ii_poest

SDT-base Contents Functions

ii_poest

Purpose

Identification of a narrow-band single pole model.

Syntax

idcom('e')
[res,po]= ii_poest(ci.Stack{'Test'},opt)

Description

ii_poest (idcom e command and associated button in the idcom GUI figure, see section 2.9) provides local curve fitting capabilities to find initial estimates of poles by simply giving an indication of their frequency.

The central frequency for the local fit is given as opt(2) or, if opt(2)==0, by clicking on a plot whose abscissas are frequencies (typically FRF of MMIF plots generated by iiplot).

The width of the selected frequency band can be given in number of points (opt(1) larger than 1) or as a fraction of the central frequency (points selected are in the interval opt(2)*(1+[-opt(1) opt(1)]) for opt(1)<1). The default value is opt(1)=0.01.

A single pole fit of the FRFs in xf is determined using a polynomial fit followed by an optimization using a special version of the id_rc algorithm. The accuracy of the results can be judged graphically (when using the idcom e command, Test and IdFrf are automatically overlaid as shown in the plot above) and based on the message passed

>> ci=idcom;iicom(ci,'CurveLoad','gartid');
>> idcom('e .01 16.5');
>> disp(ci.Stack{'IdAlt'}.po)
  1.6427e+001  1.3108e-002
LinLS: 5.337e-001, LogLS 5.480e-001, nw 18
 mean(relE) 0.00, scatter 0.47 : acceptable
Found pole at 1.6427e+001   1.3108e-002
% manual call would be [res,po]=ii_poest(ci.Stack{'Test'},[.01 16.5]);

which indicates the linear and quadratic costs (see ii_cost) in the narrow frequency band used to find the pole, the number of points in the band, the mean relative error (norm of difference between test and model over norm of response, see iiplot error) which should be below 0.1, and the level of scatter (norm of real part over norm of residues, which should be small if the structure is close to having proportional damping).

If you have a good fit and the pole differs from poles already in your current model, you can add the estimated pole (add IdAlt to IdMain) using the idcom ea command.

The choice of the bandwidth can have a significant influence on the quality of the identification. As a rule the bandwidth of your narrow-band identification should be larger than the pole damping ratio (opt(1)=0.01 for a damping of 1% is usually efficient). If, given the frequency resolution and the damping of the considered pole, the default does not correspond to a frequency band close to 2ζ_jω_j, you should change the selected bandwidth (for example impose the use of a larger band with opt(1)=.02 which you can obtain simply using idcom ('e.02')).

This routine should be used to obtain an initial estimate of the pole set, but the quality of its results should not lead you to skip the pole tuning phase (idcom eup or eopt commands) which is essential particularly if you have closely spaced modes.