We have addressed the issue of how to create better
classifiers to detect different failure types of a
hydraulic cylinder drive based on a set of temporal
measurements of key aspects of the system. We have shown
that one should try to discover a compact set of feature
extractors to identify key differences between the time
series prevailing during the different fault conditions.
In a case study we have created feature extractors and
classifiers for a test case by using visual data mining
to learn about the measurements and to discover the
differences between the measurement signals. The results
show a clear improvement on the classification results
previously achieved by using a different set of feature
extractors called statistical windows.
|Series||VTT Information Technology. Research Report|