Article: Collaborative resilience: taxonomy-informed neural networks for smart assets' maintenance in hostile Industry 4.0 environments Journal: International Journal of Mechatronics and Manufacturing Systems (IJMMS) 2024 Vol.17 No.2 pp.180 - 200 Abstract: This paper explores knowledge-informed machine learning and particularly taxonomy-informed neural networks (TINN) to enhance data-driven smart assets' maintenance by contextual knowledge. Focusing on assets within the same class that may exhibit subtle variations, we introduce a weighted Lehmer mean as a dynamic mechanism for knowledge integration. The method considers semantic distances between the asset-in-question and others in the class, enabling adaptive weighting based on behavioural characteristics. This preserves the specificity of individual models, accommodating heterogeneity arising from manufacturing and operational factors. In the adversarial learning context, the suggested method ensures robustness and resilience against adversarial influences. Our approach assumes a kind of federated learning from neighbouring assets while maintaining a detailed understanding of individual asset behaviours within a class. By combining smart assets with digital twins, federated learning, and adversarial knowledge-informed machine learning, this study underscores the importance of collaborative intelligence for efficient and adaptive maintenance strategies in Industry 4.0 and beyond. Inderscience Publishers - linking academia, business and industry through research

Title: Collaborative resilience: taxonomy-informed neural networks for smart assets' maintenance in hostile Industry 4.0 environments

Authors: Vagan Terziyan; Oleksandra Vitko

Addresses: Faculty of Information Technology, University of Jyväskylä, 40014, Jyväskylä, Finland ' Department of Artificial Intelligence, Kharkiv National University of Radio Electronics, 61166, Kharkiv, Ukraine

Abstract: This paper explores knowledge-informed machine learning and particularly taxonomy-informed neural networks (TINN) to enhance data-driven smart assets' maintenance by contextual knowledge. Focusing on assets within the same class that may exhibit subtle variations, we introduce a weighted Lehmer mean as a dynamic mechanism for knowledge integration. The method considers semantic distances between the asset-in-question and others in the class, enabling adaptive weighting based on behavioural characteristics. This preserves the specificity of individual models, accommodating heterogeneity arising from manufacturing and operational factors. In the adversarial learning context, the suggested method ensures robustness and resilience against adversarial influences. Our approach assumes a kind of federated learning from neighbouring assets while maintaining a detailed understanding of individual asset behaviours within a class. By combining smart assets with digital twins, federated learning, and adversarial knowledge-informed machine learning, this study underscores the importance of collaborative intelligence for efficient and adaptive maintenance strategies in Industry 4.0 and beyond.

Keywords: knowledge-informed machine learning; taxonomy; digital twin; maintenance; federated learning; adversarial learning; robustness; resilience.

DOI: 10.1504/IJMMS.2024.143008

International Journal of Mechatronics and Manufacturing Systems, 2024 Vol.17 No.2, pp.180 - 200

Received: 27 Jan 2024
Accepted: 28 Mar 2024
Published online: 02 Dec 2024 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article

Title: Collaborative resilience: taxonomy-informed neural networks for smart assets' maintenance in hostile Industry 4.0 environments

Keep up-to-date