Blind fault detection and source identification using higher order statistics for impacting systems
Blind fault detection and source identification using higher order statistics for impacting systems
Classical deconvolution methods for source identification can only be used if the transfer function of the system is known. For many practical situations, however, this information is not accessible and/or is time varying. The problem addressed is that of reconstruction of the original input from only the measured signal. This is known as 'blind deconvolution'. By using Higher Order Statistics (HOS), the restoration of the input signal is established through the maximisation of higher order moments (cumulants) with respect to the characteristics of the signals concerned.
This paper demonstrates the restoration of input signals that have a pulse-like form. From only the measured signal (an output of the unknown system), its normalised cumulant is constructed and employed to calculate the coefficients of the inverse filter through both a Wiener approach and global optimisation. This filter is then convolved with the measured signal to give the restored signal.
The inverse filter is determined iteratively and aspects affecting convergence and performance that are investigated include: The choice of the initial inverse filter and, order determination of the filter for both nonrecursive and recursive deconvolution operators. An experimental verification is carried out for the restoration of our impacting signal arising in the response of a cantilever beam with an end stop when randomly excited.
Techniques for the detection of non-Gaussian impacting signal from the observed signal through the higher order (>2) cumulant tensor (known as 'Higher Order Singular Value Decomposition, HOSVD') are introduced and discussed.
University of Southampton
Seo, Jong-Soo
7c03b10d-e317-4250-b1ce-1c69dedee1c3
2000
Seo, Jong-Soo
7c03b10d-e317-4250-b1ce-1c69dedee1c3
Seo, Jong-Soo
(2000)
Blind fault detection and source identification using higher order statistics for impacting systems.
University of Southampton, Doctoral Thesis.
Record type:
Thesis
(Doctoral)
Abstract
Classical deconvolution methods for source identification can only be used if the transfer function of the system is known. For many practical situations, however, this information is not accessible and/or is time varying. The problem addressed is that of reconstruction of the original input from only the measured signal. This is known as 'blind deconvolution'. By using Higher Order Statistics (HOS), the restoration of the input signal is established through the maximisation of higher order moments (cumulants) with respect to the characteristics of the signals concerned.
This paper demonstrates the restoration of input signals that have a pulse-like form. From only the measured signal (an output of the unknown system), its normalised cumulant is constructed and employed to calculate the coefficients of the inverse filter through both a Wiener approach and global optimisation. This filter is then convolved with the measured signal to give the restored signal.
The inverse filter is determined iteratively and aspects affecting convergence and performance that are investigated include: The choice of the initial inverse filter and, order determination of the filter for both nonrecursive and recursive deconvolution operators. An experimental verification is carried out for the restoration of our impacting signal arising in the response of a cantilever beam with an end stop when randomly excited.
Techniques for the detection of non-Gaussian impacting signal from the observed signal through the higher order (>2) cumulant tensor (known as 'Higher Order Singular Value Decomposition, HOSVD') are introduced and discussed.
Text
756263.pdf
- Version of Record
More information
Published date: 2000
Identifiers
Local EPrints ID: 464204
URI: http://eprints.soton.ac.uk/id/eprint/464204
PURE UUID: ed82888f-fd2a-4bb0-ba5c-8e245203b105
Catalogue record
Date deposited: 04 Jul 2022 21:33
Last modified: 16 Mar 2024 19:20
Export record
Contributors
Author:
Jong-Soo Seo
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