RTSHM-Kernel Main Screen
Two main functionalities are enabled
by the coremicro® RTSHM-kernel product:
This is a general software tool that allows characterization
of events associated with a structure's state. Features highly
correlated to structure conditions can be identified and used
for further processing by custom algorithms (user application
software). In particular an advantage of this product is that
it can be combined with Optimized Neuro Genetic Fast Estimator
(ONGFE), an ANN based software toolset, for conducting automated
diagnostics. In this case, data logs which contain vibration signals
that correspond to representative structure states can be used
(after applying the selected RTSHM-kernel's feature extraction
algorithms for training the ONGFE's diagnostic neural networks)
to perform real-time on-line diagnostics). The main screen of
the RTSHM-kernel (as shown in the figure) provides the following
main features:
- Data log feature extraction plotting and recording.
In this Graphical User Interface (GUI) section, the user has the
ability to perform analysis of a vibration signal that has been
captured in a data log. After entering the file name and selecting
acquire data, the following functions are available:
1) View the evolution of the time-domain signal where a vibration
stream feeds the measurement window.
2) Compute, plot, and record six features of the time-domain signal.
Baseline features are the signal’s mean, rms, variance,
kurtosis, crest factor, and peak amplitude for a certain window
size (default window length is 512 samples)
3) View the evolution of the frequency domain signal (via the
DFT), where vibration streams are processed for updating the screen.
Spectrum patterns can be analyzed off line for identifying and
visualizing frequency data patterns.
4) Compute, plot, and record six features of the frequency-domain
signal. Baseline features are the signal’s mean, rms, variance,
kurtosis, crest factor, and peak amplitude for a certain window
size (default window length is 512 samples)
5) Activate multi-resolution analysis (MRA) and
plot/record results for four types of wavelets. Four wavelet algorithms
are provided: (1) the standard discrete Haar transform (S-DHT);
(2) lifting scheme discrete Haar transform (LS-DHT); (3) Linear
interpolation wavelets; and (4) Daubechies D4 wavelet. Multi-resolution
analysis is then made possible by allowing the user to select
a desired MRA level for generating a set of averages and coefficients.
Time Domain Features
-
The second major section in this page is feature
extraction for designing pattern recognition algorithms (such
as neural network). In the case of methods based on Artificial
Neural Networks (ANN), the user can identify, select, and extract
parameters which are correlated with the system response (features)
for building neural network training and testing files. This process
entails the following:
1) Compute and plot a characteristic frequency response for
a representative state (such as normal operation) of a structural
component.
2) Extract as many as 11 major features from the representative
data log as well as for up to three additional logs.
3) Generate neural network training and testing
files that are built using the extracted features as inputs
and user-defined class IDs as outputs.
Feature Selection
|