Fault feature extraction method for rotating machinery based on a CEEMDAN-LPP algorithm and synthetic maximum index

• Published in: Measurement : journal of the International Measurement Confederation Vol. 189; p. 110636
• Main Authors: Lu, Na, Li, Mingliang, Zhang, Guangtao, Li, Ruiqiang, Zhou, Tingxin, Su, Chengguo
• Format: Journal Article
• Language: English
• Published: London Elsevier Ltd 15.02.2022; Elsevier Science Ltd
• Subjects: Algorithms; Bearings; Complete ensemble empirical mode decomposition with adaptive noise; Fault diagnosis; Fault tolerance; Feature extraction; Locality preserving projection; Rotating machinery; Signal processing; Synthetic maximum index; Fault diagnosis; Feature extraction; Synthetic maximum index; Locality preserving projection; Rotating machinery; Complete ensemble empirical mode decomposition with adaptive noise
• ISSN: 0263-2241; 1873-412X
• DOI: 10.1016/j.measurement.2021.110636

•A novel method is proposed to extract fault features for rotating machinery.•Synthetic maximum index (SMI) is proposed for sensitive feature selection.•The locality preserving projection (LPP) algorithm further improves results.•The proposed method performs better than the compared methods. Fault feature extraction plays an important role in rotating machinery fault diagnosis. With progress in the development of signal processing methods, more and more features can be obtained from rotating machinery signals. However, these often contain many superfluous features, and the commonly used single evaluation criterion in the feature selection procedure is often inadequate for selecting sensitive features. With this problem in mind, a synthetic maximum index, with which both the global and local properties of features can be considered, is proposed for feature selection. Combining this index with the advantages of complete ensemble empirical mode decomposition with adaptive noise and the locality preserving projection algorithm, a novel fault feature extraction method for rotating machinery is proposed. The average fault recognition accuracies of this method on three datasets are 99.79%, 99.17% and 100%, respectively. Comparing it with seven comparative methods, the results demonstrate that the proposed method has better performance.