In this paper, we present fault diagnosis algorithms for monitoring the health of complex helicopter gearboxes. These techniques are based on neuro-fuzzy algorithms for pattern classification, pattern recognition, and sensor fusion. These techniques were developed using vibration data recorded from an aft main power transmission of a U.S. Navy CH-46E helicopter. Vibration data was collected using eight accelerometers mounted at different locations of the helicopter gearbox. Frequency domain and wavelet analysis techniques were used to analyze the vibration data, and extract the necessary fault features for neural network inputs. Self Organizing Maps (SOM) neural networks were used to cluster the features coming from different sensors to distinct locations on a two-dimensional maps. Several levels of SOM were used in case of wavelet feature preprocessing in order to achieve full separation of faults at all torque levels. Feedforward backpropagation neural networks were used to identify the different faults according to where they cluster on the 2-D maps. A fuzzy sensor fusion algorithm was developed to process information from different neural network-based fault identifiers, and produce a unified decision about the health of the monitored system. A comprehensive comparison between the two systems developed using wavelet and Fourier feature vectors will be presented.
Wavelet versus Fourier preprocessing for neuro-fuzzy systems for fault diagnosis in helicopter gearboxes
1998
6 Seiten, 18 Quellen
Aufsatz (Konferenz)
Englisch
Fuzzy connectionist network for fault diagnosis of helicopter gearboxes
Tema Archiv | 1995
|Efficient fault diagnosis of helicopter gearboxes
NTRS | 1993
|Unsupervised Connectionist Network for Fault Diagnosis of Helicopter Gearboxes
British Library Conference Proceedings | 1997
|Structure-Based Connectionist Network for Fault Diagnosis of Helicopter Gearboxes
Online Contents | 1998
|Automated fault diagnosis using neuro-fuzzy systems
AIAA | 1996
|