LOCAL RATIONAL MODELLING - CAN BOOTSTRAPPED TOTAL LEAST SQUARES IMPROVE THE FRF ESTIMATE?

Description

Mechanical systems predominantly exhibit strong resonant behaviour which must be accurately characterised to prevent potential catastrophies in real world applications. Measuring the Frequency Response Function (FRF) of these systems allows to effectively assess the severity of the resonant behaviour but remains challenging due to long transient phenomena and spectral leakage. Advanced local modelling techniques such as the Local Polynomial Method (LPM) and the Local Rational Method (LRM) were introduced in the past to remedy these challenges but they do not use an appropriate model structure or introduce a bias in the FRF estimate. We resolved these issues by developing a local rational modelling technique which removes the bias on the frequency response function measurement. This technique involves the use of the Bootstrapped Total Least Squares (BTLS) estimator which achieves nearly Maximum Likelihood (ML) properties by iteratively updating its estimate. Additionally, we applied a model order selection technique which limits the occurence of pole-zero cancellations in the local model as much as possible. The performance of the proposed technique was verified by measuring the flexural vibrations of a steel beam.

Period	24 Sep 2017 → 27 Sep 2017
Event title	2017 ERNSI Workshop on System Identification
Event type	Workshop
Location	Lyon, France
Degree of Recognition	International