Asset selection and optimisation for robotic assembly cell reconfiguration

Mo, Fan (2024) Asset selection and optimisation for robotic assembly cell reconfiguration. PhD thesis, University of Nottingham.

[thumbnail of PhD_Thesis_Final_Version_Fan.pdf]
Preview
PDF (Thesis - as examined) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Available under Licence Creative Commons Attribution.
Download (61MB) | Preview

Abstract

With the development of Industry 4.0, the manufacturing industry has revolutionized a lot. Product manufacture becomes more and more customized. This trend is achieved by innovative techniques, such as the reconfigurable manufacturing system. This system is designed at the outset for rapid change in its structure, as well as in software and hardware components, to respond to market changes quickly. Robots are important in these systems because they provide the agility and precision required to adapt rapidly to new manufacturing processes and customization demands.

Despite the importance of applying robots in these systems, there might be some challenges. For example, there is data from multiple sources, such as the technical manual sensor data. Besides, robot applications must react quickly to the ever-changing process requirements to meet customer's requirements. Furthermore, further optimization, especially layout optimization, is needed to ensure production efficiency after adaptation to the current process requirements.

To address these challenges, this doctoral thesis presents a framework for reconfiguring robotic assembly cells in manufacturing. This framework consists of three parts: the experience databank, the methodology for optimal manufacturing asset selection, and the methodology for layout optimization.

The experience databank is introduced to confront the challenge of assimilating and processing heterogeneous data from numerous manufacturing sources, which is achieved by proposing a vendor-neutral ontology model. This model is specifically designed for encapsulating information about robotic assembly cells and is subsequently applied to a knowledge graph. The resulting knowledge graph, constituting the experience databank, facilitates the effective organization and interpretation of the diverse data.

An optimal manufacturing asset selection methodology is introduced to adapt to shifting processes and product requirements, which focuses on identifying potential assets and their subsequent evaluation. This approach integrates a modular evaluation framework that considers multiple criteria such as cost, energy consumption, and robot maneuverability, ensuring the selection process remains robust in changing market demands and product requirements.

A scalable methodology for layout optimization within the reconfigurable robotic assembly cells is proposed to resolve the need for further optimization post-adaption. It introduces a scalable, multi-decision modular optimization framework that synergizes a simulation environment, optimization environment, and robust optimization algorithms. This strategy utilizes the insights garnered from the experience databank to facilitate informed decision-making, thereby enabling the robotic assembly cells to not only meet the immediate production exigencies but also align with the manufacturing landscape's evolving dynamics.

The validation of the three methodologies presented in this doctoral thesis encompasses both software development and practical application through three distinct use cases. For the experience databank, an interface was developed using Protégé, Neo4j, and Py2neo, allowing for effective organization and processing of varied manufacturing data. The programming interface for the asset selection methodology was built using Python, integrating with the experience databank via Py2neo and Neo4j to facilitate dynamic and informed decision-making in asset selection. In terms of software for the layout optimization framework, two different applications were developed to demonstrate the framework's scalability and adaptability. The first application, combining Python and C# programming with Siemens Tecnomatix Process Simulate, is geared towards optimizing layouts involving multiple machines. The second application utilizes Python programming alongside the RoboDK API and RoboDK software, tailored for layout optimization in scenarios involving a single robot.

Complementing these software developments, the methodologies were further validated through three use cases, each addressing a unique aspect of the framework. Use Case 1 focused on implementing asset selection and system layout optimization based on a single objective, leveraging the experience databank. The required assets are selected, and the required cycle time for executing the whole robotic assembly operation has been reduced by 15.6% from 47.17 seconds to 39.83 seconds. Use Case 2 extended the layout optimization to single-robot operations with an emphasis on multi-criteria decision-making. The energy consumption was minimized to 5613.59 Wh after implementing optimization strategies, demonstrating a significant enhancement in energy efficiency. Compared with the baseline of 6164.98 Wh, this represents an 8.9% reduction in energy usage. For minimized cycle time, a reduction of 6.0% from the baseline of 57.11 seconds is achieved, resulting in a cycle time of 53.15 seconds. Regarding the pursuit of a maximized robot maneuverability index, an increase of 140.8% from the baseline of 0.4891235 is achieved, resulting in a maximized value of 1.1786125. Lastly, Use Case 3 tested the modular and multi-objective asset selection methodology, demonstrating its efficacy across diverse operational scenarios. Evaluations conducted with two multi-objective optimization algorithms, Non-Dominated Sorting Genetic Algorithm II and Strength Pareto Evolutionary Algorithm II, revealed interesting implications for selecting and optimizing robotic assets in response to new customer requests. Specifically, Strength Pareto Evolutionary Algorithm II identified a Pareto solution that was more cost-effective (£20,920) compared to Non-Dominated Sorting Genetic Algorithm II (£21,090), while maintaining a competitive specification efficiency score (0.865 vs. 0.879). Consequently, Strength Pareto Evolutionary Algorithm II is preferred for optimizing robotic asset selection in scenarios prioritizing cost. However, should the requirement shift towards maximizing specification efficiency, the Non-Dominated Sorting Genetic Algorithm II would be the more suitable choice.

These use cases not only showcased the practical applicability of the developed software but also underlined the robustness and adaptability of the proposed methodologies in real-world manufacturing environments.

In conclusion, this doctoral thesis presents a methodology for reconfiguring robotic assembly cells in manufacturing. By harnessing the capabilities of artificial intelligence, knowledge graphs, and simulation methodologies, it addresses the challenges of processing data from diverse sources, adapting to fluctuating market demands, and establishing further optimizations for enhanced operational efficiency in the modern manufacturing landscape. To affirm the viability of this framework, the thesis integrates software development procedures tailored to the proposed methodologies and furnishes evidence through three use cases, which are evaluated against well-defined criteria.

Item Type: Thesis (University of Nottingham only) (PhD)
Supervisors: Ratchev, Svetan
Chaplin, Jack C.
Sanderson, David
Popov, Atanas
Keywords: Robotic assembly cells; Experience databank; Optimal manufacturing asset selection; Layout optimisation; Robots in manufacturing
Subjects: T Technology > TS Manufactures
Faculties/Schools: UK Campuses > Faculty of Engineering > Department of Mechanical, Materials and Manufacturing Engineering
Item ID: 77263
Depositing User: Mo, Fan
Date Deposited: 12 Apr 2024 10:51
Last Modified: 12 Apr 2024 10:51
URI: https://eprints.nottingham.ac.uk/id/eprint/77263

Actions (Archive Staff Only)

Edit View Edit View