Open Collections

Chi, Yuanfang

University of British Columbia

The Industrial Internet of Things (IIoT) interconnects advanced sensors and devices to enhance decision-making and automate industrial operations, particularly in high-risk environments. This integration introduces vulnerabilities that can lead to catastrophic failures due to component malfunctions or cyber-physical attacks. Thus, robust fault diagnosis mechanisms are critical to maintain system reliability. Moreover, modern IIoT architectures often consist of interconnected subsystems managed by various stakeholders. A failure observed in one subsystem may have already been encountered in other subsystems and analyzed by different stakeholders. Sharing fault diagnosis knowledge, such as root causes and appropriate recovery actions, significantly enhances the overall efficiency and effectiveness of fault diagnosis in IIoT systems. This thesis introduces innovative frameworks and methodologies designed to facilitate collaborative fault diagnosis through effective knowledge sharing. It employs knowledge-based fault diagnosis where subsystems maintain their own fault knowledge in Knowledge Bases (KBs) represented by knowledge graphs. Specifically, this research first develops a distributed knowledge inference framework that employs a path-based reasoning algorithm. Data from different knowledge graphs is integrated into a neural network model, a.k.a, a reasoning model, using this advanced algorithm. This framework employs a centralized coordinator to facilitate collaborative training of a reasoning model among agents using their local KBs. This approach allows fault knowledge to be embedded in the reasoning model and subsequently shared. However, due to the diverse ownership of IIoT subsystems, the centralized coordinator of the distributed framework raises trustworthiness concerns. This thesis then introduces a decentralized knowledge inference mechanism that leverages blockchain technologies to eliminate the need for a central coordinator within the distributed framework. Moreover, this mechanism selects workers capable of providing optimal training results based on their self-evaluated capabilities with their local KBs. Finally, this thesis investigates an incentive mechanism to promote worker integrity within the decentralized knowledge inference mechanism, ensuring timely and accurate knowledge sharing for collaborative fault diagnosis. Extensive evaluations show that the proposed decentralized knowledge inference delivers a favorable reasoning model with higher overall accuracy and reduced training effort. Additionally, the thesis discusses how various incentive schemes influence participant interactions, providing insights into optimizing collaborative efforts in IIoT systems.

Text

2024-08-29

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/89082

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/