Title :
Methodology for evaluating neural networks inputs for gear fault detection
Author :
Moreno, R. ; Pintado, P. ; Chicharro, J.M. ; Morales, A.L. ; Nieto, A.J.
Author_Institution :
Area de Ing. Mec., E.T.S.I. Ind. (UCLM), Ciudad Real
Abstract :
In this paper, the artificial neural network (ANN) inputs selection for detecting and quantifying the progressive value of an incipient defect in gears is carried out by experimental design evaluation. Several parameters in time-domain (root mean squared, crest factor, energy ratio, FM0, Kurtosis, FM4, NA4, M6A, NB4) and multiscale Hilbert-wavelet transformations are evaluated as a possible inputs. Suitable inputs, according to the proposed evaluating methodology, do enhance ANN performance.
Keywords :
Hilbert transforms; condition monitoring; design of experiments; fault diagnosis; gears; neural nets; time-domain analysis; wavelet transforms; artificial neural network input selection; condition monitoring; experimental design evaluation; gear fault detection; multiscale Hilbert-wavelet transformation; time-domain analysis; Analysis of variance; Artificial neural networks; Cepstral analysis; Design for experiments; Fault detection; Gears; Maintenance; Neural networks; Torque; Wavelet transforms; Gear; detection; fault; neural network inputs;
Conference_Titel :
Mechatronics, 2009. ICM 2009. IEEE International Conference on
Conference_Location :
Malaga
Print_ISBN :
978-1-4244-4194-5
Electronic_ISBN :
978-1-4244-4195-2
DOI :
10.1109/ICMECH.2009.4957192