摘要
对模糊C-均值聚类算法改进及在齿轮故障高效、可靠识别中的应用进行研究,提出了一种新的计及模糊聚类和特征选取的在线齿轮故障识别方法—基于混合蜻蜓优化多核模糊聚类和特征子集选取的在线齿轮故障识别算法。设计多核函数和贪婪聚类中心初始化策略,以克服模糊聚类算法对初始值敏感、高维复杂数据聚类鲁棒性差的缺陷;提出特征子集选取机制,更大限度降低数据处理维度;引入混合蜻蜓优化算法,将多核函数确定、最佳特征子集和聚类中心等效为蜻蜓个体编码,通过迭代求解最优聚类效果目标函数,在完成多核函数确定和最佳特征子集选取的同时,有效提高聚类算法求解效率;构建线下训练和线上故障识别模型,利用线下训练模式得到最佳模糊聚类个数等参数,并应用于复杂多样、高维非线性海量齿轮故障监测数据线上识别。仿真结果表明,提出的改进模糊聚类算法具有更好的聚类效果和更好的齿轮故障辨识度,高维复杂数据聚类正确率提高了约(6.5~9.3)%,齿轮故障识别检测成功率提高了约(11.1~31.7)%。
The improvement of fuzzyC-means clustering algorithm and its application in the efficient and reliable identification of gear faults are studied.A new on-line gear fault recognition method is proposed,which takes into account fuzzy clustering and feature selection.The initialization strategy of multi-core function and greedy clustering center is designed to overcome the shortcomings of fuzzy clustering algorithm that is sensitive to initial value and poor robustness of high-dimensional complex data clustering.The feature subset selection mechanism is proposed to reduce the data processing dimension to a greater extent.The hybrid dragonfly algorithm is introduced to effectively improve the efficiency of clustering algorithm while completing the determination of multicore function and the selection of the best feature subset through iterative solution of the optimal clustering effect objective function,by the identification of multi-core function,the best feature subset and clustering center equivalent to dragonfly individual coding.The model of offline training and online fault recognition is constructed,and the optimal number of fuzzy clusters is obtained by offline training mode,which is applied to online fault recognition.The simulation results show that the improved fuzzy clustering algorithm has better clustering effect and better gear fault identification.The accuracy of high-dimensional complex data clustering is increased by about(6.5~9.3)%,and the success rate of gear fault identification is increased by about(11.1~31.7)%.
作者
梁颖
马泳涛
LIANG Ying;MA Yong-tao(School of Mechanical and Electrical Engineering,Zhongyuan Institute of Technology,He'nan Zhengzhou 450007,China;School of Mechanical and Power Engineering,Zhengzhou University,He'nan Zhengzhou 450001,China)
出处
《机械设计与制造》
北大核心
2022年第1期9-14,19,共7页
Machinery Design & Manufacture
基金
国家自然科学基金青年基金(51705546)。
关键词
齿轮故障识别
模糊C-均值聚类算法
特征提取
蜻蜓算法
Gear Fault Recognition
Fuzzy C-Means Clustering Algorithm
Feature Extraction
Dragonfly Algorithm
