【作者】 郑娆；

【导师】 王奎升；

【作者基本信息】 北京化工大学 ， 化工过程机械， 2010， 博士

【摘要】 化工过程中涉及多个变量、多种单元操作设备和传热、传质及化学反应等多种过程。变量之间关系复杂、过程机理复杂以及过程的多尺度都会为故障诊断带来一定的难度。为了适应化工生产过程信号的高噪声、维数高、强相关、非线性的特点,在复杂化工过程故障诊断方法的研究中,将复杂系统建模技术融入其中以及将定性模型与定量模型有机地结合都是必然的发展趋势。针对复杂化工过程故障监测与故障诊断问题,本文以实体语法系统为理论框架,建立了调控流图(RFG)、定性调控流图、调控流图定性仿真系统的形式化定义。针对复杂化工过程建模前所具备的数据基础,建立了调控流图的三种建模方法,分别为基于工艺流程图的建模、基于数学模型的建模和基于连续监测数据的建模。基于工艺流程图的建模方法可用以直接建立化工过程的定性调控流图模型,基于已有数学模型的建模方法可用以建立化工过程的定量调控流图模型,基于连续监测数据的建模方法可以结合粒子群等参数估计算法用以建立表达化工过程外在特征的定量调控流图模型。为了使定性、定量模型之间可以相互转换,本文推导了调控流图通用动力学方程,结合S-系统动力学方程,建立了两类模型互相转换的两套方法,为定性、定量相结合的复杂化工过程建模提供了基础。以CSTR系统为例,探讨了三种建模方法的具体应用。针对定性调控流图的分析问题,本文建立了调控流图的定性仿真、故障监测和故障诊断技术。并以CSTR过程、TE过程和原油蒸馏过程为例,研究了三种技术的可行性。定性仿真技术以调控流图定性仿真系统为基础,利用Datalog语言实现基于调控流图模型的自动推理。这一技术通过对变量取值空间的界定,有效解决了环的问题、多途径调节问题以及多重仿真问题,为基于调控流图的仿真奠定了基础。故障监测技术在定量调控流图模型的基础上,通过仿真数据与实际监测数据的差异进行故障监测。复杂化工过程传统数学模型由于模型复杂、计算量大,限制了这一思路的应用。本文基于连续监测数学的调控流图作为一类唯象模型,具有结构简单、计算速度快的特点,有效克服了这一瓶颈问题。故障诊断技术以定性调控流图为基础,通过调控流图的定性仿真获得每个故障的传播途径及对每个观测变量的影响,利用数据挖掘技术建立从监测变量推断故障的故障诊断模型。复杂化工过程的定性模型由于变量多、关系复杂,在反向推理过程中容易出现不确定性的结果,且计算量大,容易出现组合爆炸的问题,难以用到生产过程的实时故障诊断中。本文基于定性调控流图的故障诊断通过数据挖掘离线建立故障诊断模型,在生产过程中直接应用结构简单的故障诊断模型,具有速度快、判断直接的特点,在一定程度上解决了这一问题。在研究过程中,本文探讨了调控流图与系统动力学、SDG建模方法等相关理论与技术方法的关系,并探讨了化工过程故障诊断与复杂系统涌现性的关系。调控流图作为一种新的复杂化工过程建模方法,继承了实体语法系统的形式化特点,融合了多种方法的优点,并有效地集成了定性仿真技术、数据挖掘技术、参数估计方法,能够实现定性定量模型的相互转化,为复杂化工过程的故障监测与故障诊断提供了可行的技术方案。更多还原

【Abstract】 The chemical processes involve a wide variety of variables, process equipments and chemical reactions. Fault diagnosis of chemical process has become a big chanllenge because of complex relationships between variables and complex process mechanism. Recent trends of the study of chemical process fault diagnosis are towards complex system modeling and the integration of qualitative and quantitative models.A novel modeling method called Regulating Flow Graph (RegFlow Graph, RFG) was proposed in this dissertation in attempt to solve the problem of fault detection and diagnosis in complex chemical processes. Formal definitions of RFG, quanlitative RFG and qualitative simulation system of RFG were presented, which were based on the framework of entity grammar system.Three RFG modeling methods were developed, namely, process flow diagram-based approach (FDBA), mathematical model-based approach (MMBA) and continuous monitoring data-based approach (MDBA). FDBA can be used to build a qualitative RFG of chemical process. MMBA can be used to build a quantitative RFG. The combination of MDBA and partical swarm algorithm can be used to build a phenomenalogical model of chemical process.For the transformation between qualitative RFG and quantitative RFG, the general dynamical equation for RFG was deduced. Together with the S-system dynamics equation, two transformation methods were proposed. The applications of three RFG modeling methods in CSTR system were explored.The techniques on RFG qualitative simulation, RFG-based fault detection and RFG-based fault diagnosis were developed in the later charpters. To verify and illustrate these techniques, CSTR process, TE process and crude oil distillation process were tested as case studies.Based on the qualitative simulation systems of RFG, the automatic reasoning system of RFG was implemented using Datalog language. The problems of rings, multi-channel regulation and multiple simulation have been effectively solved.For RFG-based fault detection, faults can be identified by the difference between RFG simulated data and monitoring data. The complexity and large computional requirement of traditional mathematical models have limited their application in fault detection. The phenomenological model obtained using MDBA is simple and the caculation is fast, which can overcome the drawbacks of traditional fault detection methods. The fault disgnosis based on RFG can be accomplished using RFG qualitative simulation, which can shows the propagation path of each fault and the effect of faults on each observerd variable. The simulation results can be used to establish fault diagnosis model with data mining method. Because of the complexity of chemical processes, back-forward reasoning strategy in falut disgnosis often leads to uncertainty results and combinatorial explosion problem. The combination of RFG qualitative simulation and data ming can partly solve these problems.This dissertation has explored the relationships between RFG and the related theories and approaches, such as system dynamics and SDG modeling method. The relathionships between fault diagnosis in chemical processes and emergence in complex systems was also analysised. As a novel modeling method, RFG has the formal features of entity grammar system and contains the advantages of related modeling methods. It has overcome the limitions of traditional research methods. The integration of qualitative simulation technology, data mining technique and parameter estimation method in RFG has provided a feasible scheme for fault detection and fault diagnosis in complex chemical processes.更多还原