Machine learning based plant identification of a nonlinear two-degree-of-freedom system for active vibration control
Machine learning based plant identification of a nonlinear two-degree-of-freedom system for active vibration control
Active control solutions may be preferable to passive solutions when size or weight become a constraint in the design phase. Historically, active strategies have commonly been applied using linear controllers. However, when the response from the noise source to the error sensor becomes nonlinear, controller performance can be limited. It has previously been shown that replacing linear controllers with Neural Networks (NNs) can improve performance in such cases. Furthermore, more complex networks have been shown to improve performance further. In this paper, a model of a simple vibration control system is studied to assess the performance and behaviour of linear models against Multi-Layer Perceptrons (MLPs), Recurrent Neural Networks (RNNs) and Long Short Term Memory (LSTM) networks implemented as plant models. The system is simulated with a fixed nonlinear cubic stiffness, with the magnitude of the input plant identification noise varied. The performance of the plant models is discussed in the time and frequency domains.
Pike, Alexander
1cd3f629-7971-4b9c-9b4a-636df608bbe0
Cheer, Jordan
8e452f50-4c7d-4d4e-913a-34015e99b9dc
12 September 2022
Pike, Alexander
1cd3f629-7971-4b9c-9b4a-636df608bbe0
Cheer, Jordan
8e452f50-4c7d-4d4e-913a-34015e99b9dc
Pike, Alexander and Cheer, Jordan
(2022)
Machine learning based plant identification of a nonlinear two-degree-of-freedom system for active vibration control.
In Proceedings of ISMA 2022.
Record type:
Conference or Workshop Item
(Paper)
Abstract
Active control solutions may be preferable to passive solutions when size or weight become a constraint in the design phase. Historically, active strategies have commonly been applied using linear controllers. However, when the response from the noise source to the error sensor becomes nonlinear, controller performance can be limited. It has previously been shown that replacing linear controllers with Neural Networks (NNs) can improve performance in such cases. Furthermore, more complex networks have been shown to improve performance further. In this paper, a model of a simple vibration control system is studied to assess the performance and behaviour of linear models against Multi-Layer Perceptrons (MLPs), Recurrent Neural Networks (RNNs) and Long Short Term Memory (LSTM) networks implemented as plant models. The system is simulated with a fixed nonlinear cubic stiffness, with the magnitude of the input plant identification noise varied. The performance of the plant models is discussed in the time and frequency domains.
Text
IN21_Latex_Paper-1
- Accepted Manuscript
Restricted to Repository staff only
Request a copy
More information
Published date: 12 September 2022
Venue - Dates:
International Conference on Noise and Vibration Engineering 2022, KU Leuven Social Sciences Building, Leuven, Belgium, 2022-09-12 - 2022-09-14
Identifiers
Local EPrints ID: 470921
URI: http://eprints.soton.ac.uk/id/eprint/470921
PURE UUID: 71fa80f6-5afb-4f48-8966-77a3deba4b88
Catalogue record
Date deposited: 21 Oct 2022 16:31
Last modified: 17 Mar 2024 03:23
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
