基于联邦分层优化学习的设备故障诊断方法
Equipment fault diagnosis method based on federated hierarchical optimization learning
刘晶 1王晓茜 2唐震 2吕华 2季海鹏3
作者信息
- 1. 河北工业大学人工智能与数据科学学院,天津 300400;河北省数据驱动工业智能工程研究中心,天津 300400
- 2. 河北工业大学人工智能与数据科学学院,天津 300400
- 3. 天津开发区精诺瀚海数据科技有限公司大数据部门,天津 300400;河北工业大学材料科学与工程学院,天津 300400
- 折叠
摘要
随着工业物联网的快速发展,联邦学习能够实现数据隐私下的多工厂联合设备故障诊断,然而不同工厂工况、计算资源、数据质量等异构性较大,导致传统联邦学习故障诊断的训练效率与准确率不高.针对上述问题,提出基于联邦分层优化学习的设备故障诊断方法.建立设备故障诊断分层架构进行层局部聚合缓解异构问题;提出本地个性更新选择算法减少模型偏移,提升联合诊断准确率;提出基于迭代阈值的局部聚合模型,通过动态迭代与局部聚合完成中心模型聚合.经实验分析验证,该方法显著提高了联合故障诊断的训练效率与准确率,具有良好的鲁棒性,满足多工厂高效设备故障诊断的工业需求.
Abstract
With the rapid development of the industrial Internet of Things,federated learning can realize the fault diagnosis of multi-plant joint equipment under data privacy.However,the heterogeneity of different plant conditions,computing resources,data quality,etc.is large,resulting in the low training efficiency and accuracy of traditional federated learning fault diagnosis.Aiming at the above problems,a method of equipment fault diagnosis based on federated hierarchical optimization learning was proposed.Ahierarchical architecture for equipment fault diagnosis was established for local aggregation to alleviate the heteroge-neous problem.A local personalized update selection algorithm was proposed to reduce model bias and improve the accuracy of joint diagnosis.A local aggregation model based on iteration threshold was proposed,and the central model aggregation was completed through dynamic iteration and local aggregation.The experimental analysis shows that this method significantly improves the training efficiency and accuracy of joint fault diagnosis,and has good robustness,meeting the industrial require-ments of efficient equipment fault diagnosis in multiple factories.
关键词
设备故障诊断/联邦学习/数据异构/联合建模/分层优化/个性更新/动态迭代Key words
equipment fault diagnosis/federal study/data heterogeneity/joint modeling/hierarchical optimization/personality update/dynamic iteration引用本文复制引用
基金项目
京津冀基础研究合作专项(E2021203250)
河北省自然科学基金(F2022202021)
出版年
2024
NETL
NSTL
万方数据