ӣ��Ӱ��

Topic

A Framework for Autonomic Digital Twin Orchestration and Management Systems

Department of Computer Science

Date & location

• Tuesday, November 18, 2025

• 9:30 A.M.

• Engineering Computer Science Building

• Room 467 and Virtual Defence

Reviewers

Supervisory Committee

• Dr. Hausi Müller, Department of Computer Science, ӣ��Ӱ�� (Co-Supervisor)

• Dr. Norha M. Villegas, Department of Computer Science, UVic (Co-Supervisor)

• Dr. Alex Thomo, Department of Computer Science, UVic (Member)

• Dr. Xiaodai Dong, Department of Electrical and Computer Engineering, UVic (Outside Member) 

External Examiner

• Dr. Ying (Jenny) Zou, Department of Electrical and Computer Engineering, Queen’s University 

Chair of Oral Examination

• Dr. Nikki Macdonald, School of Public Administration, UVic

   

Abstract

The advancement and implementation of the Digital Twin (DT) concept are poised to disrupt multiple application domains across industry and society. DTs enable the augmentation of machine, system, and human capabilities by enriching data-driven decision making and forecasting through the continuous aggregation, interpretation, and exploitation of relevant phenomena from mirrored counterparts—i.e., Real Twins (RTs). The accelerated adoption of DT technologies—from smart urban infrastructures to complex IT environments—has been propelled by the synergistic convergence of innovations in the Internet of Things (IoT), discriminative (traditional) Artificial  Intelligence (AI), Generative AI (GenAI), simulation technologies, and Cloud Computing. Central to the DT vision is the notion of Digital Twin Operation & Management Systems (DTOMSs), software-intensive infrastructure responsible for realizing the potential of DTs and preserving sustained fidelity between RTs and their virtual representations. However, the inherent dynamism and unpredictability of RT environments pose significant challenges to the relatively static nature of contemporary DTOMS architectures. These systems often fall short in reflecting evolving RT contexts, anticipating behavioural drift, and adapting to runtime uncertainties— capabilities that are essential to unlock the full potential of DT-based systems.

This dissertation addresses the limitations of conventional DTOMSs by advocating a transition toward autonomic DTOMSs (i.e., ADTOMSs). Grounded in the principles of self-adaptive systems, autonomic computing, and continuous software engineering, we investigate how ADTOMSs can dynamically represent, reason about, and evolve in response to contextual and continuous variations in mirrored RTs and their operational environments. We propose foundational architectural constructs, runtime DT modelling mechanisms, GenAI-based knowledge exploitation techniques, and model evolution strategies. Collectively, these contributions constitute a framework that advances the engineering of DTOMSs toward autonomic systems with improved complexity and uncertainty management.

The contributions of this research are fourfold. First, we propose a reference model and accompanying reference architecture that delineate the core design elements of ADTOMSs, incorporating self-management capabilities aimed at mitigating system complexity and reducing human intervention and cognitive load. Second, we introduce an adaptive model evolution mechanism that enables ADTOMSs to incrementally refine internal representations in response to the evolving dynamics of their corresponding RTs. Third, we develop a dynamic DT context modelling and knowledge representation framework to support continuous monitoring and adaptive reasoning over conditions captured or simulated from mirrored RTs. Fourth, we design an automated reasoning framework, leveraging Continuous Experimentation (CExp) and GenAI, to extract actionable insights from heterogeneous data sources, which facilitates early anomaly detection and behavioural forecasting. 

Methodologically, this research adopts an exploratory sequential mixed-methods approach, integrating conceptual modelling, systematic literature analysis, and empirical validation through case-driven experimentation. The proposed contributions are evaluated within the domains of smart urban transit and IT environments, demonstrating their feasibility, adaptability, and practical relevance across heterogeneous operational contexts. 

Building upon the contributions of this dissertation, several promising research directions emerge that can inspire further academic exploration and practical innovation. First, our modelling infrastructure provide the foundations for context management in GenAI settings. This paradigm leverages our modelling approach to treat prompts, retrieval corpora, model snapshots, inference runs, and their artifacts as first-class elements, enabling reproducible, auditable, and drift-aware reasoning within DT-based systems. Second, our reference model and reference architecture create interesting opportunities for the realization of a systematic, policy-governed mapping from control objectives and observed symptoms to well-scoped reasoning tasks, making tacit operational cues explicit and ensuring that queries to the reasoning layer remain aligned with goals and context. Third, the CExp practices incorporated in our conceptualization of ADTOMSs provide a sound basis for a hybrid quantum orchestration twin that operationalizes the use of quantum and classical resources as a managed control objective, using controlled experiments and provenance aware scheduling to decide when and how to employ each under fidelity, latency, and cost constraints. Together, these directions extend our contributions toward trustworthy, explainable, and efficient autonomy in advanced DT-based systems. 

In summary, this dissertation advances the conceptual and technological foundation of DTOMSs by introducing autonomic principles into their operational lifecycle. The resulting ADTOMS paradigm establishes a robust basis for the continuous, autonomic evolution of DTs, positioning them as resilient, adaptive, and long-lived software-intensive systems capable of operating effectively under uncertainty. This work contributes to the broader vision of self-managing data-intensive systems and offers novel engineering strategies for advancing DT practices across dynamic application domains.