不确定性下Bayes-DeepONet旋转机械故障报警方法研究  

Bayesian DeepONet Fault Alarming Method for Rotating Machines with Uncertainties

作　　者：赵慧君 姜孝谟 王志成 Hui-jun Zhao;Xiao-mo Jiang;Zhi-cheng Wang(Faculty of Vehicle Engineering and Mechanics,Dalian University of Technology;State Key Laboratory of Structural Analysis Optimization and CAE Software for Industrial Equipment,Provincial Key Laboratory of Digital Twin for Industrial Equipment of Liaoning,School of Energy and Power Engineering,Dalian University of Technology;Laboratory of Ocean Energy Utilization of Ministry of Education,School of Energy and Power Engineering,Dalian University of Technology)

机构地区：[1]大连理工大学运载工程与力学学部 [2]工业装备结构分析、优化与CAE软件国家重点实验室,工业装备数字孪生辽宁省重点实验室,大连理工大学能源与动力学院 [3]海洋能源利用与节能教育部重点实验室,大连理工大学能源与动力学院

出　　处：《风机技术》2023年第6期54-59,共6页Chinese Journal of Turbomachinery

基　　金：国家领军人才项目(82211402);工业装备数字孪生国家重点实验室项目(3006-02020000)。

摘　　要：深度学习方法在旋转机械故障诊断中得到广泛应用,但现有的方法很少考虑数据和模型不确定性对诊断结果的影响,以致于实际应用中模型的泛化能力不足,精度无法保证。本文基于Dropout的近似变分推理与深度算子神经网络(DeepONet)模型无缝对接,提出一种基于不确定性下贝叶斯DeepONet模型的旋转机械故障预警新模型。使用大型压缩机案例,进行了方法验证,结果表明:考虑数据和模型的不确定性,所提方法具有非常高的准确性,实现对机组运行状态的实时监测和精准预警。Deep learning models have been widely used in fault alarming and diagnosis of rotating machinery,but most of the current methods do not consider the uncertainties in both the data and the model,thus leading to the inaccurate results.This paper presents a novel Bayesian deep operator network(DeepONet)model under uncertainties for fault alarming of rotating machines by seamless integrating the Bayesian theory,dropout and DeepONet model.Numerical results have shown the advantages of the proposed methods by using the data and events of a real-world centrifugal compressor.This study provides a promising tool in smart maintenance of large rotating machines by enabling early fault warning considering uncertainties.

关 键 词：不确定性 旋转机械 贝叶斯 DeepONet 故障预警 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TH133.3[自动化与计算机技术—控制科学与工程]