列车轴承故障声音信号单通道去相关分离矫正

Single-channel de-correlation, separation and correction of train bearing fault sound signals

道旁声学监测信号具有多声源、陡畸变和强噪声的特点,这给基于声音信号的列车轴承故障诊断带来阻碍。针对道旁声学监测系统(track-side acoustic detection system,TADS)轮轨接触噪声干扰和多普勒畸变问题,提出了一种单通道去相关声源分离矫正方法(single channel de-correlation blind source separation-time-domain interpolation resampling,SCDBSS-TIR)。采用单个麦克风正对轮对轴承采集含噪单通道观测信号,利用奇异谱分析转换为双通道观测信号;利用多个时延相关矩阵特征值分解进行声源分离;利用时域插值拟合矫正法进行多普勒畸变矫正。采用单麦克风-双声源圆周运动模型开展仿真和试验验证,仿真和试验表明,在多普勒效应影响下,所提方法的分离效果优于经典的正定盲源分离法,同时对带内噪声的消除效果良好,所提方法有望在TADS中得到应用。

Trackside acoustic monitoring signals have characteristics of multiple sound sources,steep distortion and strong noise,they bring obstacles to fault diagnosis of train bearings based on sound signals.Here,aiming at wheel-rail contact noise interference and Doppler distortion in TADS system,a single channel de-correlation blind source separation-time-domain interpolation resampling(SCDBSS-TIR)method was proposed.A single microphone was used to collect single-channel observation signals with noise directly facing wheel-set bearing.Firstly,the singular spectrum analysis was used to convert single-channel observation signals into two-channel ones.Then,multiple time delay-correlation matrixes’eigenvalue decompositions were used to separate sound sources.Finally,the time domain interpolation fitting correction method was used to correct Doppler distortion.Simulation and experimental verification were conducted by using the circular motion model of single microphone-two sound sources.The results showed that under influences of Doppler effect,the separation effect of the proposed method is better than the classical positive definite blind source separation method;the proposed method’s elimination effect of in-band noise is good;the proposed method is expected to be applied in TADS.