A novel neural-network approach of analog fault diagnosis based on kernel discriminant analysis and particle swarm optimization

作者：Xiao, Yingqun^*; Feng, Lianggui

来源：Applied Soft Computing, 2012, 12(2): 904-920.

DOI：10.1016/j.asoc.2011.10.002

• Analog circuits
• Fault diagnosis
• Maximal class separability
• Kernel principal component analysis (KPCA)
• Kernel linear discriminant analysis (KLDA)
• Particle swarm optimization
• Neural network

Kernel principal component analysis (KPCA) and kernel linear discriminant analysis (KLDA) are two commonly used and effective methods for dimensionality reduction and feature extraction. In this paper, we propose a KLDA method based on maximal class separability for extracting the optimal features of analog fault data sets, where the proposed KLDA method is compared with principal component analysis (PCA), linear discriminant analysis (LDA) and KPCA methods. Meanwhile, a novel particle swarm optimization (PSO) based algorithm is developed to tune parameters and structures of neural networks jointly. Our study shows that KLDA is overall superior to PCA, LDA and KPCA in feature extraction performance and the proposed PSO-based algorithm has the properties of convenience of implementation and better training performance than Back-propagation algorithm. The simulation results demonstrate the effectiveness of these methods.

• 出版日期2012-2
• 单位中国人民解放军国防科学技术大学