Material Identification of Particles in Space-Borne Electronic Equipments Based on Principal Component Analysis - Intelligent Computing for Sustainable Energy and Environment - page 464

directly linked to the physical parameters of the material. Moreover, ZCR is
defined as a measure of the number of times within a given period that the
amplitude of the acoustic signal passes through a value of zero [10]. ZCR is an
important parameter for acoustic signal classification. It is computed as follows:

n− 1

Z i = sign [ x i ( m− 1) ∗x i ( m )] . (1)

m =1

where n is the number of samples in the ith pulse.
The power spectrum of the signal reflects the signal frequency energy dis-
tribution. When the frequency components change, the power spectrum of the
centroid position will also change. In other words, the power spectrum disperses
or concentrates as the frequency components increases or decreases. Hence, the
power spectrum of the centroid and the spectral energy distribution can reflect
the frequency domain characterics. In this paper, the spectrum centroid ( FC ),
mean square frequency ( MSF ), spectrum variance ( VF ) and peak frequency are
chosen as the frequency domain features. They are defined as follows:

N 2

f i · PSD i

i = N 1

FC = . (2)

N 2

PSD i

i = N 1

N 2

f i ·

PSD i

i = N 1

MSF = . (3)

N 2

PSD i

i = N 1

N 2

( f i −FC ) 2 ·PSD i

i = N 1

FC 2 .

VF = = MSF − (4)

N 2

PSD i

i = N 1

f i is the magnitude of the ith frequency, PSD i is the power spectral amplitude
of the ith frequency.
All the aforementioned features in both time and frequency domain were
used as an initial variable pool, and then the PCA was used to further select
representative features from this pool. The single and accumulative contribution
rate of principal components obtained by PCA is shown in Fig. 3. In this study,
the R N threshold of 0.75 leads to the number of principal components N =2for
the clustering analysis.
By comparison with wire, aluminum (Al) and tin particles as in the Fig. 4, it
is feasible to distinguish the different material particles using PCA. For the wire
particles, material hardness is lower than that of tin and aluminum. However, it
is dicult to classify tin and aluminum due to the overlapping patterns.