The use of multi-fidelity simulation optimisation for real-time management of a manufacturing line
The use of multi-fidelity simulation optimisation for real-time management of a manufacturing line
Manufacturers often rely on simulation models to support decision-making for a wide range of tasks from system design to routine operations. There is scope for simulation modelling to extend further to deal with situations that look for fast responses, ideally in real time. Simulation models for manufacturing systems are usually complex and large, meaning they take a long time to run. They are not able to support real-time decision-making when working with conventional simulation optimisation procedures which require multiple replications for producing the recommendations. A multi-fidelity simulation optimisation framework is designed to address this problem. It includes a low-fidelity metamodel to guide the search using the high-fidelity simulation model, which aims to reduce the replications of the high-fidelity model needed in the optimisation. The proposed method has been tested on a well-known simulation model of an inventory system and a new model of a production line. The results from both the textbook example and the manufacturing system show that it satisfies the need for real-time optimisation and performs well. In the production line case, we aim to optimise the repair policy when multiple machines break down simultaneously. The order to repair the failed machines could affect how the system would recover and would lead to different short-term system throughput. By optimising the repair policy, the system throughput is optimised. In order to carry out the optimisation in real time, a ''hot-start'' simulation model is structured for the production line system. The state space of the metamodel for the production line is modified and reduced in order to fit the metamodel with fewer data. An adaptive sequential sampling method is developed to efficiently sample from the stochastic simulation model.
University of Southampton
Cao, Yiyun
cfc96d2e-fd5c-44cf-95b1-71537a56964c
2024
Cao, Yiyun
cfc96d2e-fd5c-44cf-95b1-71537a56964c
Currie, Christine
dcfd0972-1b42-4fac-8a67-0258cfdeb55a
Onggo, Stephan
8e9a2ea5-140a-44c0-9c17-e9cf93662f80
Cao, Yiyun
(2024)
The use of multi-fidelity simulation optimisation for real-time management of a manufacturing line.
University of Southampton, Doctoral Thesis, 131pp.
Record type:
Thesis
(Doctoral)
Abstract
Manufacturers often rely on simulation models to support decision-making for a wide range of tasks from system design to routine operations. There is scope for simulation modelling to extend further to deal with situations that look for fast responses, ideally in real time. Simulation models for manufacturing systems are usually complex and large, meaning they take a long time to run. They are not able to support real-time decision-making when working with conventional simulation optimisation procedures which require multiple replications for producing the recommendations. A multi-fidelity simulation optimisation framework is designed to address this problem. It includes a low-fidelity metamodel to guide the search using the high-fidelity simulation model, which aims to reduce the replications of the high-fidelity model needed in the optimisation. The proposed method has been tested on a well-known simulation model of an inventory system and a new model of a production line. The results from both the textbook example and the manufacturing system show that it satisfies the need for real-time optimisation and performs well. In the production line case, we aim to optimise the repair policy when multiple machines break down simultaneously. The order to repair the failed machines could affect how the system would recover and would lead to different short-term system throughput. By optimising the repair policy, the system throughput is optimised. In order to carry out the optimisation in real time, a ''hot-start'' simulation model is structured for the production line system. The state space of the metamodel for the production line is modified and reduced in order to fit the metamodel with fewer data. An adaptive sequential sampling method is developed to efficiently sample from the stochastic simulation model.
Text
YiyunCao_Southampton_PhD_Thesis_20240119 (1)
- Version of Record
Text
Final-thesis-submission-Examination-Mr-Yiyun-Cao
Restricted to Repository staff only
More information
Published date: 2024
Identifiers
Local EPrints ID: 486566
URI: http://eprints.soton.ac.uk/id/eprint/486566
PURE UUID: cf2642f1-04d1-4975-9706-73f5bddc3ff3
Catalogue record
Date deposited: 26 Jan 2024 17:37
Last modified: 18 Mar 2024 03:55
Export record
Download statistics
Downloads from ePrints over the past year. Other digital versions may also be available to download e.g. from the publisher's website.
View more statistics