Myeongsu Kang a, Gopala Krishnan Ramaswami b, Melinda Hodkiewiczc, Edward Cripps d, Jong-Myon Kim e, Michael Pecht a
a Center for Advanced Life Cycle Engineering (CALCE), University of Maryland, College Park, MD, USA
b Department of Physics, National University of Singapore, Singapore, Singapore
c School of Mechanical and Chemical Engineering, University of Western Australia, Crawley, WA, Australia
d School of Mathematics and Statistics, University of Western Australia, Crawley, WA, Australia
e Department of IT Convergence, University of Ulsan, Ulsan, South Korea
Abstract:
Machine learning techniques are indispensable in today’s
data-driven fault diagnosis methodolgoies. Among many machine techniques, knearest
neighbor (k-NN) is one of the most widely used for fault diagnosis due
to its simplicity, effectiveness, and computational efficiency. However, the lack
of a density-based affinity measure in the conventional k-NN algorithm can
decrease the classification accuracy. To address this issue, a sequential k-NN
classification methodology using distance- and density-based affinity measures
in a sequential manner is introduced for classification.
Keywords: Data-driven fault diagnosis, Density-based affinity measure,
k-Nearest neighbor, Machine learning