science:
Lars Kotthofg and others who did the actual work
Artifjcially Intelligent Manufacturing Center larsko@uwyo.edu Leiden, 29 August 2019
Lars Kotthofg and others who did the actual work
Artifjcially Intelligent Manufacturing Center larsko@uwyo.edu Leiden, 29 August 2019
2
▷ treat tunable process as black box – no knowledge of inner workings required ▷ intelligently and iteratively select parameter settings likely to improve performance ▷ mature techniques used in many areas of AI
3
▷ reduce graphene oxide to graphene through laser irradiation ▷ allows to create electrically conductive lines in insulating material ▷ laser parameters need to be tuned carefully to achieve good results
4
5
6
7
8
▷ evaluate small number of initial (random) confjgurations ▷ build surrogate model of parameter-performance surface based
▷ use model to predict where to evaluate next ▷ repeat ▷ allows targeted exploration of new confjgurations
9
ei −1.0 −0.5 0.0 0.5 1.0 0.0 0.4 0.8 0.00 0.01 0.02 0.03
x type
prop seq
type
y yhat ei
Iter = 2, Gap = 1.5281e−01
Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. http://arxiv.org/abs/1703.03373. 10
ei −1.0 −0.5 0.0 0.5 1.0 0.0 0.4 0.8 0.000 0.005 0.010 0.015 0.020
x type
prop seq
type
y yhat ei
Iter = 3, Gap = 1.5281e−01
Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. http://arxiv.org/abs/1703.03373. 11
ei −1.0 −0.5 0.0 0.5 1.0 0.0 0.4 0.8 0.000 0.005 0.010
x type
prop seq
type
y yhat ei
Iter = 4, Gap = 1.3494e−02
Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. http://arxiv.org/abs/1703.03373. 12
ei −1.0 −0.5 0.0 0.5 1.0 0.0 0.4 0.8 0.000 0.005 0.010 0.015
x type
prop seq
type
y yhat ei
Iter = 5, Gap = 1.3494e−02
Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. http://arxiv.org/abs/1703.03373. 13
ei −1.0 −0.5 0.0 0.5 1.0 0.0 0.4 0.8 0.000 0.002 0.004 0.006
x type
prop seq
type
y yhat ei
Iter = 6, Gap = 2.1938e−06
Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. http://arxiv.org/abs/1703.03373. 14
ei −1.0 −0.5 0.0 0.5 1.0 0.0 0.4 0.8 0e+00 5e−04 1e−03
x type
prop seq
type
y yhat ei
Iter = 7, Gap = 2.1938e−06
Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. http://arxiv.org/abs/1703.03373. 15
https://www.automl.org/book/
16
▷ laser power (1 mW to 4400 mW), duration for irradiating spot (710 ms to 20 210 ms), pressure in reaction chamber (10 psi to 100 psi) ▷ ≈7.8 billion confjgurations ▷ individual graphene oxide sample allows for max 361 evaluations, about 2 weeks of human operator time
17
4 6 10 20 30 40 50
Iteration Ratio
18
4 6 2 4 6 8
Iteration Ratio
During Training After 1st prediction + Prediction
50 um 50 um
▷ improvement of factor of two over best result in literature ▷ good results even with small amount of initial data (19 evaluations) ▷ code can be used by domain experts with no background in machine learning
19
4 3 45 5 7 6 8 2 15 13 14 12 17 26 21 22 27 19 20 18 25 9 24 31 40 32 35 29 38 41 34 33 11 30 28 10 16 23 46 36 42 37 39 47 48 44 43 2 4 6
Parameter Space Ratio
20
▷ extend parameter space with gas in reaction chamber – air, argon, nitrogen ▷ extend ranges of other parameters ▷ more and longer experimental campaigns
21
2 3 4 5 10 20 30 40 50
Iteration Ratio
22
2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 1 2 3 4 5
(time − 7080.843) * −1 + (power − 2536.714) * −0.006 + (pressure − 576.286) * −0.016 + (gas − 1.9) * 0
Ratio
23
▷ optimize parameters of pattern generator for energy absorption of material ▷ six numeric parameters ▷ computational evaluation of candidates
24
2 3 4
Iteration fitness
25
1 2 3 4
(f − 0.086) * 0 + (k − 0.05) * 0 + (du − 0) * 0 + (dv − 0) * 0 + (y − 93.618) * −0.007 + (x − 81.067) * −0.01 + (sequence − 338.5) * −1 + (part − 6.644) * −0.018
target
26
27
▷ automate experimental setup ▷ application to other materials ▷ more in-depth investigation of Bayesian Optimization performance (and other approaches) ▷ inform understanding of process by what surrogate model has learned
28
▷ optimization of wear of buttons ▷ density functional theory (DFT) calculations of properties of graphene ▷ optimization of DFT calculations
29
▷ sparsity of data ▷ multi-scale measurements ▷ combination of optimization with experiments and simulations
30
Tutorial on AI for Materials Science @ IJCAI 2019 https://www.cs.uwyo.edu/~larsko/aimat-tut/ Simulator optimizers available ▷ build surrogate model based on (relatively) large amount of data ▷ Bayesian Optimization based on this surrogate model ▷ playground to try your own approaches
31
4 6 2 4 6 8
Iteration Ratio
During Training After 1st prediction + Prediction
50 um 50 um 32
Artifjcially Intelligent Manufacturing Center @ University of Wyoming www.uwyo.edu/aim
33