Diagnosis of machine scream-identification of standard number of bearings and diagnosis of the failure modes

The authors describe a new acoustic method to identify standard numbers of ball bearings and diagnose their failure modes from the scream produced when they are rotating. They apply neural network techniques to the identification and diagnosis. An architecture for the hierarchically structured multilayer neural networks is presented. The tasks of identification of the numbers and diagnosis of the failure mode are allotted to each of the hierarchical stages. The network in the first stage directly receives a dB line scream spectrum and identifies the standard number of the ball bearing. A network in the second stage corresponding to the standard number identified in the first stage diagnoses the fault from the information on the repetitive frequency and stationarity or nonstationarity characteristics of the scream. The identification and diagnosis of bearings were carried out both by simulation and by experiments. It was found that the hierarchical neural network could effectively identify the standard number and diagnose the faults