基于JADE-ICA的滚动轴承多故障信号盲源分离  

JADE-ICA-based blind source separation of multi-fault signals of rolling bearings

作　　者：席剑辉[1] 崔健驰 蒋丽英[1] 

机构地区：[1]沈阳航空航天大学自动化学院,沈阳110036

出　　处：《振动与冲击》2017年第5期231-237,共7页Journal of Vibration and Shock

基　　金：国家自然科学基金青年基金(60804025);辽宁省教育厅科学技术研究项目(L2014069;L2013070);沈阳市科技创新团队项目(SRC201204)

摘　　要：研究一种基于源信号高阶统计信息的矩阵联合近似对角化独立元分析(JADE-ICA)方法,并将其应用于滚动轴承故障声发射(AE)信号的盲源分离。滚动轴承的声发射源信号一般具有衰减性和准周期性,多组信号间还具有时差性,信号被多个传感器接收。通过最大程度的联合近似对角化,可以使源信号与分离信号有效的一一对应,克服非线性和时差的影响;通过高阶统计的高斯噪声不敏感性可以有效抑制随机观测噪声对分离结果的影响。选用相关系数、二次残差、性能指数和频谱特征构成系列时频域评价指标对分离结果进行较为全面的验证。仿真结果证明了该方法的可行性和有效性。Here, a joint approximate diagonalization of eigen-matrix and independent component analysis （JADE- ICA） method based on the high-order statistics of source signals was studied. It was applied to the blind source separation of rolling bearing faults＇ acoustic emission （AE） signals. The multi-AE source signals of rolling bearings were collected with multi-sensor. It was shown that the boiling bearing multi-AE source signals have characteristics, such as, time difference among multi-signals, with decay and quasi periodicity. Through the maximum joint approximate diagonalization, the one-to-one match between source signals and separated signals was realized to overcome the influence of nonlinearity and time difference. With the insensitivity of high-order statistics to Gaussian noise, the effects of random measured noise on the separated results were effectively suppressed. Correlation coefficient, quadratic residual, performance index and spectral characteristics were chosen to form a set of time-frequency domain evaluation indexes to verify the separated results. The simulation results verified the feasibility and effectiveness of the proposed method.

关 键 词：JADE 滚动轴承 故障诊断 声发射 

分 类 号：TP306[自动化与计算机技术—计算机系统结构]