PSI - Issue 17

Pavel Steinbauer et al. / Procedia Structural Integrity 17 (2019) 799–805 Author name / Structural Integrity Procedia 00 (2019) 000 – 000

803

5

5. Evaluation algorithm

The pole health state is evaluated using several methods, which are finally aggregated. The first method simply checks, whether static values of the acceleration vector have changed from initialization. The pole tilt caused by severe damage, ground distortion, traffic accident or similar is safely detected. The acceleration is measured in two axis into vectors y 1 and y 2 . = | asin( ) − asin([ ( 1 ); ( 2 )]) | (1) Frequency analyses are performed only in case of sufficient motions, remarkable excitation of the pole max( ( 1 ) − ( 1 ) , ( 2 ) − ( 2 )) < ℎ (2) The eigenfrequency shift detection is based on peak detection and correlation analysis. The time domain data are processed into frequency domain using fast fourier transform algorithm

and treated by coarse graining to remove random outliers

t psd f

( ) ( ) y f

( y f f f Hz =  −  +   : f f ) [ ]

(3)

The spectra is then averaged from several measurements to suppress measurement noise.

pro 2 k =

: N Tag

Tag

Tag

1,

0,

0



=

=

=

long

short

test

PSD

1 k

(

)

1 k k = +

,

(4)

PSD PSD =

psd

=

+

T

T

T

t

k

1

−

The peak detection is based search algorithm detecting significant peaks in the PSD spectra. The 50Hz region is filtered from the averaged spectra PSD T at first to remove electromagnetic interference. Then the averaged spectra PSD T is iteratively clipped from the maximum to minimum. In each iteration, newly detected frequencies are added into peak set for further analyses.

For PSD i = max(PSD T ) : min(PSD T ), f peak =find(PSD T >PSD i ); PEAK=[PEAK, f peak ]; If length(PEAK)>=N peakmax, Exit; end;

(5)

where N peakmax is selected maximum of evaluated peaks. Using correlation analysis, in analogy to cross entropy concept (e.g. RAMOS, Daniel, et al. (2018) ) , the probability of random quantity is equivalent to statistical frequency of excitated eigen frequencies ( ) ( ) b i T i p x PSD f  ( ) ( ) b i a i q x psd f  (6) Where 1 b i N = K a b N is number of frequency intervals and ( ) b T i PSD f a ( ) b a i psd f are statistical frequencies of excitated eigen frequencies in the i-th frequency interval, obtained by accumulation of values binarized