Open-Circuit Fault Diagnosis of Neutral Point Clamped Three-Level Inverter Based on Sparse Representation

• Published in: IEEE access Vol. 6; pp. 73499 - 73508
• Main Authors: Yu, Yunjun, Pei, Shilei
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2018; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Circuit faults; Circuits; Clamping; Data mining; Diagnostic systems; Dictionaries; Fault diagnosis; fault identification; Feature extraction; Identification methods; Inverters; Neutral point clamped three-level inverter; Representations; Signal analysis; Signal processing; Singular value decomposition; Support vector machines
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2018.2883219

Decomposition of the signal on the orthogonal or nonorthogonal basis of the signal space is the traditional method for fault feature extraction in the field of inverter fault diagnosis. These signal analysis methods make the result of signal decomposition not sparse and they are not self-adaptive. In recent years, sparse representation has received considerable attention in signal processing because the method can overcome the shortcomings of traditional methods by decomposing the signal on an over-complete dictionary instead of on an orthogonal or nonorthogonal basis. This paper proposes a combination of sparse representation and support vector machine (SVM) for the fault diagnosis of neutral point clamped (NPC) three-level inverter. First, the three-phase phase voltage signals are sampled as the characteristic signals for analysis. Then, the K-SVD algorithm is used as the fault feature extraction technology to obtain an over-complete dictionary and the sparse representation coefficients of the characteristic signals. The latter are used as the feature information for the characteristic signals. Finally, the SVM with powerful generalization capability is used as the fault identification method to identify NPC three-level inverter fault types according to the extracted feature information and analyze the fault diagnosis effect. Simulation experiments show that the combination of sparse representation and the SVM for fault diagnosis of NPC three-level inverters has the advantage of high diagnostic accuracy. The fault diagnosis method proposed in this paper is compared with other methods to further verify its superiority in the fault diagnosis of NPC three-level inverters.