基于WT与LSSVC的信号消噪方法及其在转子故障诊断中的应用

Wavelet denoising using least squares support vector classifier and its application to rotor fault diagnosis

噪声是影响转子早期故障特征有效识别与准确提取的主要因素。针对转子故障信号的消噪问题,提出一种基于最小二乘支持向量分类器的小波消噪方法。该方法以信号与噪声在小波域内的统计特征为理论依据,通过构造最小二乘支持向量分类器,以分类方式实现小波域内的信号与噪声判别,并对信号与噪声小波系数采取衰减策略以弱化噪声污染。模拟信号消噪分析与转子故障信号消噪实例表明,该方法可有效抑制信号中的噪声干扰,在信噪比与均方误差意义下的消噪性能优于小波阈值消噪方法。

Noise is the primary factor that affects the accuracy of rotor fault diagnosis.In order to extract the fault feature from noise-contaminated fault signals of rotor,a denoising method using wavelet transform（WT） and least squares support vector classifier（LSSVC） is proposed.Firstly,WT is applied to the noisy signal to obtain the scaling coefficients and the wavelet coefficients.The wavelet coefficients that represent the noise are identified by the trained LSSVC and then set to zero.The denoised signal is reconstructed by applying the inverse WT to the scaling coefficients and the processed wavelet coefficients.Extensive numerical experiments on simulated signals and rotor fault signals are carried out to confirm the effectiveness of the proposed method.The experimental results indicate that the method outperforms those by using conventional threshold-based wavelet in noise reduction.