Impact and Opportunities of Artificial Intelligence Techniques in the Steel Industry: Ongoing Applications, Perspectives and Future Trends (Advances in Intelligent Systems and Computing)

This book collects perceptions and needs expectations and experiences concerning the application of Artificial Intelligence (AI) and Machine Learning in the steel sector. It contains a selection of themes discussed within the Workshop entitled “Impact and Opportunities of Artificial Intelligence in the Steel Industry” organized by the European Steel Technology Platform as an online event from October 15 until November 5, 2020. The event aimed at analyzing the diffusion of AI technologies in steelworks and at providing indications for future research, development and innovation actions addressing the sector demands. The chapters treat general analyses on transversal themes and applications for process optimization, product quality enhancement, yield increase, optimal exploitation of resources and smart data handling. The book is devoted to researchers and technicians in the steel or AI fields as well as for managers and policymakers exploring the opportunities provided by AI in industry. Preface Organization Workshop Chairs Workshop Board Contents About the Authors Challenges and Frontiers in Implementing Artificial Intelligence in Process Industry 1 Introduction 2 Challenges for AI in Process Industry 2.1 Technological and Infrastructural Challenges 2.2 Data as Primary Resource that Drives Artificial Intelligence 2.3 Security Challenges 2.4 Social Challenges Caused by Staff Acceptance 2.5 Social Challenges Caused by Expert Acceptance 3 Frontiers for AI 3.1 Models 3.2 Forecasting of Production Chain 3.3 Optimisation 3.4 Semantics 4 Summary References Data Pre-processing for Efficient Design of Machine Learning-Based Models to be Applied in the Steel Sector Abstract 1 Introduction 2 Typical Pre-processing Stages 2.1 Outlier Detection 2.2 Redundant Variables Elimination 2.3 Variable Selection 3 Steel Industrial Applications 3.1 First Case Study: Inclusion Classification 3.2 Second Case Study: Ripple Defect Detection 3.3 Third Case Study: Mechanical Property Prediction 4 Results 4.1 Results on First Case Study 4.2 Results on Second Case Study 4.3 Results on Third Case Study 5 Conclusions References Quantifying Uncertainty in Physics-Informed Variational Autoencoders for Anomaly Detection 1 Introduction 2 Stochasticity in Production Processes 2.1 Analytical Treatment of Fluctuations 2.2 Example and Its Perturbation 2.3 Classical Approach for Extracting Uncertainties 2.4 Limitations 3 Autoencoders and Uncertainty 3.1 Standard Autoencoder Concept 3.2 Variational Autoencoders for Dimensionality Reduction 3.3 Variational Autoencoder with Physics-Informed Input Modification 4 Interpretation of Resulting Distributions 4.1 Clustering in the Autoencoder Latent Space 4.2 Probability Distribution in the Latent Space 5 Conclusion and Outlook References Mapping of Standardized State Machines to Utilize Machine Learning Models in Process Control Environments Abstract 1 Introduction 2 Utilization of Machine Learning in Control Applications 2.1 Control Applications in the Context of Industry 4.0 2.2 Reinforcement Learning for Plant Control 2.3 Component Based Process Control 3 Scenario - Learning to Control a Palette Transport System of a Cold Rolling Mill Plant 3.1 Migration from the Real Plant to the Gym 3.2 The Prototypical Gym 3.3 ML Operation Mode Enhancement 4 Interaction Scheme of the Machine Learning Model and Process Control Components 4.1 ML Model Interactions 4.2 Mapping to Unified State Machines 5 Summary and Outlook Acknowledgement References Quality 4.0 - Transparent Product Quality Supervision in the Age of Industry 4.0 Abstract 1 Introduction 2 Problem Description 3 The Quality 4.0 Approach 4 Quality Data Generation Service (QGS) 4.1 Data Plausibility Protection 4.2 Plausibility Value Calculation 4.3 Outlier Detection 5 Quality Allocation Service (QAS) 6 Quality Exchange Service (QXS) 7 Implementation 7.1 The FADI Framework 7.2 A Sample Implementation of FADI 7.3 The CI/CD Pipeline of FADI 8 Conclusion References AI and ML Techniques for Generation and Assessment of Products Properties Data Abstract 1 Introduction 2 Artificial Intelligence Techniques for Quality Data 3 Case Studies 3.1 Billet Crack Risk Assessment Through Final Hydrogen Content Estimation 3.2 Flatness Defects Detection Through DNN Based Image Analysis 3.3 Jominy Profile Predictor with Punctual Reliability Assessment 4 Conclusions References The Use of Advanced Data Analytics to Monitor Process-Induced Changes to the Microstructure and Mechanical Properties in Flat Steel Strip Abstract 1 Introduction 2 Data Preparation and Pre-processing 3 Selection of Input Variables 3.1 Process Knowledge 3.2 Genetic Algorithm 3.3 Decision Trees 3.4 Variables Redundancy Filter 4 Prediction Models and Results 5 Discussion and Conclusions Acknowledgements References Unsupervised Deep Learning for Detection of Non-uniform Surface Defect Distributions in Flat Steel Production Abstract 1 Introduction 2 Characterization of Surface Defect Distributions 3 Deep Learning Clustering Model 3.1 Deep Embedding Clustering (DEC) 4 Clustering of an Artificial Defect Distribution Dataset 4.1 Generation of the Artificial Distribution Dataset 4.2 Network Characteristics 4.3 Clustering Results 5 Clustering of a Real Defect Distribution Dataset 5.1 Generation of the Real Distribution Dataset 5.2 Clustering Results 6 Conclusions References Machine Learning-Based Models for Supporting Optimal Exploitation of Process Off-Gases in Integrated Steelworks Abstract 1 Introduction 2 Overview of the GASNET Project and Software 3 Materials and Methods 3.1 Modelling Methodologies 3.2 Optimization Strategies 4 Description of Developed Model Library 4.1 Off-Gases Production Models 4.2 Model of BOF Steam Production 4.3 Model of Electrical Power Production by BFG Expansion Turbine 4.4 Model for BFG Demand by Hot Blast Stoves 4.5 Model for Steam Injection in BF Cold Blast Wind 4.6 Model for Electricity Demand by BF Fans 4.7 Model for Energy Demands by RH Vacuum Degassing 5 Results and Discussion 5.1 Prediction of Gas Production and Demand 5.2 Prediction of Steam and Electricity Production and Demand 5.3 Optimization Results 6 Conclusions Acknowledgments References Industrial Cyber Security at the Network Edge: The BRAINE Project Approach Abstract 1 Introduction 2 Use Cases 2.1 Securing Edge-Based Solutions for the Aluminium Industry 2.2 Securing Digital Twin for Industry 4.0 at the Edge 3 BRAINE’s Approach 3.1 Overall Architecture 4 BRAINE Cyber Security Approach 4.1 Security Threat Model 4.2 Proactive/Reactive Cyber Security Methodologies Based on AI 5 Specific BRAINE Solutions for Industrial Cyber Security at the Edge 5.1 HW Acceleration for AI and Secure Communications 5.2 Programmable Networking for Industrial Cyber Security at the Edge 6 Concluding Remarks Acknowledgement References Smart Steel Pipe Production Plant via Cognitive Digital Twins: A Case Study on Digitalization of Spiral Welded Pipe Machinery Abstract 1 Introduction 2 Background 2.1 Digital Twins and Cognitive Digital Twins 2.2 Digital Twins in the Steel Industry 3 The CogniTwin Project 4 The NOKSEL Case 4.1 Objectives 4.2 Success Criteria 4.3 Work Accomplished 4.4 Future Work 5 Conclusion Acknowledgement References TSorage: A Modern and Resilient Platform for Time Series Management at Scale Abstract 1 Introduction 2 Existing Work 3 TSorage 4 Architecture 5 Case Study 6 Conclusion and Future Work Acknowledgement References Author Index This book collects perceptions and needs expectations and experiences concerning the application of Artificial Intelligence (AI) and Machine Learning in the steel sector. It contains a selection of themes discussed within the Workshop entitled Impact and Opportunities of Artificial Intelligence in the Steel Industry Borganized by the European Steel Technology Platform as an online event from October 15 until November 5, 2020. The event aimed at analyzing the diffusion of AI technologies in steelworks and at providing indications for future research, development and innovation actions addressing the sector demands. The chapters treat general analyses on transversal themes and applications for process optimization, product quality enhancement, yield increase, optimal exploitation of resources and smart data handling. The book is devoted to researchers and technicians in the steel or AI fields as well as for managers and policymakers exploring the opportunities provided by AI in industry