Method of Redundant Features Eliminating Based on K-Means Clustering

When diagnosing mechanical faults using multi signal features, in order to keep the most efficient signal features and eliminate redundant ones, and make the mechanical fault diagnosis achieve a balance between computational complexity and diagnostic accuracy, a redundant features eliminating method based on k-means clustering was proposed. By optimizing the result of k-means clustering we obtained weights of all the inputted features, after setting appropriate threshold, compare it with all feature weights, eliminate those features whose weight is smaller than the threshold. After compared this method with some advanced methods, it shows that when using the SVM multi-classifier for mechanical fault diagnosis, this method produces a higher accuracy; and since eliminating part of features, it makes the time needed for diagnosis shorter.