数据不完备下基于Informer的离心鼓风机故障趋势预测方法  

Centrifugal blower fault trend prediction method based on Informer with incomplete data

作　　者：张友 李聪波[1] 林利红[2] 钱静 易茜[1] ZHANG You;LI Congbo;LIN Lihong;QIAN Jing;YI Qian(State Key Laboratory of Mechanical Transmission,Chongqing University,Chongqing 400044,China;College of Mechanical and Vehicle Engineering,Chongqing University,Chongqing 400044,China)

机构地区：[1]重庆大学机械传动国家重点实验室,重庆400044 [2]重庆大学机械与运载工程学院,重庆400044

出　　处：《计算机集成制造系统》2023年第1期133-145,共13页Computer Integrated Manufacturing Systems

基　　金：国家自然科学基金资助项目(51975075);重庆市技术创新与应用示范专项资助项目(cstc2018jszx-cyzdX0146)。

摘　　要：离心鼓风机在运行过程中,监测数据缺失会导致故障趋势预测滞后和预测精度下降。针对该问题,提出一种考虑数据不完备的离心鼓风机故障趋势预测方法。首先,基于张量分解对缺失监测数据进行填补,获得离心鼓风机的完备监测数据;其次,基于填补后的完备监测数据利用深度置信网络(DBN)构建能表征离心鼓风机健康状态的健康指标;最后使用Informer方法预测健康指标的未来走势,实现离心鼓风机的故障趋势预测。案例分析结果表明,相比缺失数据,利用填补后的数据所建立的预测模型能更早预测故障的发生,同时所提出的预测方法较Transformer、长短时记忆(LSTM)和门控循环单元(GRU)等常用传统方法预测精度更高。During the operation of the centrifugal blower, the missing monitoring data will cause the fault trend prediction lag and prediction accuracy to decrease. To solve this problem, a fault trend prediction method considering incomplete data was proposed. The missing monitoring data were filled by tensor decomposition to obtain the complete monitoring data. Based on the completed monitoring data, the Deep Belief Network(DBN) was used to construct the health indicators that could characterize the health status of the centrifugal blower. The Informer method was used to predict the future trend of health indicator and realized fault trend prediction of the centrifugal blower. Experiment results showed that the prediction model used the filled data could predict the occurrence of faults earlier than incomplete data. Meanwhile, the prediction accuracy of the proposed method was higher than Transformer, Long Short Term Memory(LSTM), Gate Recurrent Unit(GRU) and other conventional methods.

关 键 词：离心鼓风机 故障趋势预测 不完备数据 Informer方法 张量分解 

分 类 号：TH17[机械工程—机械制造及自动化]