Contour graphics identification using neural networks

The shape of shaft orbit is the one of the most important graphic representation of the fault characteristic. It contains various fault information of diagnosis. This paper presents a novel fault diagnosis method of rotating machinery based on shaft orbit identification. Contour coding technique is an error-free coding technique, and it has properties of rotation, scaling and translation invariance after normalized. In the present study, binary image contour coding combined with perimeter method is used to extract shaft orbit feature vector. Then, radial basic function networks (RBF´s) are used to training feature vectors and diagnose the fault. By comparison experiment with the back-propagation (BP) network, the result indicates the proposed approach is faster and achieved satisfactory accuracy.