非完备先验知识下的滑动轴承摩擦状态识别  

Plain bearing friction state recognition without complete prior knowledge

作　　者：张峻宁[1] 张培林[1] 李兵[2] 吴定海[1] 杨玉栋[3] 

机构地区：[1]军械工程学院七系,石家庄050003 [2]军械工程学院四系,石家庄050003 [3]武汉军械士官学校四系,武汉430075

出　　处：《航空动力学报》2017年第7期1704-1711,共8页Journal of Aerospace Power

基　　金：国家自然科学基金(51205405;51305454)

摘　　要：针对监测系统通常无法全部获取轴承摩擦退化状态的先验知识,无法建立全摩擦状态的识别模型,从状态间的相似性出发,提出一种无先验知识下的基于灰色B型绝对关联度(AGRDB)和稀疏编码的滑动轴承状态识别方法。针对稀疏表示不具有监督性的缺陷,在稀疏编码的目标函数中引入AGRDB算法,训练类间距离最大、类内距离最小的正常润滑和严重摩擦的编码;在相同字典下建立具有一致判别性的稀疏表示模型,通过比较当前状态与正常润滑、严重摩擦的稀疏编码与重构误差,进一步识别当前轴承的状态,仿真信号和柴油机轴承实验的结果表明:所提方法能够在较少先验知识下识别出滑动轴承的早期摩擦状态(100~216min)和严重摩擦状态(216~384min),且算法简单,适合较少样本下的滑动轴承摩擦故障在线监测。Given that the prior knowledge of all kinds of bearing friction degradation model cannot be attained usually, starting from the similarities of different states, a bearing friction fault＇s state recognition algorithm without prior knowledge was proposed based on sparse representation and absolute grey relational degree of B-mode （AGRDB）. First, for the defects of sparse representation without supervision, the AGRDB was involved in the sparse representation, to get normal and severe friction codes under the largest distance between classes and smallest distance within the classes. Second, sparse representation model with discriminant sex was established under the same dictionary. And current state of the bearing was identified by comparing sparse coding and reconstruction error of normal lubrication and serious friction. Finally, the results of simulation signal and diesel engine bearing experiment show that the proposed method can better identify the sliding bearing early friction state （100-216 min） and serious friction state （216- 384 min） under the less prior knowledge. And this algorithm is suitable for plain bearing fault monitoring under less samples.

关 键 词：非完备先验知识 滑动轴承 状态识别 灰色B型绝对关联度(AGRDB) 稀疏编码 

分 类 号：V21[航空宇航科学与技术—航空宇航推进理论与工程] TH117.2[机械工程—机械设计及理论]