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: 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

2

Automated Parameter Tuning workings required improve performance 3 ▷ treat tunable process as black box – no knowledge of inner ▷ intelligently and iteratively select parameter settings likely to ▷ mature techniques used in many areas of AI

Optimizing Graphene Oxide Reduction material results 4 ▷ reduce graphene oxide to graphene through laser irradiation ▷ allows to create electrically conductive lines in insulating ▷ laser parameters need to be tuned carefully to achieve good

From Graphite/Coal to Carbon Electronics 5 Overview of the Process

Experimental Setup 6

Evaluation of Irradiated Material 7

Morphology of Irradiated Material 8

Bayesian Optimization with Surrogate Models on this 9 ▷ 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

Bayesian Optimization with Surrogate Models Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A http://arxiv.org/abs/1703.03373. Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. 10 Iter = 2, Gap = 1.5281e−01 0.8 ● ● y 0.4 type ● init ● ● prop 0.0 seq type y 0.03 yhat ei 0.02 ei 0.01 0.00 −1.0 −0.5 0.0 0.5 1.0 x

Bayesian Optimization with Surrogate Models Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A http://arxiv.org/abs/1703.03373. Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. 11 Iter = 3, Gap = 1.5281e−01 0.8 ● ● y 0.4 type ● init ● ● prop 0.0 seq type y 0.020 yhat ei 0.015 ei 0.010 0.005 0.000 −1.0 −0.5 0.0 0.5 1.0 x

Bayesian Optimization with Surrogate Models Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A http://arxiv.org/abs/1703.03373. Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. 12 Iter = 4, Gap = 1.3494e−02 0.8 ● ● y 0.4 type ● init ● ● prop 0.0 seq type y yhat 0.010 ei ei 0.005 0.000 −1.0 −0.5 0.0 0.5 1.0 x

Bayesian Optimization with Surrogate Models Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A http://arxiv.org/abs/1703.03373. Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. 13 Iter = 5, Gap = 1.3494e−02 0.8 ● ● y 0.4 type ● ● init ● prop 0.0 seq type y 0.015 yhat ei 0.010 ei 0.005 0.000 −1.0 −0.5 0.0 0.5 1.0 x

Bayesian Optimization with Surrogate Models Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A http://arxiv.org/abs/1703.03373. Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. 14 Iter = 6, Gap = 2.1938e−06 0.8 ● ● y 0.4 type ● init ● ● prop 0.0 seq type y 0.006 yhat ei 0.004 ei 0.002 0.000 −1.0 −0.5 0.0 0.5 1.0 x

Bayesian Optimization with Surrogate Models Bischl, Bernd, Jakob Richter, Jakob Bossek, Daniel Horn, Janek Thomas, and Michel Lang. “MlrMBO: A http://arxiv.org/abs/1703.03373. Modular Framework for Model-Based Optimization of Expensive Black-Box Functions,” March 9, 2017. 15 Iter = 7, Gap = 2.1938e−06 0.8 ● ● y 0.4 type ● init ● ● prop 0.0 seq type y yhat 1e−03 ei ei 5e−04 0e+00 −1.0 −0.5 0.0 0.5 1.0 x

https://www.automl.org/book/ 16

Tuned Parameters (710 ms to 20 210 ms), pressure in reaction chamber (10 psi to 100 psi) evaluations, about 2 weeks of human operator time 17 ▷ laser power (1 mW to 4400 mW), duration for irradiating spot ▷ ≈ 7.8 billion confjgurations ▷ individual graphene oxide sample allows for max 361

Tuned Parameters 18 ● ● ● ● ● ● ● 6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Ratio ● ● ● ● ● ● ● ● ● ● ● 4 ● ● ● ● ● ● ● 2 0 10 20 30 40 50 Iteration

Tuned Parameters evaluations) machine learning 19 ● ● 6 ● ● ● ● ● ● Ratio 4 + Prediction 2 • Actual 50 um 50 um 0 2 4 6 8 Iteration After 1 st prediction During Training ▷ improvement of factor of two over best result in literature ▷ good results even with small amount of initial data (19 ▷ code can be used by domain experts with no background in

Explored Parameter Space 20 48 47 12 18 29 1 5 9 7 21 17 35 6 37 42 22 19 31 38 33 28 25 10 32 11 13 2 27 34 8 16 20 30 39 Ratio 6 3 44 43 46 45 14 26 24 40 4 4 36 41 ● 23 ● 15 ● ● ● ● ● 2 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Parameter Space

Tuned Parameters – Kapton argon, nitrogen 21 ▷ extend parameter space with gas in reaction chamber – air, ▷ extend ranges of other parameters ▷ more and longer experimental campaigns

Tuned Parameters – Kapton 22 5 4 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 3 ● ● Ratio ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 2 ● ● ● ● ● 1 ● ● 0 0 10 20 30 40 50 Iteration

Explored Parameter Space – Kapton 23 ● 5 14 8 42 2 33 20 4 4 35 29 22 16 11 9 36 5 ● 17 1 38 15 39 19 40 6 3 Ratio ● 28 24 18 12 23 3 10 7 13 ● 49 44 46 34 ● 50 43 ● 47 31 27 2 30 48 37 25 32 ● ● 26 ● ● 21 1 ● 45 ● 41 ● ● ● ● ● ● ● ● 0 (time − 7080.843) * −1 + (power − 2536.714) * −0.006 + (pressure − 576.286) * −0.016 + (gas − 1.9) * 0

Design of New Materials absorption of material 24 ▷ optimize parameters of pattern generator for energy ▷ six numeric parameters ▷ computational evaluation of candidates

ML-Optimized Generator Parameters 25 4 ● ● ● ● ● 3 ● ● ● ● fitness ● ● ● ● ● ● 2 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 1 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Iteration

26 ML-Optimized Generator Parameters 4 3 target 2 1 0 (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

ML-Optimized Generator Parameters 27

Outlook performance (and other approaches) learned 28 ▷ automate experimental setup ▷ application to other materials ▷ more in-depth investigation of Bayesian Optimization ▷ inform understanding of process by what surrogate model has

Other Projects graphene 29 ▷ optimization of wear of buttons ▷ density functional theory (DFT) calculations of properties of ▷ optimization of DFT calculations

Challenges and Opportunities 30 ▷ sparsity of data ▷ multi-scale measurements ▷ combination of optimization with experiments and simulations

Do Try This at Home Tutorial on AI for Materials Science @ IJCAI 2019 https://www.cs.uwyo.edu/~larsko/aimat-tut/ Simulator optimizers available data 31 ▷ build surrogate model based on (relatively) large amount of ▷ Bayesian Optimization based on this surrogate model ▷ playground to try your own approaches

Summary 32 ● ● 6 ● ● ● ● ● ● Ratio 4 + Prediction 2 Actual • 50 um 50 um 0 2 4 6 8 Iteration After 1 st prediction During Training

AIM Artifjcially Intelligent Manufacturing Center @ University of Wyoming www.uwyo.edu/aim 33