Multi-service collaboration and composition of cloud manufacturing customized production based on problem decomposition

Hao Yue, Yingtao Wu, Min Wang, Hesuan Hu, Weimin Wu, Jihui ZhangarXiv 2024 · DOI · arXiv

Introduction

Cloud Manufacturing (CMfg) represents a paradigm shift in production, transforming manufacturing resources and capabilities into on-demand services. A critical challenge within this ecosystem is the Service Composition Optimal Selection (SCOS)—the process of identifying the most efficient combination of virtualized resources to fulfill bespoke customer orders.

Traditional SCOS models typically rely on a one-to-one mapping, where each production sub-task is assigned to a single service provider. However, this approach often fails in complex, customized production environments where a single provider may lack the capacity or speed to meet high-volume demands. This paper addresses this gap by introducing a framework for multi-service collaboration, allowing multiple providers to work simultaneously on a single sub-task to eliminate bottlenecks.

A Three-Layered Framework for Cloud Manufacturing

To facilitate multi-service collaboration, the authors propose a robust three-layered architecture designed to bridge the gap between high-level requirements and low-level resource execution.

The three-layered framework for service composition and optimization.
  1. The First Layer (User Layer): Captures customization requirements and generates specific production orders.
  2. The Middle Layer (Cloud Service Center): Acts as the intelligence hub, performing task decomposition, service matching, and optimization of service composition.
  3. The Bottom Layer (Service Application): Manages the physical execution of services using machines, robots, workers, and edge computing devices.

Optimization via Problem Decomposition based Genetic Algorithm (PDGA)

The introduction of multi-service collaboration exponentially increases the "search space" for optimal solutions, making standard optimization tools inefficient. To solve this, the researchers developed the Problem Decomposition based Genetic Algorithm (PDGA).

The PDGA employs three core strategies:

  • Decomposition: Instead of optimizing the entire production chain as a single entity, the algorithm breaks the problem down. Each sub-task maintains its own "population" of potential service combinations, allowing for more granular search.
  • Stratified Sequencing: The algorithm prioritizes objectives hierarchically: first minimizing completion time, followed by production cost, and finally the number of services utilized.
  • Search Bootstrap Function: This specialized mechanism prevents the algorithm from prematurely converging on solutions that optimize time at the expense of cost, ensuring a diverse set of Pareto-optimal (non-dominated) solutions.

Case Study: Smart Clothing Customization

The framework's efficacy was tested using a simulated Smart Clothing Customization scenario. This case study involves a production chain of six distinct sub-tasks, ranging from initial data collection to final manufacturing.

Workflow of the smart clothing customization case study.

The sub-tasks include:

  1. Data collection (ST1): Smart body measurement and demand collection.
  2. Design and Processing (ST2-ST4): Clothing design, loose quantity processing, and CAD drawing.
  3. Layout (ST5): Optimizing material usage.
  4. Manufacturing (ST6): Tailoring, sewing, ironing, inspection, and packaging.

Key Results and Performance

The PDGA was benchmarked against NSGA-II, a widely recognized multi-objective genetic algorithm. The results demonstrated significant improvements across several metrics:

  • Computational Efficiency: PDGA converged faster than NSGA-II, finding superior solutions in approximately 7,900–8,000ms compared to over 8,700ms for the benchmark.
  • Resource Optimization: Solutions generated by PDGA were more streamlined, utilizing an average of 7–8 services per order, whereas NSGA-II averaged 12.4. Reducing the number of services minimizes administrative overhead and potential points of failure.
  • Solution Quality: PDGA consistently produced a "denser" Pareto front, offering decision-makers a wider variety of high-quality options that balanced time and cost more effectively than traditional methods.

Conclusion and Future Work

By enabling multi-service collaboration and utilizing a decomposition-based optimization approach, this research provides a scalable solution for the complexities of modern customized manufacturing. While the current model is robust, future work will focus on addressing stochasticity—accounting for real-world service delays or outages—and expanding the framework to handle multi-order complexity, where multiple custom orders compete for the same resource pool simultaneously.

Summary updated

BibTeX
@misc{yue2024multi,
  author = {Hao Yue and Yingtao Wu and Min Wang and Hesuan Hu and Weimin Wu and Jihui Zhang},
  title = {{Multi-service collaboration and composition of cloud manufacturing customized production based on problem decomposition}},
  year = {2024},
  eprint = {2406.19608},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2406.19608},
}
Figure