【作者】 张然；

【导师】 潘艳秋；

【作者基本信息】 大连理工大学 ， 化学工程（专业学位）， 2018， 硕士

【摘要】 互联网的高速发展掀起了一场新的工业革命,整个工业生产体系提升到了一个崭新的水平。石化行业在此背景下正朝着大型化、一体化和智能化方向发展。压缩机组作为石化企业重要的动力源,故障的发生将造成重大的经济损失或人员伤亡。因此,保证压缩机组安全运行对企业的稳定发展具有极其重要的意义,深入开展石化企业大型压缩机组安全运行研究十分必要。本文结合中石化智能工厂基于大数据的设备性能劣化评估及诊断提升项目,根据国内外的研究状况,调研了国内某石化企业所用压缩机组的工作原理和结构特征,建立振动和工艺数据采集模型,结合常见故障的振动特征和表现形式,给出一个状态监测分析实例,为企业做出评估报告,提供维修决策依据。在此基础上构建了压缩机组健康状态评价框架,建立基于主成分分析法、相关性分析法、单因素方差分析法和k-means聚类分析法的压缩机组横向、纵向健康评估模型,选择评价指标,说明应用场景与功能。根据健康评估模型,对压缩机组划分了4个子系统,利用采集的数据进行压缩机组健康状态的定量体检和评分,并做出可视化展示。为方便企业人员实施监控,用较少的数据量替代多个量进行实时展示,用主成分分析法对压缩机组的振动特征值进行降维处理,选取了6个主成分作为新的特征值代表12个原始变量,实现50%的测点降维;用相关性分析法对机组的振动测点进行降维处理,结果表明同一台压缩机组的轴振动值显著相关、轴位移的水平和垂直振动值高度相关。提出用单因素方差分析法对5台不同压缩机的轴位移进行历史数据横向对比,得出5台压缩机组的健康水平排序。再用k-means聚类分析法对机组历史的状态等级进行归类,观测机组基于时间序列的运行状态,并与监测系统中的机组历史报警时间做对比,结果表明当聚类数k=3时的聚类结果与发生故障时的监测系统实际报警时间基本吻合,且能很好地反映设备实际的故障状态。最后对以上分析在石化大数据平台ProMACE~?工业云的开发实现以及下一步故障预测分析做出展望。更多还原

【Abstract】 The rapid development of the Internet has created a new industrial revolution and the entire industrial production system has been upgraded to a new level.In this background,the petrochemical industry is now moving toward large-scale,integrated,and intelligent.As an important power source for petrochemical factories,the compressor units will cause major economic losses or casualties when the fault happens.To ensure the safe operation of the compressor units is of great significance for the stable development of the factories.Therefore,it is necessary to carry out a thorough and careful research on the safe operation of the compressor units for the petrochemical enterprise.This work was carried out under the evaluation and diagnosis of equipment performance degradation promotion project based on big data of Sinopec Smart Factory.According to the research status at domestic and abroad,the working principle and structural characteristics of the compressor group used by a petrochemical enterprise were investigated.The vibration and process data acquisition model was established.Combining the vibration characteristics and manifestations of common faults,an example of condition monitoring analysis was presented.It is an evaluation reports for the enterprise and providing the basis for maintenance decisions.Moreover,a framework for evaluating the health status of the compressor units was funded.A horizontal and vertical health assessment model for the compressor units based on principal component analysis method,correlation analysis method,one-factor variance analysis method and k-means clustering method were established.The evaluation indicators were selected and the application scenes and functions was explained.According to the health assessment model,a compressor unit was divided into four sub-systems and the collected data were used to value the health status of the compressor units which was shown in a visual display.It has been proposed to use less amount of data instead of multiple quantities for real-time display which would be convenient for monitoring.The principal component analysis method was used to reduce the dimensionality of vibration characteristic values.6 principal components were selected as the new eigenvalues to represent the 12 original variables where 50%of the measurement points were reduced in dimension.The correlation values analysis method was used to reduce the dimensional values of the vibration measurement points of the units.The results showed that the shaft vibration of the same compressor group were significantly correlated and the horizontal and vertical axle displacements are highly correlated.The one-factor variation analysis method was used to compare the historical data of the axle displacement of the five different compressor units,and the health status of these five compressors were ranked.Furthermore,to observe the operation status of the compressor units based on time and to compare the units’historical alarm time in the monitoring system,the k-means cluster analysis method was used to classify the historical status of the units.The results indicated that when the clustering number k=3,the clustering results are basically consistent with the actual alarm time in the monitoring system when the fault occurs,and it can commendably reflect the actual fault state of the equipments.In the end,the above analysis will be prospected in the development of the petrochemical big data platform ProMACE~?industrial cloud and the fault prediction analysis.更多还原