Structural Health Monitoring Using Smart Sensors - ideals ...
Structural Health Monitoring Using Smart Sensors - ideals ...
Structural Health Monitoring Using Smart Sensors - ideals ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
10 0<br />
10 -2<br />
10 -4<br />
0 500 1000 1500 2000<br />
10 -12<br />
10 -14<br />
10 -16<br />
0 500 1000 1500 2000<br />
discrete sample point<br />
Figure 7.1. FFT accuracy check (Number of FFT data points: 4,096).<br />
Fourier amplitude |F|<br />
Fourier amplitude<br />
error ratio |F-F0|/|F0|<br />
give numerically identical FFT results, considering that double precision data has about 15<br />
effective significant digits.<br />
The FFT on double precision data needs n B memory space to store the data to be<br />
analyzed; the data is then replaced with the FFT result. The Imote2’s 256 kB of on-board<br />
memory can hold about 32,000 double data points. If FFT is applied to a longer record,<br />
larger RAM or virtual memory is necessary. Note that the Imote2 has 32 MB of RAM,<br />
which can hold a large amount of data. At the time of this research, however, only 256 kB<br />
of memory is accessible.<br />
7.1.2 Singular value decomposition<br />
Singular Value Decomposition (SVD) is a part of both the ERA and DLV methods.<br />
While SVD in the ERA and mass perturbation DLV method are applied to real valued<br />
matrices, the SDLV method may involve SVD of a complex matrix, depending on how<br />
the observation matrix, C, is reconstructed through modal analysis. A double precision<br />
SVD function for real matrices, as well as one for complex matrices needs to be available.<br />
The source code for SVD of a real matrix is found in Numerical Recipes in C and<br />
CLAPACK, while CLAPACK also provides source codes for SVD of a complex matrix.<br />
First, the matrix is transformed to a bi-diagonal form using the Householder reduction.<br />
The bidiagonal matrix is then diagonalized to obtain the singular values. The source code<br />
from Numerical Recipes in C and CLAPACK are modified for implementation on the<br />
Imote2.<br />
The accuracy of the implementation is examined. The singular values of the matrix<br />
are chosen as accuracy indicators; singular values of various magnitudes are inspected. An<br />
arbitrary matrix is first constructed and decomposed by SVD on MATLAB. The singular<br />
values are then replaced with a decreasing geometric series; singular values of this matrix<br />
104