声学故障诊断中的宽带相关噪声信号分离方法

Separation of Wide-Band Coherent Noise Signals in Acoustic Based Diagnosis

为了提取混合噪声中待诊设备的噪声信号 ,利用传声器组构造线性测量阵列流型 ,建立多声源机器系统的宽带相关噪声混合模型。使用聚焦变换技术消除频率对参数估计的影响 ,并同时消除相关信号功率谱矩阵的奇异性 ,采用宽带相关 MUSIC法估计噪声方差与声源方位 ,从混合信号频谱中分离噪声源信号频谱。该方法可减小其他噪声源信号的干扰 ,提高待检设备噪声信号的信噪比。

In acoustic based diagnosis,a measured accoustic signal is the mixing of all the sounds from different sources,and the signal carrying fault features is drowned by the mixed signal.In this paper the signal is measured by microphone linear array manifold and a mixed model of wide-band coherent sound sources is established.Through focus transform matrix,the steering matrix is focused on a common frequency and also the sigular spectrum density matrix of correlated noises is transformed into nonsingular one.Then,the noise variance and the location of sound sources are estimated by wide-band coherent MUSIC algorithm.The fequency spectrum of the signal carrying fault features can be recovered from that of the mixed signal.The interference of other sound sources is eliminated and the SNR of the signal is improved significantly.The experiment results made in a semi-anechoic chamber demonstrate that the mixed model and the wide-band coherent MUSIC algorithm are effective in the separation.