Elevating rotating machinery fault analysis: A multifaceted strategy with FFT, PCA, ANN, and K-means

Regis Cardoso, Willian Daniel de Mattos, Giancarlo Marchesini, Eduardo Camilo Inacio, Rodrigo Marques de Figueiredo, Sandro José Rigo

Abstract: The advancement of machine learning techniques has revolutionized the area of intelligent predictive maintenance, especially in industrial contexts, where fault diagnosis and classification stand out. This enables the development of systems that proactively identify incipient failures, recommending timely machine shutdowns to mitigate unsafe conditions within both the process and the surrounding environment. This research work proposes a novel predictive maintenance approach for classifying failures on rotating machinery that combines Fast Fourier Transform (FFT) and Principal Component Analysis (PCA) with supervised and unsupervised learning algorithms. An extensive evaluation of the proposed approach is carried out using real vibration data from a planetary gearbox system, introduced in the Data Challenge of the 2023 IEEE International Conference on Prognostics and Health Management (ICPHM23). Results demonstrate the effectiveness of the proposed multifaceted strategy with both supervised and unsupervised learning algorithms, outperforming state-of-the-art approaches, with a considerable low complexity methodology.

External IDs:dblp:journals/cee/CardosoMMIFR25