Use of Machine Learning to Improve Condition Monitoring and Vibration Analysis

Klaus Stohl, Primetals Technologies Austria GmbH; Bradley Kintner and Borui Li, ITR

Sophisticated condition monitoring and analysis is a cornerstone of any digitalization strategy. On-line monitoring of machine feature data is state of the art. For a full analysis of complex machinery (e.g., variable load, speed, product), however, a human expert is still needed to determine whether there is a problem, its severity, its root cause and, most importantly, what shall be done about it. More than 35 years of data allowed for the application of machine-learning (ML) techniques to extend what is possible automatically. This paper focuses on several examples of successful ML applications and the resulting benefits for steel producers.

Synthetic Images of Longitudinal Cracks in Steel Slabs Via Wasserstein Generative Adversarial Nets Used Toward Unsupervised Classification

Diego Andrade, ANT Automation LLC

In the training of neural networks (NN) for automatic classification of slab surface defects, the learning rate becomes onerous and slow, being necessary to apply qualified human resources with in-depth knowledge of steel quality. Images can be affected by sources: illumination, camera angles, quality, sensitivity and environmental changes, thus forcing the NN in the most unfavorable case to never finish learning. With the proposed framework using recent advances in machine learning, synthetic images are produced using a Wasserstein generative adversarial network to create sets of images for steel slab defects while validating these synthetic images using inception score methodology.

Artificial Intelligence Services in Steel Production - On Premises and in the Cloud

Sonja Strasser, Primetals Technologies Austria GmbH

Physics-based metallurgical models have been used for decades in the steel industry. Advanced methods of machine learning and data sciences allow combining the strength of pure data models and proven physics models and exploiting their benefits. Expert know-how in steel is being integrated with data science expertise to provide the next level of digital services for the metals industry. This paper focuses on practical examples how to use artificial intelligence in steel production to gain deeper insight, develop new control schemes, find root causes, improve deviation forecasting, and to optimize operations, quality results and predictive maintenance.

A Mini-Tutorial for Managers and Practitioners: Machine Learning for Process Improvement in Iron and Steel Manufacturing

Anil Gandhir, Qualicent Analytics Inc.

As a toolkit and a guide for executives, managers and practitioners, this paper describes why and how advanced analytics and machine learning (ML) are applied to help achieve significant breakthroughs for process improvement in steel plant operations, measured in terms of quantity and quality of output. The discussion will focus on practical aspects of implementing ML in production environments while also attending to business requirements of profitability and cost reduction. Specifically, ML methods will be described that allow model accuracies of ~90%.