【摘要】 由于依靠单一的物理特征难以全面反映机械的故障信息,针对这一问题,对机器学习中常用的故障特征提取方法进行了研究,在此基础上提出了一种基于完备总体经验模态分解(CEEMD)、散布熵(DE)和Hjorth参数的混合特征滚动轴承故障诊断方法。首先,基于CEEMD对轴承原始信号进行分解,得到了若干个固有模态函数(IMF)分量;然后,根据与原信号的相关性选取敏感IMF分量,并求出其DE和Hjorth参数,形成散布熵特征向量和Hjorth参数矩阵,再对Hjorth参数矩阵进行奇异值分解,提取出奇异值作为特征向量,并将该向量与散布熵特征向量形成混合特征向量;最后,利用基于粒子群优化算法的最小二乘支持向量机(LSSVM),对滚动轴承不同故障特征向量进行了训练和识别。研究结果表明:该方法能够准确地诊断出滚动轴承的故障类型和程度,突出不同故障的特征;与采用单一特征的方法相比,该方法能更准确地辨别出滚动轴承的故障信息,采用该方法获得的故障识别率可达到100%。更多还原

【Abstract】 Aiming at problems of a single physical feature being difficult to fully reflect fault information, the common fault feature extraction methods in machine learning were studied. A mixed feature fault diagnosis method based on complete ensemble empirical mode decomposition(CEEMD), dispersion entropy(DE) and Hjorth parameters was proposed. Firstly, the bearing signal was decomposed by CEEMD to obtain intrinsic mode function(IMF) components. Secondly, the sensitive IMFs were chosen by using the correlation with the original signal to calculate its DE and Hjorth parameters, and formed DE feature vector and Hjorth parameter matrix. Then the singular value decomposition(SVD) was applied to transform Hjorth parameter matrix into a singular value vector, and a mixed feature vector was formed with the singular value and DE feature vector. Finally, the least square support vector machine(LSSVM) based on particle swarm optimization(PSO) was used to train and identify different fault feature vectors. The results indicate that the method can accurately diagnose the fault type and degree of rolling bearings, and highlight the characteristics of different faults. Comparing with the method using single feature, the recognition rate reaches 100% after using this method, it can identify the fault information more accurately.更多还原