Efficient feature extraction technique for diagnosing broken bars in three-phase induction machines. (February 2019)

• Record Type: Journal Article
• Authors: Liboni, Luisa H.B.; Flauzino, Rogério Andrade; da Silva, Ivan Nunes; Marques Costa, Eduardo C.
• Is Part Of: Measurement. Volume 134(2019)
• Page Start: 825
• Page End: 834
• Publication Date: 2019-02
• Subjects: Broken bars -- Fault diagnosis -- Induction machine -- Orthogonal decomposition -- Support Vector Machines (SVMs) -- Feature extraction; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2018.12.005
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: Fault analysis is made in the time domain so that frequency-domain transforms are not required. Decomposition products are sensitive to operating conditions and power consumption. Diagnostic system eases the classification task undertaken by statistical methods or intelligent systems. The tool can be used as a feature extractor for nonlinear event detection. Abstract: An original diagnostic method is proposed for identifying broken rotor bars based on a new technique for decomposing electric signals obtained by measurements, the Orthogonal Component Decomposition technique. This decomposition shows to be an efficient fault-specific feature extractor, and a broken bar simulation is used to determinate the relationships between the decomposition products and the fault phenomena. The efficiency of the information extraction is evaluated by Kruskall-Wallis variance analyses and Support Vector Machines on experimental signals, where the fault occurrence is detected, and the fault severity, given by the number of broken bars, is also diagnosed. The experimental signals are obtained by measurements from induction machines operating with different torque levels and driven either directly by the grid or frequency inverters. The proposed diagnostic method does not depend on frequency analysis and spectral resolution and comprises the Orthogonal Component Decomposition and Support Vector Machines. Furthermore, the results demonstrate the effectiveness of the information extraction since high diagnostic accuracy is achieved in low-torque situations and in a broad range of slip values. Additional analyses support that the proposed decomposition should be further investigated for many other applications on event detection.