It is important to extract fault features when machine would be in fault state. In order to sepa- rate different fault vibration signals from measured mixtures and diagnose the fault features of the machine ef- fectively according to the separated signals, a blind source separation (BSS) method using kernel function based on finite support samples was proposed. The method is stronger adaptability for the score functions esti mated according to finite support observed signal samples. The simulation results prove that the proposed BSS algorithm is able to separate hybrid mixtures that contain both sub-gaussian and super-gaussian sources. It is shown that the algorithm has better separation performance when compared with other BSS ones: The results of an experiment under the motor' s composite fault states with pedestal looseness fault and rotor unbalance fault show that this method is feasible for fault diagnosis.%机械设备发生故障时，故障特征的提取是很重要的．为了从观测信号中分离出不同的故障特征源信号，并根据分离信号准确地进行故障诊断，从观测信号样本出发，提出了基于有限支持样本核函数的盲源分离（FSS—kernelBSS）方法．此方法利用有限的观测样本估计信号的概率分布，得到了评价函数，具有很好的自适应能力．仿真试验结果表明：此方法能成功地分离超、亚高斯混合信号，与其他盲源分离方法相比，此方法具有更好的分离性能．将该方法用于转子不平衡和支座松动的复合故障信号的盲分离，分离出了各复合故障的主要频谱．分离结果表明：此方法应用于机械设备复合故障诊断中是可行的．
展开▼
