Mechanical fault detection and classification using pattern recognition based on bispectrum algorithm

• Published in: Advanced Technologies for Sustainable Systems Vol. 4; pp. 147 - 165
• Main Authors: Habib, Michael R., Hassan, Mohammed A., Abul Seoud, Rania A., Bayoumi, Abdel M.
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2016; Springer International Publishing
• Series: Lecture Notes in Networks and Systems
• Subjects: Alternative & renewable energy sources & technology; Bispectrum; Condition Based Maintenance; Information retrieval; Logistic Regression; Machine Learning; Principle Component Analysis; Rotating Machinery Fault Detection; Tool making; Vibration Analysis
• ISBN: 3319487248; 9783319487243
• ISSN: 2367-3370; 2367-3389
• DOI: 10.1007/978-3-319-48725-0_15

Higher order spectral analysis of vibration signals is an efficient tool in condition monitoring and fault detection and diagnosis of rotating machinery. In this paper, features extracted from vibration bispectrum are used in fault classification of critical rotating components in the AH-64D helicopter tail rotor drive train system. Different classifiers are used to compare the performance of the proposed algorithm based on bispectrum to the traditional algorithms based on linear auto- and cross-power spectral analysis techniques. Principal component analysis (PCA) is used to reduce the size of features extracted from vibration bispectrum and linear spectral analysis, then the reduced set is used to train different classifiers. Using different criteria such as accuracy, precision, sensitivity, F score, true alarm, and error classification accuracy (ECA), the performance of the proposed algorithm is evaluated and compared against similar classification algorithms. The proposed method is verified using real-world data collected from a dedicated AH-64D helicopter drive-train research test bed at the CPM center, University of South Carolina. The proposed algorithm increases the accuracy of fault detection to 96.88%, precession to 95.83%, sensitivity to 95.83%.