Application of Supervised and Unsupervised Machine Learning to the Classification of Damaged Rotor‐Bearing Systems

Abstract

AbstractIn this letter, supervised machine learning classifiers are compared with clustering algorithms in the categorization of malfunctioned rotor systems. For this purpose, a dataset containing 660 faulted rotor‐bearing systems—that are, unbalanced, misaligned, and cracked—is used. Samples are created utilizing the finite element method in MATLAB. A sequential forward selection (SFS) method is employed to reduce the number of features in the signal‐processing stage after the feature extraction phase in the time, frequency, and time–frequency domains. The outcomes of three supervised algorithms—support vector machine, k‐nearest neighbors, and ensemble learning, as well as one unsupervised procedure, e.g., k‐Means clustering, are compared. In the latter method, to find the optimal number of clusters, the Calinski–Harabasz criterion is applied. The findings represent that, even though the supervised methods' acquired accuracies are noticeably higher (97.7% in the validation stage), using clustering algorithms can be beneficial in a variety of real‐time condition monitoring applications in rotating machines where the type, extent, and location of the damage are unknown.