基于数字孪生的炼化关键装备虚实融合诊断机制与早期预警方法研究

The harsh operating environment and complex structure of key equipment in petrochemical industry pose limitations on the number and installation locations of physical sensors to timely acquire the equipment status, and also cause the difficulty of effective fault symptoms detection and early fault diagnosis and prognosis. Thus, the equipment presents high operational risk and downtime rate. To improve fault diagnosis and prognosis, flue gas turbine, a key equipment in petrochemical industry, is selected as the research object. This project starts with analysis of cyber-physical fusion and studies the multi-physics mixed modeling and dynamic update mechanism, then puts forward ubiquitous sensing technique based on real-time synchronization and faithful mapping, studies accurate twin data-driven fault diagnosis models and fault cause tracing analysis methods, establishes long-term fault warning and prediction model using the fusion and interaction of physical entity and virtual models. This project focuses on the multi-physics mixed modeling and dynamic update mechanism, twin data-driven fusion analysis methods, the fusion and interaction mechanism of physical entity and virtual models. The fundamental framework of fault diagnosis and prognosis based on the fusion of physical entity and virtual model under complex operating conditions is then constructed and validated via fault simulation, experimental studies and theoretical analysis. It is anticipated to provide a new idea and new method for effective monitoring data acquisition and fault diagnosis and prognosis models based on the fusion of physical entity and virtual models. It is of significance to make maintenance strategy and ensure the equipment safety in oil and gas industry.

本课题以故障频发的炼化关键装备-烟气轮机为研究对象，针对工况严苛、结构复杂、系统耦合导致的有效监测信息获取难、故障征兆发现难和早期故障诊断预测难等问题，引入数字孪生技术，从信息物理系统融合分析角度入手，解决多因素耦合下多物理场孪生模型实时映射机制、虚实数据同步驱动的多源信息融合与高效挖掘、面向故障预测的虚实模型交互融合机理等关键科学问题，研究多物理场作用下数字孪生建模方法，提出基于实时映射孪生模型的智能感知策略、研究虚实数据驱动的精准诊断与溯源分析方法，构建虚实融合的长周期预测预警模型。最终通过故障仿真、模拟实验和理论分析，构建严苛工况下炼化关键装备虚实融合智能诊断机制与早期预警基础性研究框架。项目预期为严苛工况下有效监测数据获取及虚实融合诊断预测模型提供新思路、新方法，对制定科学的维修决策、保障设备安全可靠运行具有重要的理论和现实意义。

针对烟气轮机中存在的样本小、非线性、运行工况复杂、有效监测数据获取困难等问题，本项目基于数字孪生技术将物理空间中的物理实体在信息空间中进行全要素重建，形成具有感知、分析、执行能力的数字孪生体，为烟气轮机安全监测与智能诊断难题的解决提供了新思路。首先根据烟气轮机的生产工艺过程，合理划分烟气轮机系统的功能和边界，结合现场资料及数据对烟气轮机各功能系统各部件进行故障模式分析和风险等级评定，进而基于多物理场耦合理论完成烟气轮机典型部件的三维立体数字孪生物理模型构建。其次，为使数字孪生模型能够实时反映实际空间中的物理系统，提出了一种基于多参数敏感度分析的模型更新方案，将孪生模型模拟的动态响应和物理系统间所测量的瞬态响应之差最小值视为目标函数，利用试验获得的多参数样本，重构能够反映修正参数与目标变量关系的响应面并迭代优化，实现有限元模型参数的更新。再者，利用构建的实时映射模型分析烟气轮机故障与信号表征间的逻辑映射关系，结合烟气轮机数据来源多样、工况多变等情况，分别提出了基于多模态深度学习的多源信息融合诊断模型以及基于增强多分辨率深度卷积网络的智能诊断模型，全面提升模型性能以及模型的泛化能力，并增强数据融合的可解释性，解决特征提取-数据融合-诊断决策难以联合优化的问题。最后，基于数字孪生模型研究多物理场耦合作用下设备性能退化过程以及故障动态演化过程，提出基于物理引导神经网络模型的烟气轮机部件劣化趋势预测方法，实现烟气轮机退化状态的精准估计与早期预警。