Combination of discrete cosine transform with neural network in fault diagnosis for rotating machinery

The orbits of shaft centerline are pieces of indispensable information for the rotating machinery fault diagnosis, in general, a special shape of orbits of shaft centerline corresponds to a special fault type. The application of neural network is helpful to identify the orbits of shaft centerline, and the discrete cosine transform is helpful to reduce the input dimension of neural network. The paper discusses the method of combining the discrete cosine transform technique with the neural network, to compress the input data while the resolving power of input network is improved, so as to keep the input dimension of neural network invariant. The feasibility of improved back-propagation algorithm which makes the convergence faster is proved. Learned with the experimental simulated fault data, the neural network system can be used to identify orbits of shaft centerline in a higher identification rate automatically.