Feature extraction for bearing prognostics using weighted correlation of fault frequencies over cycles

In the predictive maintenance of rolling element bearings, finding a good prognostic feature to predict the remaining useful life is of critical importance, since it enables the maintenance scheduling in advance while ensuring the life without failure. In this article, a method for health index extraction is proposed for the bearing prognostics using the weighted correlation of fault frequencies over cycles. Although there have been numerous studies toward this purpose, this article is distinct in three aspects, which is more favorable for practical applications: First, the feature is simple and is rooted on the physical faults. Second, the method is applied after the anomaly detection. Third, the method is examined by the three run-to-failure cases with different types of bearings: one ball and two roller elements, which may suggest more general applicability. The trends of the proposed health index and prognostic performance are evaluated against the traditional features: root mean square and kurtosis. As a result, it is found that the new health index shows superior performance in view of the prognosis, which exhibits gradual and monotonic increase in overall performance with respect to the cycles since the inception of anomaly detection at the early stage.