【作者】 钱静；

【导师】 林利红；

【作者基本信息】 重庆大学 ， 工程（机械工程）（专业学位）， 2022， 硕士

【摘要】 离心鼓风机作为环保、化工等领域中不可或缺的关键设备,正朝着大型化、复杂化、自动化方向发展,由于其零部件互相耦合,运行工作环境恶劣加之自身不断疲劳退化,极易发生故障,一旦发生故障将造成巨大的经济损失甚至人员伤亡。随着传感器技术及工业互联网的发展,利用数据驱动方法挖掘大量监测数据中的隐藏健康信息,是实现离心鼓风机退化趋势预测及其剩余使用寿命预测的有效途径。在设备监测数据收集过程中,往往由于传感器故障、网络传输故障等原因导致数据出现缺失,导致离心鼓风机的退化趋势及剩余使用寿命的预测精度下降,甚至出现错误的预测结果。因此,在考虑数据缺失情况下对离心鼓风机缺失数据进行填补,并基于填补之后的完整数据进行离心鼓风机的退化趋势预测和剩余寿命预测具有重要的意义。首先,介绍了离心鼓风机缺失数据填补流程,描述了离心鼓风机数据采集系统和监测数据标准化采集与控制架构,分析了离心鼓风机监测数据缺失原因及缺失模式;构建了离心鼓风机缺失数据四维张量,并基于Tucker分解方法实现离心鼓风机原始张量的重构和监测数据的填补。其次,提取填补后的完整监测数据的特征并进行归一化,使用深度置信网络构建离心鼓风机的健康指标;为解决现有预测模型在长时间序列预测时历史信息挖掘不充分问题,使用基于注意力机制建模的Informer深度学习方法对离心鼓风机健康指标进行预测,提高离心鼓风机退化趋势预测的准确率。然后,针对填补后的完整监测数据,基于3σ原则对离心鼓风机的退化起始点进行检测,针对退化起始点之后时刻监测数据,使用卷积神经网络充分挖掘多种监测信息的隐藏健康特征,并结合门控循环单元预测方法构建离心鼓风机的剩余使用寿命预测模型。最后,结合实验说明考虑缺失数据下对离心鼓风机进行退化趋势预测和寿命预测的必要性,对比了数据缺失与数据填补情况下对离心鼓风机进行退化趋势预测和剩余使用寿命预测的影响,并与其他常见预测方法结果进行对比,验证本文方法的有效性与优越性。更多还原

【Abstract】 Centrifugal blower as indispensable large equipment in areas of environmental protection and chemical industry,is developed toward largescale,complication and automation.Due to the mutual coupling between its components,the harsh working conditions and the constant fatigue degradation,faults are easy to happen,which will cause huge economic losses and even casualties.With the development of sensor technology and internet industry,using the method of data driven to mine the hidden health information in mass monitoring data is the effective way to realize the degradation trend prediction and remaining useful life prediction of the centrifugal blower.However,in the process of data collection,monitoring data is often lost due to sensor faults,network transmission faults and other reasons,which will lead to degradation trend of centrifugal blower and prediction accuracy of remaining service life decline or even wrong prediction results.Therefore,it is of great significance to fill in the missing data of centrifugal blower under the condition of data missing,and to predict the degradation trend and remaining useful life of centrifugal blower based on the complete data filled.Firstly,the missing data filling process of centrifugal blower was introduced,the centrifugal blower data acquisition system and the standardized monitoring data acquisition and control architecture were described,and the causes and modes of missing monitoring data of centrifugal blower were analyzed.The four-dimensional tensor of missing data of centrifugal blower was constructed,and the original tensor of centrifugal blower was reconstructed and the monitoring data was filled based on Tucker decomposition method.Secondly,the characteristics of the complete monitoring data after filling were extracted and normalized,and the health index of the centrifugal blower was constructed using the deep confidence network.At the same time,to solve the problem of insufficient historical information mining of existing prediction models in long time series prediction,the Informer method based on attention mechanism modeling was used to predict the health indicators of the centrifugal blower,and the accuracy of the centrifugal blower degradation trend prediction was improved.Thirdly,the degradation starting point of the centrifugal blower was detected based on the 3σ principle.For the monitoring data at the time after the degradation starting point,the convolution neural network was used to fully excavate the hidden health characteristics of various monitoring information.Combined with the prediction method of the gating cycle unit,the remaining useful life prediction model of the centrifugal blower was constructed.Finally,combined with the experiment,the necessity of the degradation trend prediction and remaining useful life prediction of the centrifugal blower considering missing data was illustrated.The effects of missing data and data filling on the degradation trend prediction and life prediction of the centrifugal blower were compared,and the results were compared with those of other common prediction methods to verify the effectiveness and superiority of the proposed method.更多还原