Research on ELM Soft Fault Diagnosis of Analog Circuit Based on KSLPP Feature Extraction

Abstract

In order to improve the capability of soft fault diagnosis in an analog circuit, an integrated diagnosis method based on KSLPP feature extraction and ELM is proposed. The KSLPP feature extraction ability is firstly used to construct the principal component feature set from the fault sample set. Then, the advantage of ELM on solving the complicated nonlinearity problem is applied to build the fault identification model from the principal component feature. Finally, the sample sets of soft fault diagnosis for the analog circuit are respectively established by waveform parameter and wavelet packet transform to conduct the diagnostic test for the built model. The simulation of Elliptic Filter shows that, based on the fault sample set gotten by waveform parameter, the total correct rate of the integrated method is 98.2%; on the basis of the fault sample set built by wavelet packet transform, the total correct rate is 100%. The feasibility and effectiveness of this method are also validated.

Keywords

• Extreme learning machine
• Kernel supervised locality preserving projection
• feature extraction
• analog circuit
• fault diagnosis