摘要
现代智能制造产线集成化程度高,产线设备结构和功能复杂,工艺流程存在耦合性,导致故障隔离定位困难,易产生故障传播风险并威胁产线安全稳定高效运行.故障根因诊断旨在基于产线实时状态信息,通过准确识别故障传播路径与源头,提升产线和关键设备的系统可靠性,提高生产运行效率.然而,现有方法多局限于因果相关性分析,缺乏深层次的因果推理能力,难以揭示故障后状态变量之间真实的因果关系,限制了故障传播分析与根因诊断的可信度.为此,本文提出一种基于深度因果推理的故障根因诊断方法 (deep causal inference-based root cause diagnosis, DCIRCD).该方法首先通过结构因果模型(structural causal model, SCM)构建监测变量间的因果关系,结合干预与反事实推理实现因果推理的3个层级,提升了监测变量状态变化的因果解释性.特别地,通过引入孪生网络简化反事实推理过程,将其转化为关联因果模型上的贝叶斯推断(Bayesian inference, BI),降低了计算复杂度.所提方法在航空机载叶轮制造产线上进行了可行性与有效性验证,基于产线的状态报警数据,DCIRCD方法能够准确诊断故障根因,其结论与历史维修记录和系统机理相符.
Modern intelligent manufacturing production lines are highly integrated,with complex equipment structures and functions,as well as coupled process flows,leading to difficulties in fault isolation and localization.This complexity increases the risk of fault propagation,threatening the safe,stable,and efficient operation of the production line.Root cause diagnosis aims to enhance system reliability and operational efficiency by accurately identifying fault propagation paths and sources based on real-time production line state information.However,existing methods are mostly limited to causal correlation analysis and lack deep causal reasoning capabilities,making it difficult to reveal the true causal relationships between state variables after faults occur.This limitation reduces the credibility of fault propagation analysis and root cause diagnosis.To address this,this paper proposes a deep causal inference-based root cause diagnosis(DCIRCD)method.This method first constructs causal relationships between monitoring variables using a structural causal model(SCM)and combines intervention and counterfactual reasoning to achieve three levels of causal inference,improving the interpretability of state changes in monitoring variables.Notably,by introducing a twin network to simplify the counterfactual reasoning process,the method transforms it into Bayesian inference on an associated causal model,reducing computational complexity.The proposed method was validated for feasibility and effectiveness on an aviation airborne impeller manufacturing production line.Based on the production line’s state alarm data,the DCIRCD method accurately diagnosed fault root causes,with results consistent with historical maintenance records and system mechanisms.
作者
陆宁云
黄守金
姜斌
李洋
Ningyun LU;Shoujin HUANG;Bin JIANG;Yang LI(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China;State Key Laboratory of Mechanics and Control for Aerospace Structures,Nanjing 211106,China;School of Mechanical and Electrical Engineering and Automation,Shanghai University,Shanghai 200444,China)
出处
《中国科学:信息科学》
北大核心
2025年第7期1673-1686,共14页
Scientia Sinica(Informationis)
基金
江苏省基础研究计划重点项目(批准号:BK20243045)
国家自然科学基金(批准号:62273176)
国家重点研发计划(批准号:2021YFB3301300)资助项目。
关键词
因果推理
反事实推理
故障根因诊断
孪生网络
causal inference
counterfactual inference
fault root cause diagnosis
twin networks
