【作者】 孙希；

【导师】 李鑫滨；

【作者基本信息】 燕山大学 ， 控制理论与控制工程， 2012， 硕士

【摘要】 滚动轴承是各种旋转机械中应用最广泛的一种通用机械部件，其工况监控与故障诊断具有重要意义。本文研究基于小波包的轴承故障特征提取及基于模糊C均值(Fuzzy C-Means，简称FCM)多分类器组的故障识别方法，以提高轴承故障诊断系统的准确率。首先，对滚动轴承的故障机理进行了分析，并分别采用时域和基于傅里叶变换的频域方法对轴承振动信号的特征进行提取与分析。分析结果表明，基于时域和傅里叶变换的频域方法对信号的故障特征描述各有局限性。进而，本文采用基于小波包的信号能量特征提取方法。其次，由于多分类器组的鲁棒性较好，能克服因某些原因造成的数据失真或缺失引起的误判或漏判。在故障识别中采用FCM多分类器对故障特征进行分类识别，得到各类的聚类中心及各个故障样本对于各故障类型的隶属度，从而实现模糊聚类划分。FCM算法具有较高的搜索能力，但它是一种局部搜索算法，且对聚类中心的初值十分敏感，并容易陷入局部极值。为避免分类时陷入局部最优，本文利用粒子群优化(Particle Swarm Optimization，简称PSO)算法的全局搜索能力对FCM的聚类中心进行优化，进而得到一种基于PSO优化的FCM多分类器组故障特征识别方法。最后，在进行FCM多分类器组融合时采用模糊积分(Fuzzy Integral，简称FI)融合方法，用以提高多分类器融合系统的分类精确率。在基于FI的多分类器融合系统中，模糊测度对融合系统的性能有很大的影响，通常情况模糊测度为预先人为给定。因此，本文同样采用PSO算法优化模糊测度，改进基于FI融合的多分类器组的分类精度和分类效率。仿真研究表明基于小波包和FCM多分类器组的故障诊断方法能够有效地提高轴承故障诊断的识别准确率。更多还原

【Abstract】 The rolling bearing is the most widely used mechanical components in a variety ofrotating machinery, its condition monitoring and fault diagnosis possess significance. Inorder to improve the accuracy of the bearing fault diagnosis system, bearing fault featureextraction based on wavelet packet and fault recognition based on FCM classifier group.First of all, the analysis of the failure mechanism of rolling bearing, time domain andfrequency domain based on Fourier transform are used in the bearing vibration signalfeature extraction and analysis. The analysis showed that the failure characteristics of thesignal based on time domain and Fourier transform have their own limitations.Furthermore, using the signal energy feature extraction methods based on wavelet packet.Secondly, for the robustness, multiple classifier group is able to overcome the falsepositives caused by data distortion or missing due to some reason. Using multipleclassifiers of FCM achieve the classification of fault characteristics in the faultidentification. FCM algorithm has better search capabilities which is a local searchalgorithm and so sensitive to the initial value of the cluster center that is easy to fall intolocal minima. In order to avoid the local optimum, using Particle Swarm Optimizationalgorithm, which of the global search ability optimize the cluster centers of FCM andachieve the failure feature recognition method of FCM classifier group based on PSO.Finally, in order to improve the classification accuracy rate of multiple classifiersfusion system and the robustness of the system, fuzzy integral fusion method is usedduring the FCM classifier group fusion. Based on fuzzy integral classifier fusion system,fuzzy measure has a great impact on the performance of the fusion system and is given onthe pre-anthropogenic. Therefore, the PSO algorithm is used to optimize the fuzzymeasure and improve the classification accuracy and efficiency based on fuzzy integralfusion group. Simulation studies show that the fault diagnosis method based on waveletpacket and FCM classifier group is effectively improve the recognition accuracy.更多还原