摘要
针对强噪声背景的高频振动信号,给出一种利用小波包消噪和频带分割技术,根据信号能量在小波包空间的分布特性,提取故障信号特征信息的方法.在小波包空间自适应软阈值去噪,消除白噪声;运用频带分割消除有色噪声,计算各子空间的能量,抽取低维特征矢量,作为小波网络的输入.该方法既提高了小波包神经网络的故障识别性能,又简化了决策网络结构,提高了收敛速度.
As for high frequency vibration signals submerged in strong noise, one method of fault feature extraction is put forward with Daub4 orthogonal Wavelet Package Transform (WPT) basing on both energy distribution analyzing and de-noising in wavelet packet subspaces. A perfect method is designed to eliminate white noise with adaptive soft-threshold wavelet packet shrinking, de-noise pink noise according to frequency bands split, analyze the signal energy distribution of wavelet packet subspaces, and extract the character low dimension eigenvectors of fault information as input data of the Wavelet Neural Network (WNN). As a result, this approach simplifies the WNN structure and speeds its convergence process so as to enhance the WNN capability of Fault Detection and Isolation (FDI).
出处
《哈尔滨工业大学学报》
EI
CAS
CSCD
北大核心
2005年第4期561-564,共4页
Journal of Harbin Institute of Technology
关键词
故障识别
软阈值
特征信息
小波包谱
小波神经网络
Feature extraction
Hydraulic turbines
Interference suppression
Neural networks
Vibrations (mechanical)
White noise