A Newly Designed Diagnostic Method for Mechanical Faults of High-Voltage Circuit Breakers via SSAE and IELM

• Published in: IEEE transactions on instrumentation and measurement 2021, Vol.70, p.1-13
• Main Authors: Gao, Wei, Qiao, Su-Peng, Wai, Rong-Jong, Guo, Mou-Fa
• Format: Article
• Language: English
• Subjects: Artificial neural networks; Circuit breakers; Circuit design; Circuit faults; Classifiers; Constraint of label; Diagnostic systems; Entropy; Fault detection; Fault diagnosis; Feature extraction; Feature recognition; High voltages; high-voltage circuit breaker (HVCB); identification of unknown types of fault (UTFs); integrated extreme learning machine (IELM); Machine learning; stacked autoencoder (SAE); Time-frequency analysis; Vibration; vibration signal; Vibrations
• ISSN: 0018-9456; 1557-9662
• DOI: 10.1109/TIM.2020.3011734

In general, vibration signals generated by the switching operation of a high-voltage circuit breaker (HVCB) contains important information to reflect its mechanical status. A method for mechanical fault diagnoses of an HVCB based on a semisupervised stacked autoencoder (SSAE) and an integrated extreme learning machine (IELM) is proposed in this study. First, the complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) is used to decompose the vibration signal to obtain the time-frequency energy matrix. Then, an SSAE model is applied to automatically extract the characteristic information from the energy matrix. As a result, two-level classifiers can be constructed. The first level is utilized to identify normal or abnormal states, and the second level is selected to identify various types of faults in the abnormal state. The classifiers of these levels are composed of binary IELM. The advantages of the proposed method are that it not only can automatically extract the high-recognition features from the time-frequency energy matrix of high dimension to complete the identification of the existing fault types in the training set but also can accurately identify the samples of unknown types of faults. Experimental results show that the proposed method can effectively diagnose mechanical faults of an HVCB, and the classification accuracy reaches 99.5%.