投影寻踪方法在设备预知维护中的应用研究
2016-10-29程晓涵汪爱明苏一新孟国营
程晓涵 汪爱明 苏一新 孟国营
摘要: 提出了一种融合投影寻踪的自回归分析方法实现设备的预知维护。该预知维护方法是从设备关键部件处提取振动信号,经分析和计算得出24种特征指标用以描述设备运行状态;对24种特征指标分别提取一个时间序列并各自进行自回归分析,得到各自对应的预测因子;利用投影寻踪将前述预测因子投影到二维空间,然后分别建立预测因子投影值与相对应的特征指标值的拟合函数,进而推算出24种特征指标的未来值;再通过对24种特征指标的未来值在最佳投影方向矩阵下进行投影,根据投影值的分布情况判断设备未来运行状态是否存在异常,从而实现设备的预知维护。最后利用美国西储大学轴承数据中心网站公开发布的轴承探伤测试数据集中的内圈故障数据和山西省某洗煤厂主井皮带机的减速器故障数据进行了验证。关键词: 故障诊断; 特征指标; 投影寻踪; 评价体系; 预知维护方法
中图分类号: TH165+.3; TP277文献标志码: A文章编号: 10044523(2016)04063107
DOI:10.16385/j.cnki.issn.10044523.2016.04.010
引言
目前设备的日常运行和维护工作基本上是依托于各种在线或者离线的监测监控系统,并主要依靠门限报警方式,通过对不同的监测参数设置一定的极限值进行超值报警,但当达到报警值时设备已经出现了不同程度的损坏或性能降低,导致设备存在很大的安全隐患,并使维护维修工作存在严重的被动性,因此需要寻求方法实现设备运行状态的预知,及时采取维护措施。
设备运行状况是否正常可以由不同的特征指标进行衡量,不同的特征指标对各类故障类型和不同的损伤程度可能表现出不同的灵敏度和响应特点,且相关性难以界定[12],因此文章选取了24种特征指标对设备状态进行描述,提出了一种基于投影寻踪方法的设备预知维护方法,即采用基于24种特征指标的融合投影寻踪方法的自回归分析,实现了对设备零部件未来状态的初步预测,并根据预测结果提出相应的维护维修建议。
Abstract: Equipment maintenance is passive when using traditional threshold alarm, so early detection of fault and prescient maintenance are badly in need. A method of regression analysis combined with projection pursuit method for equipments is put forward. The first step of this method is analyzing the vibration signal and calculating 24 characteristic indexes to describe the equipment running status. By extracting a time series respectively of these 24 characteristic indexes and regression analysis, predictors for each characteristic index are obtained. Predictors are respectively projected to twodimensional space using projection pursuit method, and then fitting functions between projection values of predictors and characteristic index are established, hereunder future values of 24 characteristic indexes are inferred. Then projecting these 24 future values under the best projection direction matrix and observation of distribution of projection values, whether the future status of the equipment abnormal could be judged and prescient maintenance of coal mine equipment could be realized. At last, the method proposed in this paper is verified by inner ring fault test data publicly released by the Case Western Reserve University Bearing Data Center website and failure data when gear reducer served to belt convey at main shaft of coal washery of one mining group in Shanxi Province was broken.
Key words: fault diagnosis; characteristic indexes; projection pursuit method; evaluation index system; predictive maintenance method