Abstract
This thesis explores the concept of service autonomy in cloud manufacturing (CMfg) scheduling, addressing the challenges and opportunities it presents in optimizing operational goals. Service autonomy refers to the independent decision-making capabilities of manufacturing services (MSs) within a CMfg system.This literature review systematically examines the related core CMfg themes, providing a comprehensive understanding of its current research landscape. It synthesizes CMfg's characteristics and advantages, establishes the critical importance of task scheduling and service autonomy and their associated challenges, positions bilevel optimization as a key manifestation of autonomy for strategic coordination, details QoS prioritization and resource utilization as pivotal operational objectives, and discusses energy efficiency and dynamic environments. Ultimately, this exploration identifies three critical research gaps for this thesis.
The first study addresses the concept of service autonomy by proposing a bilevel optimization model for CMfg scheduling with parallel batch processing. The upper level aims to maximize the overall Quality of Service (QoS) of the CMfg system, while the lower level focuses on maximizing the resource utilization rate of each MS. The study introduces an Integer and Categorical Particle Swarm Optimization (ICPSO) algorithm enhanced by Deep Q-Networks (DQN) to solve the problem. Experiments demonstrate that the proposed ICPSO-DQN algorithm outperforms other variants on the testing instances, effectively balancing exploration and exploitation. The method includes perturbation mechanisms, a subgoal-oriented search strategy, and an ϵ reset rule to improve its exploration capabilities. The experimental results further demonstrate that the proposed subgoal-oriented search strategy effectively improves both the upper- and lower-level objectives, thereby benefiting multiple stakeholders in the CMfg system. Although the decision-making at the lower level in the bilevel optimization model is autonomous, this study, from an upper-level methodological perspective, achieves a dual enhancement that ensures all parties involved can derive value, creating a mutually advantageous outcome.
The second study introduces a novel energy-efficient scheduling model that incorporates inter-stage waiting duration optimization to reduce energy consumption. A service-autonomous framework is proposed to decompose the solution into two phases: job sequencing and assignment on the cloud side and temporal decision-making on the edge side. A Genetic Algorithm assisted by Dueling Double Deep Q-Networks (named as D3QGA) is used to solve the first phase, while dynamic programming addresses the second phase. The service-autonomous framework leverages the characteristics of service autonomy to enhance the overall efficiency of the optimization process. By decentralizing decision-making to the manufacturing services (MSs), the framework allows each service to optimize its own timing of operations while still contributing to the overall energy efficiency goals of the CMfg system. The results of numerical experiments demonstrate that the proposed model and solution methods effectively reduce ISW-related energy consumption with minimal trade-offs on other energy components. The service-autonomous framework yields solutions that are remarkably close to optimal, and the D3QGA algorithm shows significant advantages over the basic genetic algorithm on varied instance scales by dynamically balancing exploration and exploitation.
The third study addresses a dynamic CMfg scheduling problem with hybrid autonomous services, where MSs have varying levels of autonomy. The problem is modeled as a Markov Decision Process (MDP), and a Multi-Head Soft Actor-Critic (MHSAC) algorithm is proposed to solve it. The algorithm categorizes actions into assignment and insertion decisions, applying conditional gradient updates to enhance stability. Experiments demonstrate that MHSAC outperforms other state-of-the-art reinforcement learning algorithms in solving the proposed problem, particularly under high market demand. The study also explores the benefits of integrating into cloud platforms for both suppliers and customers, showing improved resource utilization and reduced order tardiness. This research provides valuable insights into managing service autonomy in dynamic CMfg environments to achieve operational efficiency and customer satisfaction.
In summary, this thesis provides valuable insights into leveraging service autonomy in CMfg scheduling, offering practical solutions for improving operational and energy efficiency. Service autonomy, while often presenting challenges to global optimization due to the independent decision-making capabilities of MSs, can be effectively harnessed to enhance overall system performance and achieve business goals. The proposed models and algorithms demonstrate significant improvements in balancing system-level objectives with individual service autonomy, highlighting the potential value of managing and embracing service autonomy.
| Date of Award | 15 Oct 2025 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Zhen Tan (Supervisor), Lei Zhang (Supervisor) & Chandra Irawan (Supervisor) |