基于K-means和高斯混合模型聚类的齿轮箱故障识别研究

Gearbox fault identification based on K-means and Gauss mixed model clustering

针对传统的基于振动信号的机械故障诊断技术过于复杂、诊断时间过长等问题,提出了结合K-means和高斯混合模型聚类方法的齿轮箱轴承和齿轮故障快速识别方法。首先,通过经验模态分解方法分解振动信号,利用相关分析选取了对振动信号局部特征表达最佳的IMF分量,IMF分量的均方根值和原始振动信号的均方根值,共同构成了振动信号特征集;然后,利用K-means算法确定了振动信号特征集的可分类别数;最后,基于振动信号特征集及其可分类别数,利用高斯混合模型聚类构造了齿轮箱运行状态的多维高斯分布函数,建立了齿轮箱在各运行状态下的从属概率模型,并根据从属概率大小,实现了齿轮箱故障的快速识别。实验和研究结果表明:针对实验环境下齿轮箱轴承和齿轮典型故障识别,基于K-means和高斯混合模型聚类的齿轮箱故障识别方法平均识别准确率为94.3%,高于基于模糊c均值聚类方法的故障识别平均准确率(84.5%)。

Aiming at the problem that the traditional mechanical fault diagnosis technology based on vibration signal was too complex and the diagnosis time was long,a fast identification method of gearbox bearing and gear fault based on K-means and Gaussian mixture model clustering method was proposed.First,the vibration signal was decomposed by empirical mode decomposition(EMD),and the IMF component which best expressed the local characteristics of the vibration signal was selected by correlation analysis.The vibration signal feature set was constituted by the root mean square values of the IMF component and the original vibration signal.Then,the classifiable number of the vibration signal feature set was determined by using K-means algorithm.Finally,based on the vibration signal feature set and its classifiable number,the multi-dimensional Gaussian distribution function of gearbox operation state was constructed by using Gaussian mixture model clustering,and the dependent probability model of gearbox was established in each running state.According to the value of subordinate probability,the gearbox fault could be identified quickly.The experimental and research results show that the average accuracy of the gearbox fault identification method based on K-means and Gaussian mixture model clustering is 94.3%,which is higher than the gearbox fault identification method based on fuzzy c-means clustering(84.5%)in the experimental environment.