Artificial Intelligence-Aided Materials Design: AI-Algorithms and Case Studies on Alloys and Metallurgical Processes

Cover
Half Title
Title Page
Copyright Page
Dedication
Contents
Foreword
Preface
Acknowledgments
Authors Biographies
1. Introduction: Artificial Intelligence and Materials Design
	1.1 Data-Driven Materials Science: Initiatives, Goals, and Progress
	1.2 Multiscale-Materials Modeling
		1.2.1 Initiatives in the United States
		1.2.2 The Integrated Computational Materials Engineering Expert Group (ICMEg)
	1.3 Virtual Materials Design (VMD)
	1.4 Challenges (MGI/ICME/VMD)
		1.4.1 Potential Solution Through Data Science/Artificial Intelligence
	1.5 This Book and Its contribution Towards MGI/ICME and Virtual Materials Design (VMD)
	1.6 Aim and Scope of Current Book
		1.6.1 Discovering and Optimizing Chemistry/Processing-Structure-Property (C/PSP) Linkages
			1.6.1.1 State of the Art
			1.6.1.2 Challenge
			1.6.1.3 Research Plan
		1.6.2 Data Science/Artificial Intelligence Techniques for Accelerating the Discovery and Deployment of New Materials
			1.6.2.1 State of the Art
			1.6.2.2 Challenge
			1.6.2.3 Research Plan
		1.6.3 Data from Production: Industrial Furnaces and Rolling Mill
			1.6.3.1 Iron-Making Blast Furnace
			1.6.3.2 Basic Oxygen Furnace (BOF) or LD (Linz Donawitz) Steelmaking Furnace
			1.6.3.3 Rolling Mill
		1.6.4 Nano Mechanics and Atomic Force Microscope (AFM) Imaging
		1.6.5 Thermo-Mechanical Treatment
		1.6.6 Phase Field
		1.6.7 Atomistic Simulations
			1.6.7.1 Vienna Atomistic Simulation Package (VASP)
			1.6.7.2 Schrodinger Materials Design Suite
		1.6.8 Efficient Way of Choosing a Supervised Machine Learning Algorithm
			1.6.8.1 State of the Art
			1.6.8.2 Challenges
			1.6.8.3 Research Plan
		1.6.9 CALPHAD Approach
	1.7 CASE STUDY #1: Uncertainty Quantification and Propagation Studied Under the Framework of CALPHAD Approach Using PyCalphad and ESPEI Software
		1.7.1 Uncertainty Quantification
		1.7.2 Uncertainty Propagation
		1.7.3 Conclusion: Uncertainty Quantification and Propagation
2. Metallurgical/Materials Concepts
	2.1 Materials/Alloy Design
		2.1.1 Effect of Chemical Composition
		2.1.2 Effect of Processing
		2.1.3 Effect of Structure
		2.1.4 Effect of Composition, Processing, and Structure on Properties (C/P-S-P)
	2.2 CALPHAD Approach
		2.2.1 Precipitation Kinetics: Nucleation and Growth of Grains/Crystals of Desired Phase
		2.2.2 CALPHAD-Based Software and Database: Application and Limitations
			2.2.2.1 THERMOCALC (Andersson et al. 2002)
			2.2.2.2 FACTSAGE (Bale et al. 2002)
				2.2.2.2.1 Pareto-Optimal Set in CALPHAD-Based Software FACTSAGE
			2.2.2.3 JMatPro®: (JMatPro 2021)
			2.2.2.4 PyCalphad/ESPEI (Otis and Liu 2017, Otis et al. 2021)
				2.2.2.4.1 Phase Diagram in PyCalphad
	2.3 Nanomechanics and Nanotribology
		2.3.1 Nanoindentation
		2.3.2 Atomic Force Microscopy (AFM) imaging
		2.3.3 Modeling and Simulation: Case Study in This Book
	2.4 Magnetism and Magnetic Terminologies
		2.4.1 Magnetization (M) (Buschow and de Boer 2003)
		2.4.2 Magnetic Field (H) (Buschow and de Boer 2003)
		2.4.3 Magnetic Flux Density (Dilon 2014)
		2.4.4 Core Loss (Pcv) (Herzer 1993)
		2.4.5 Coercivity (Hc) (Cullity and Graham 2009, Buschow and de Boer 2003)
		2.4.6 Magnetic Anisotropy (Herzer 1993)
		2.4.7 Magnetostriction (Λs) (Buschow and de Boer 2003)
		2.4.8 Permeability (Μ) (Cullity and Graham 2009, Buschow and de Boer 2003)
		2.4.9 Curie Temperature (Cullity and Graham 2009)
		2.4.10 Hard Magnets (Jha et al. 2017, Mcguiness et al. 2015)
		2.4.11 Soft Magnets (Buschow and de Boer 2003, Herzer 1993)
	2.5 Extractive/Process Metallurgy
		2.5.1 Blast Furnace Iron Making
			2.5.1.1 Hot Metal Production
			2.5.1.2 Furnace Variables
				2.5.1.2.1 Iron Ore
				2.5.1.2.2 Coke (Biswas 1981, Smith 2017)
				2.5.1.2.3 Flux (Ghosh and Chatterjee 2015)
			2.5.1.3 Blast Furnace Process Description (Ghosh and Chatterjee 2015, Zhou et al. 2016)
				2.5.1.3.1 Indirect Reduction Zone
				2.5.1.3.2 Direct Reduction and Melting Zone
			2.5.1.4 Productivity
		2.5.2 Steelmaking
			2.5.2.1 Basic Oxygen Furnace (BOF): LD (Linz Donawitz) Furnace (Turkdogan 1996)
				2.5.2.1.1 Input: Raw Materials Charged into the Converter
				2.5.2.1.2 Output: Materials Produced After Blowing Operation Is Complete
				2.5.2.1.3 Reactions in BOF (Ghosh and Chatterjee 2015)
					2.5.2.1.3.1 Reactions at Oxygen Jet Impact for Top-Blown Converters
					2.5.2.1.3.2 Removal of Impurity During Blowing Operation
				2.5.2.1.4 Slag
	2.6 Nickel-Base Superalloys
		2.6.1 Critical Phases
		2.6.2 Effect of Alloying Elements
	2.7 Aluminum Alloys
	2.8 Titanium Alloys
		2.8.1 High-Temperature Applications
			2.8.1.1 Phases
			2.8.1.2 Titanium Aluminides (Gupta and Pant 2018, Wu 2006)
		2.8.2 Biomaterials
	2.9 Utilizing Data from Other Approaches
		2.9.1 Density Functional Theory/Ab-initio Method
			2.9.1.1 VASP
			2.9.1.2 Schrodinger Materials Design Suite
	2.10 CASE STUDY # 2: Inverse Determination of Phase-Field Model Parameters for a Targeted Grain Size Through AI Algorithms
		2.10.1 Phase-Field Approach (Moelans et al. 2015, Chang and Moelans 2015)
			2.10.1.1 Background
			2.10.1.2 Phase-Field Variables Identification for This Study
			2.10.1.3 Application of AI Algorithms
			2.10.1.4 Application of AI-Based Algorithm on Phase-Field Data
				2.10.1.4.1 Generation of New Input Parameters
			2.10.1.5 Parallel Coordinate Chart (PCC)
			2.11 Conclusions
3. Artificial Intelligence Algorithms
	3.1 Introduction
	3.2 Data Mining
		3.2.1 Uncertainties
			3.2.1.1 Uncertainty Propagation
			3.2.1.2 Uncertainty Mitigation
			3.2.1.3 Uncertainty Management
			3.2.1.4 Case Studies on Uncertainties
		3.2.2 Collecting Data from Various Sources
		3.2.3 Pre-Processing/Cleaning Data from Various Sources
		3.2.4 Scaling Data
		3.2.5 Statistics and Visualization
		3.2.6 Preparation/Selection of Data for Development of Surrogate Models Through Machine Learning:= (Box and Draper 1987)
	3.3 Designing Experiments or Predicting Experimental Parameters Through Data Mining
		3.3.1 Random Sequence Generators: (mode FRONTIER)
		3.3.2 Sobol's Algorithm (Sobol 1967)
		3.3.3 Adaptive Space Filler (mode FRONTIER)
		3.3.4 Importance of Support Points in Developing Accurate Predictive Models (Box and Draper 1987)
	3.4 Surrogate Models/Meta-Models
		3.4.1 Selection of Machine Learning Algorithms Based on Nature of Data
		3.4.2 Performance Measurements of a Metamodel
			3.4.2.1 R Square (R2)
			3.4.2.2 Relative Average Absolute Error (RAAE)
			3.4.2.3 Relative Maximum Absolute Error (RMAE)
			3.4.2.4 AIC (Akaike Information Criterion): (mode FRONTIER)
		3.4.3 Statistical Approaches: (mode FRONTIER)
	3.5 Machine Learning: Supervised (Box and Draper 1987, Mueller et al. 2016)
		3.5.1 Artificial Neural Networks (ANNs) (Basheer and Hajmeer 2000)
		3.5.2 Deep Learning (Lecun et al. 2015)
			3.5.2.1 Tensorflow and Keras
		3.5.3 Genetic Programming
		3.5.4 K-Nearest Neighbor
		3.5.5 Kriging Model (modeFRONTIER 2021)
		3.5.6 Gaussian Processes (modeFRONTIER 2021, Schulz et al. 2018, Tancret 2012)
		3.5.7 Hybrid Meta-models (modeFRONTIER 2021)
		3.5.8 Radial Basis Functions (RBF) (modeFRONTIER 2021)
		3.5.9 Polynomial Regression Model (PR) (modeFRONTIER 2021)
	3.6 Machine Learning: Unsupervised (modeFRONTIER 2021)
		3.6.1 Principal Component Analysis (PCA) (modeFRONTIER 2021)
		3.6.2 Hierarchical Clustering Analysis (HCA) (modeFRONTIER 2021)
		3.6.3 Self-Organizing Maps (SOM) (modeFRONTIER 2021, Jha et al. 2017)
	3.7 Multi-Objective Optimization
		3.7.1 Pareto-Optimality
			3.7.1.1 Pareto-Optimal Set
			3.7.1.2 Evolutionary Algorithms (EA) for Multi-Objective Optimization
	3.8 Evolutionary Algorithms (EAs)/Genetic Algorithms (GAs)
		3.8.1 Non-Dominated Sorting Genetic Algorithm (NSGA-II)
			3.8.1.1 Elitist Preserving Strategy
			3.8.1.2 Crowding Distance Selection
		3.8.2 Predator-Prey Genetic Algorithm (PPGA) (Pettersson et al. 2007)
		3.8.3 Multi-Objective Particle Swarm Optimization (MOPSO)
		3.8.4 Simulated Annealing (SA)
			3.8.4.1 Multi-Objective Simulated Annealing (MOSA)
			3.8.4.2 Archived Multi-Objective Simulated Annealing (AMOSA)
		3.8.5 Evolution Strategies (ESs)
			3.8.5.1 Multi-Membered ES (MMES)
			3.8.5.2 Recombinative ES
		3.8.6 Differential Evolution (DE)
			3.8.6.1 Initialization of Initial Population
			3.8.6.2 Differential Mutation
			3.8.6.3 Crossover
			3.8.6.4 Selection
		3.8.7 Strength Pareto Evolutionary Algorithm 2 (SPEA 2)
	3.9 Inverse Design
	3.10 Multi-Criterion Decision Making (MCDM) (Miettinen 1999)
	3.11 CASE STUDY #3: Designing New Molecules/Refrigerants by Computational Chemistry and AI Approach Using Schrodinger Materials Science Suite Software
		3.11.1 Background
		3.11.2 Global Warming Potential and Infrared Absorption Spectra
		3.11.3 Conclusions
4. Case Study #4: Computational Platform for Developing Predictive Models for Predicting Load-Displacement Curve and AFM Image: Combined Experimental-Machine Learning Approach
	4.1 Introduction: Background
	4.2 Case Studies
		4.2.1 Cold Sprayed Aluminum Based Bulk Metallic Glass (AL-BMG) Coating (Pitchuka et al. 2014, 2016)
		4.2.2 Al-5CNT Coating (Bakshi et al. 2010)
	4.3 Results: Application of Machine Learning Algorithms
		4.3.1 Case Study 4.2.1: Load-Displacement Curve Prediction for AL-BMG as Sprayed/Cast Coating
			4.3.1.1 ANN Model Development in This Work
		4.3.2 Case Study 4.2.1: AFM Image Prediction for AL-BMG as Sprayed/Cast Coating
		4.3.3 Case Study 4.2.2: Load-Displacement Curve Prediction for Al-5CNT Coating
		4.3.4 Case Study 4.2.2: AFM Image Prediction for Al-5CNT Coating
		4.3.5 HA CNT
	4.4 Conclusions
5. Case Study #5: Design of Hard Magnetic AlNiCo Alloys: Combined Machine Learning-Experimental Approach
	5.1 Introduction: Description of Case Study and Goals
	5.2 Design Variables (composition), Targeted Properties, Metamodeling, and Multi-Objective Optimization Problem Formulation
		5.2.1 Data Acquisition
		5.2.2 Targeted Properties and Multi-Objective Problem Formulation (Jha 2016)
		5.2.3 Software Used in This Work
	5.3 Application of Supervised Machine Learning Algorithms
	5.4 Multi-Objective Optimization for Determining Novel Composition of New Magnets
	5.5 Application of Unsupervised Machine Learning Algorithms
	5.6 Comparison of Experimental and Predictions from AI-Based Approaches with Physical Models
		5.6.1 Comparison with Commercial Alloys
		5.6.2 Comparison: AI Predictions vs Experiments
	5.7 Conclusions
6. Case Study #6: Design and Discovery of Soft Magnetic Alloys: Combined Experiment-Machine Learning-CALPHAD Approach
	6.1 Introduction
	6.2 Ab-Initio DFT: Estimation of Magnetic States for Various Phases
	6.3 CALPHAD: Thermodynamics/Phase Transformation and Kinetics
		6.3.1 Identification of Stable and Metastable Phases (Jha et al. 2017)
		6.3.2 Precipitation Kinetics Simulation (Jha et al. 2017, 2018)
	6.4 Application of Machine Learning Algorithms in Optimizing Precipitation Kinetics of Desired Fe3Si Phase (Α″ With D03 Structure) (Jha et al. 2018)
	6.5 Experimental Validation
		6.5.1 Isothermal Annealing at 540°C for 1 Hour (Jha et al. 2019)
			6.5.1.1 CALPHAD Model Parametrization
		6.5.2 Annealing Time of 0.1 Seconds (Diercks et al. 2020)
	6.6 Discovery of New Soft Magnetic Alloys Through the AI Approach
		6.6.1 Unsupervised Machine Learning
		6.6.2 Supervised Machine Learning and Multi-Objective Optimization
	6.7 Conclusions
7. Case Study #7: Nickel-Based Superalloys: Combined Machine Learning-CALPHAD Approach
	7.1 Introduction
	7.2 CALPHAD Approach: Thermodynamics/Phase Transformation
	7.3 Application of Machine Learning Algorithms
		7.3.1 Visualization and Correlation
		7.3.2 Inverse Design of Chemistry/Composition for Given Stress and Time to Rupture
	7.4 Inverse Determination of Chemistry of Alloy for a Given Value of Stress and Time to Rupture
	7.5 Conclusion from Inverse Design
8. Case Study #8: Design of Aluminum Alloys: Combined Machine Learning-CALPHAD Approach
	8.1 Introduction: Background
	8.2 Materials and Methods: CALPHAD Approach: Thermodynamics/Phase Transformation and Precipitation Kinetics
		8.2.1 CALPHAD: Identification of Stable and Metastable Phases
	8.3 Results: Application of Machine Learning Algorithms
		8.3.1 Deep Learning Artificial Neural Network (DL-ANN) Model
		8.3.2 Self-Organizing Map (SOM) Analysis
		8.3.3 6XXX Alloys
	8.4 Conclusions
9. Case Study #9: Titanium Alloys for High-Temperature Applications: Combined Machine Learning-CALPHAD Approach
	9.1 Problem Formulation
	9.2 CALPHAD Approach: Thermodynamics/Phase Transformation
		9.2.1 Final Data from CALPHAD That Will Be Used for SOM Analysis
	9.3 Application of Machine Learning Algorithms
	9.4 Titanium Aluminides
		9.4.1 Conventional Experimental Design
			9.4.1.1 Challenges
		9.4.2 Proposed Method
	9.5 Conclusions
10. Case Study #10: Design of β-Stabilized, ω-Free, Titanium-Based Biomaterials: Combined Machine Learning-CALPHAD Approach
	10.1 CALPHAD Approach: Thermodynamics/Phase Transformation
	10.2 Application of Machine Learning Algorithms for Predicting Novel Compositions/Temperature for Improving Phase Stability of Desired β-Phase While Avoiding Detrimental α- and ω-Phases
		10.2.1 Problem Formulation
			10.2.1.1 Ti-Ta-Nb-Sn-Mo-Zr-Fe-Cr-V System
			10.2.1.2 Ti-Nb-Ta System
	10.3 Supervised Machine Learning in RapidMiner
	10.4 Unsupervised Machine Learning in modeFRONTIER: Self-Organizing Maps (SOMs)
	10.5 Unsupervised Machine Learning in modeFRONTIER: Principal Component Analysis (PCA)
	10.6 Multi-Objective Optimization in modeFRONTIER
		10.6.1 Variable Screening
		10.6.2 Development of Supervised Machine Learning Models
		10.6.3 Multi-Objective Optimization
	10.7 Conclusions
11. Case Study #11: Industrial Furnaces I: Application of Machine Learning on Industrial Iron-Making Blast Furnace Data
	11.1 Blast Furnace Iron-Making
		11.1.1 Blast Furnace Operation: Analytical Model and Data-Driven Model
		11.1.2 CO2 Emissions from a Blast Furnace
		11.1.3 Hot Metal Temperature and Silicon Content in Hot Metal
	11.2 Dataset Used in This Chapter
	11.3 Furnace 1: 2-Month Data Was Used
		11.3.1 Supervised Machine Learning in RapidMiner
		11.3.2 Unsupervised Machine Learning in modeFRONTIER
		11.3.3 Multi-Objective Optimization in modeFRONTIER
			11.3.3.1 Variable Screening Through ANOVA
			11.3.3.2 Model Development by Supervised Machine Learning
			11.3.3.3 Multi-Objective Optimization Through Genetic Algorithm
	11.4 Furnace 2: 1-Month Data Was Used
		11.4.1 Analysis of Data with Respect to Operating Parameters and Objectives
		11.4.2 Unsupervised Machine Learning: Self-Organizing Maps
	11.5 Furnace 3: 1-Year Data Was Used for This Analysis
		11.5.1 Data Processing
			11.5.1.1 Variable Screening
		11.5.2 Unsupervised Machine Learning: Self-Organizing Maps
		11.5.3 Multi-Objective Problem Formulation
			11.5.3.1 Case 1: Two Objectives and One Constraint
			11.5.3.2 Case 2: Two Objectives and Two Constraints
12. Case Study #12: Industrial Furnaces II: Development of GUI/APP to Determine Additions in LD Steelmaking Furnace
	12.1 Introduction: LD Steelmaking
	12.2 Development of GUI: Formulation of Analytical Model
		12.2.1 Calculations: Assumptions
		12.2.2 Material Balance
	12.3 GUI/APP in MATLAB
	12.4 Simulation of LD steelmaking process in Thermo-Calc
	12.5 CALPHAD approach: Determining effect of various microalloying elements on crack succeptiblity coefficient of Al-Si killed microalloyed steel
	12.6 Conclusion
13. Case Study #13: Selection of a Supervised Machine Learning (Response Surface) Algorithm for a Given Problem
	13.1 Introduction: Background
		13.1.1 Purpose of This Study: Benefits of Supervised ML-Based Predictive Models
		13.1.2 Purpose of This Study: Challenges or Limitations
	13.2 Method
		13.2.1 Modules Used in modeFRONTIER
		13.2.2 Accuracy Measurements
	13.3 Results: Error Metrics for Various Supervised Machine Learning Models
		13.3.1 Performance Analysis for Response Surfaces Schittkowski Test Cases (2-D) 1-24
		13.3.2 Suggested Supervised ML-Based Method for Various Schittkowski Test Cases
	13.4 KNIME Software: Schittkowski Test Problem 47
	13.5 Conclusions
14. Case Study #14: Effect of Operating Parameters on Roll Force and Torque in an Industrial Rolling Mill: Supervised and Unsupervised Machine Learning Approach
	14.1 Introduction: Rolling Data
	14.2 Unsupervised Machine Learning: Self-Organizing Maps (SOMs)
	14.3 Supervised Machine Learning: K-Nearest Neighbor Algorithm in modeFRONTIER
		14.3.1 Variable Screening
		14.3.2 Predictive Models Developed Through Supervised Machine Learning
	14.4 KNIME: Random Forest Algorithm
	14.5 Hierarchical Clustering Analysis (HCA)
	14.6 Principal Component Analysis
	14.7 Plate Mill: Multi-objective Optimization of YS, UTS and Elongation Percent of Micro-alloyed Steel Plates
	14.8 Coiling: Multi-objective Optimization of YS, UTS and Elongation Percent of AHSS Coils
	14.9 Conclusions
15. Case Study #15: Developing Predictive Models for Flow Stress by Utilizing Experimental Data Generated from Gleeble Testing Machine: Combined Experimental-Supervised Machine Learning Approach
	15.1 Introduction: Thermomechanical Data from Gleeble Testing Machine
	15.2 Supervised Machine Learning: Development of Predictive Model for Stress
		15.2.1 Variable Screening
		15.2.2 Development of Prediction Model for Stress
	15.3 JMatPro Analysis and Comparison with Present Work
		15.3.1 JMatPro Analysis
		15.3.2 JMatPro Analysis and Experimental Results
		15.3.3 Experiments, JMatPro Analysis, and Supervised Machine Learning
	15.4 Distribution of Stress Studied Through Supervised Machine Learning Model Predictions
		15.4.1 Prediction for New Experimental Parameters
		15.4.2 Stress Mapping: Supervised Machine Learning Predictions
16. Computational Platforms Used in This Work
	16.1 Computational Infrastructure
		16.1.1 Computer for Artificial Intelligence (AI)-Based Work
		16.1.2 CALPHAD-Based Work
		16.1.3 Online Platforms Used for Generating Results
		16.1.4 Android Phone (Redmi Note 7 Pro)
		16.1.5 Apple iPhone 6S Plus
		16.1.6 Supercomputer/High-Performance Computing (HPC)
	16.2 ESTECO modeFRONTIER
	16.3 MATLAB (MATLAB 2019)
		16.3.1 Deep Learning Toolbox
		16.3.2 PlatEMO: Evolutionary Multi-Objective Optimization (Tian et al. 2017)
	16.4 Sigma Technologies: IOSO (Egorov 1998)
	16.5 Cyberdyne: KIMEME
	16.6 Python Programming Language Packages
		16.6.1 TensorFlow and Keras: Open Source
		16.6.2 PyTorch (Paszke et al. 2017)
	16.7 R-Studio (R-Studio 2021)
	16.8 IBM SPSS (IBM SPSS 2015)
	16.9 WEKA: Open Source (Hall et al. 2009)
	16.10 CITRINE Informatics (O'Mara et al. 2016)
	16.11 KNIME (KNIME 2021)
	16.12 RapidMiner (RapidMiner 2021)
	16.13 Atomic Scale Simulation-Based Software
		16.13.1 Materials Project (Jain et al. 2013)
		16.13.2 Vienna Atomic Simulation Package (VASP) (VASP 2021)
		16.13.3 Schrodinger Materials Science Suite (Bochevarov et al. 2013)
	16.14 Case Study #16: Developing AI-Based Model for 100 Variable Problems in ABACUS AI
		16.14.1 ABACUS AI: Open Source and Proprietary (Abacus AI 2021)
		16.14.2 Deep Learning Model
	16.15 Conclusions
References
Index