基于大数据的设备故障全矢预测模型研究

发布时间：2018-12-21 12:20

【摘要】：旋转机械是机械装备的重要组成部分,一旦出现故障,会导致整个设备或生产过程停止运行,甚至造成严重的安全事故和重大经济损失。因此,机械设备故障预测受到人们的关注和研究。传统频谱分析方法仅依赖单通道振动信息,丢失了信息的完整性,而全矢谱技术采用同源双通道信息融合的思想,保证频谱包含完整、全面的振动信息;现代设备监测系统通常采集大量监测信号,但在设备预测过程中,并不能充分利用历史监测信息,导致中长期预测可信度低,时间序列聚类方法能把近似状态的时刻聚类,将大量历史监测信息简化,使得预测过程中利用更多历史信息;为克服全矢-ARIMA(FV-ARIMA)和全矢-SVR(FV-SVR)预测模型的缺陷,提出改进的全矢-SVR预测模型。本文以设备监测大数据为研究对象,全矢谱技术和时间序列聚类为理论支撑,结合改进的全矢-SVR预测模型,对设备故障预测进行研究。主要研究工作如下:(1)详细研究了全矢谱技术的理论和算法,给出Hilbert-全矢谱的算法步骤,并将其应用于滚动轴承的退化分析中,验证了Hilbert-全矢谱具有良好的包络解调效果,所求得的特征主振矢能够表征振动强度,区分故障类型。(2)研究设备监测数据的特点,并给出设备监测大数据的概念;研究数据的平滑处理方法和时间序列聚类分析方法,并将其应用到真实的时序序列中,获得良好的平滑处理效果和聚类效果。(3)研究ARIMA模型和SVR预测的基本理论和算法;给出全矢预测模型的基本流程;通过对滚动轴承的状态预测,分析并总结全矢-ARIMA和全矢-SVR预测模型的优点和缺点。(4)针对全矢预测模型的缺点,提出改进的全矢-SVR预测模型;结合时间序列聚类分析和改进的全矢预测模型,构建基于大数据的中长期设备故障全矢预测模型;采用滚动轴承运行过程中的全部历史数据,分别对改进的全矢-SVR预测模型和基于大数据的中长期设备故障全矢预测模型进行实验验证,结果显示,两种预测方法均取得良好的预测效果。
[Abstract]:Rotating machinery is an important part of machinery and equipment, once it fails, it will cause the whole equipment or production process to stop running, and even cause serious safety accidents and major economic losses. Therefore, mechanical equipment fault prediction has attracted people's attention and research. The traditional spectrum analysis method only relies on the single channel vibration information and loses the integrity of the information, while the whole vector spectrum technology adopts the idea of homologous and dual channel information fusion to ensure that the spectrum contains complete and comprehensive vibration information. Modern equipment monitoring system usually collects a large number of monitoring signals, but in the process of equipment prediction, the historical monitoring information can not be fully utilized, which leads to the low credibility of the medium and long term prediction, and the time series clustering method can cluster the approximate state of time. Simplifying a large number of historical monitoring information to make use of more historical information in the prediction process; In order to overcome the defects of total vector ARIMA (FV-ARIMA) and total vector SVR (FV-SVR) prediction models, an improved total vector SVR prediction model is proposed. This paper takes big data as the research object, total vector spectrum technology and time series clustering as the theoretical support, combined with the improved full-vector SVR prediction model, to study the equipment fault prediction. The main research work is as follows: (1) the theory and algorithm of total vector spectrum technology are studied in detail, and the steps of Hilbert- complete vector spectrum algorithm are given, and applied to the degradation analysis of rolling bearings. It is verified that Hilbert- full-vector spectrum has a good envelope demodulation effect, and the characteristic principal vibration vector can characterize vibration intensity and distinguish fault types. (2) the characteristics of equipment monitoring data are studied, and the concept of equipment monitoring big data is given. The data smoothing method and time series clustering analysis method are studied and applied to real time series to obtain good smoothing effect and clustering effect. (3) the basic theory and algorithm of ARIMA model and SVR prediction are studied. The basic flow chart of the full-vector prediction model is given. The advantages and disadvantages of full-vector ARIMA and full-vector SVR prediction models are analyzed and summarized through the state prediction of rolling bearings. (4) aiming at the shortcomings of the full-vector prediction model, an improved full-vector SVR prediction model is proposed. Combined with time series clustering analysis and improved full-vector prediction model, the full-vector prediction model of medium- and long-term equipment fault based on big data is constructed. Using all the historical data of rolling bearing operation, the improved full-vector SVR prediction model and the full-vector prediction model of medium-long term equipment fault based on big data are tested, and the results show that, The two prediction methods have achieved good prediction results.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH17

【参考文献】