Mechanical Faults Diagnosis of High-Voltage Circuit Breaker via Hybrid Features and Integrated Extreme Learning Machine

• Published in: IEEE access Vol. 7; pp. 60091 - 60103
• Main Authors: Gao, Wei, Wai, Rong-Jong, Qiao, Su-Peng, Guo, Mou-Fa
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 2019; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Aliasing; Artificial neural networks; Bandpass filters; Circuit breakers; Circuit faults; Circuit reliability; complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN); Diagnostic systems; Distribution functions; Electric equipment; Electric power supplies; Electric power systems; Entropy; Fault diagnosis; Feature extraction; Frequencies; High voltages; High-voltage circuit breaker; Hilbert transformation; integrated extreme learning machine (IELM); Machine learning; Mathematical analysis; Matrix algebra; Matrix methods; Normalizing; singular entropy; Support vector machines; Time-frequency analysis; time-frequency entropy; vibration signal; Vibrations
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2019.2915252

As key electrical equipment in the power system, the normal operation of a high-voltage circuit breaker is related to the reliability and economy of the power supply. In this paper, a mechanical fault diagnostic method for a high-voltage circuit breaker via the hybrid feature extraction and the integrated extreme learning machine (IELM) is investigated. First, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the vibration signal to obtain intrinsic mode functions (IMF). Then, the sub-band reconstruction of each order IMF component is performed by combining the Hilbert transform and the band-pass filter in order to obtain the time-frequency matrix. Moreover, mechanical fault feature vectors can be formed by the time-frequency entropy and the singular entropy, which are extracted by transforming the time-frequency matrix into the energy matrix and normalizing the frequency bands via the normal cumulative distribution function (NCDF). In addition, an IELM is built for the fault classification. The advantages of the proposed CEEMDAN scheme in combination with band-pass filtering can eliminate the modal aliasing, reduce the number of auxiliary noise additions, and improve the decomposition efficiency. Besides, the performance of the singular entropy normalized by the NCDF is more stable, and the IELM composed of multiple weak classifiers can solve the shortcomings of the traditional extreme learning machine. The experimental results based on measured data show the proposed method can effectively diagnose the mechanical failure via small samples of high-voltage circuit breakers.