【作者】 崔琳；

【导师】 赵劲松；

【作者基本信息】 北京化工大学 ， 控制理论与控制工程， 2009， 博士

【摘要】 随着石油化工的规模化、集成化、自动化的发展,以及人们对安全发展的要求不断提高,化工安全问题已经成为研究热点。作为公认的完备性最好的安全评价方法之一,危险性与可操作性(Hazard andOperability,HAZOP)分析可以比较全面而系统地识别化工过程中的安全隐患。因此,HAZOP已经成为国际石化行业安全评价的最佳实践。传统的HAZOP分析是一种依赖专家知识的人工分析,以小组会议讨论、“头脑风暴”式进行。HAZOP分析耗时长、人员劳动强度大,分析一致性和完备性难以保证。为解决上述问题,人们已经开发了多种自动化HAZOP分析计算机辅助系统,尤其是基于符号有向图(Signed Directed Graph,SDG)模型的方法取得了相当多研究成果。但是由于SDG模型知识表达能力不足,这些专家系统的HAZOP分析结果存在正确识别率不高的问题,因此尚未被工业用户广泛接受。针对以上问题本文在自动HAZOP分析方面做了大量研究,按照模型表达、模型推理和系统集成的顺序,自底向上地分层次提出新的理论和方法,构建了对应的自动分析原型系统。文中使用两个工业过程的实例,通过对比专家分析结果与自动分析结果,验证了所提出的方法的可行性和正确性。主要取得的研究成果如下:1、针对现有各种定性模型知识表达能力不足的问题,本文提出了新的定性因果模型——分层有向图(Layered Directed Graph,LDG)模型,该模型使用有向图来表达专家知识,在标准有向图基础上引入了称为“层”的维度,大大扩展了模型的表达能力,能够表达HAZOP分析中的所有知识,解决了模型容量的问题。LDG模型的构成、要素,建模方法在文中做了详细论述。为表达化工过程中实际存在的种种复杂的因果关系,LDG模型方法中提出了称为“条件因果关系”的要素。为实现知识的重用,提高LDG模型的建模效率,本文提出了建立可扩充、可继承的LDG模型库的策略,并给出了与LDG模型相关的存储、表示等具体实现。2、对应提出的LDG模型,本课题研究开发了LDG模型推理机。该推理机使用LDG模型,能够进行跨模型的定性逻辑推理,自动得到全化工流程的HAZOP分析报告。文中对LDG模型推理的算法做了详细论述,包括备选基本算法介绍,算法比较,算法的修正,多模型互联推理算法,推理结果的排序筛选算法等。推理机使用了模块化结构,由核心推理算法和外围算法处理模块构成,有效解决了复杂条件的推理,异常物料处理,设备、流程定量信息的应用,以及控制回路、安全措施对分析结果的修正等多种推理问题。文中对推理机各个模块的算法亦做了详细讨论。通过对该推理机的性能测试和分析,本文给出了该推理机的可用性数据。文中给出了对应一个实际化工过程的一组LDG模型,并使用推理机对其执行推理,对比得到的LDG分析、SDG分析的结果及人工分析的结果,说明了LDG模型推理的优势。3、本文提出了将LDG模型系统与外部CAD/CAM系统集成的系统框架,并开发了一个LDG原型专家系统,在该框架下LDG模型推理系统可通过集成从其他系统中自动获取分析所需的数据,可以减少人员劳动,降低错误率。该系统框架应用模块化结构,将不同CAD/CAM系统的差异封装形成统一的跨平台、跨语言的接口供LDG模型系统使用。文中以与一种化工CAD软件Intergraph SmartPlant P&ID的集成为例,展示了该框架各个模块的详细构成,说明了从CAD/CAM系统中获取数据,分类数据,数据匹配等数据处理算法,测试了数据处理的运行效率。并以一个实际流程为例,说明了该LDG原型专家系统进行的从工艺管道及仪表流程图(Piping and Instrument Diagram,P&ID)到HAZOP分析报告的全部过程。更多还原

【Abstract】 With the development of petrochemical process in scaling up, integration and automation,and the continuously increasing requirements for safe development,chemical process safety has been a hot research area.As one of the best approaches for process safety analysis,hazard and operability(HAZOP) analysis can thoroughly and systematically identify the potential hazards that exist in a chemical process.Therefore, it has been the best practice in the PHA of chemical process industry (CPI).Traditionally,HAZOP is a "brainstorming" activity that is done by an expert team consisting of 6～8 people coming from different areas such as process design,process instrumentation,maintenance and safety management.Due to its time-consuming and effort-consuming nature,the consistency and thoroughness of the analysis results are hardly guaranteed.To address the above issue,there has been an extensive motivation for automating HAZOP since the beginning of late 1980s. Among these efforts,signed directed graph(SDG) model based expert systems for HAZOP analysis are the most successful.Due to their shortcomings in knowledge representation that has hardly been recognized,however,they have not been widely deployed or accepted in the CPI.In this thesis,a new modeling methodology,the relative deduce algorithm and an integrated system framework are presented,consisting a complete resolvent of the automatic HAZOP study problem.A prototype system implementing the above framework is developed,and the correctness and completeness of the approach are checked.The research results of this study are listed below:1) Aiming at the drawback existing in the qualitative models mentioned above,a new qualitative model named as "Layered Directed Graph(LDG)" is presented in this paper.It uses digraph to express the professional knowledge.Based on the standard digraph,a new dimension named as "layer" is introduced to the model,which extends the model’s capability of knowledge expression.The LDG model’s structure,essential elements are presented in this paper,while the model construction methods are discussed in detail.A mechanisim named as "conditional cause-effect" is presented to express the complicated causality existing in chemical process.A strategy of building LDG model library is also presented.The knowledge within the model library is inheritable and reusable,which makes the construction of new model more easy and effective.2) Based on the LDG model methodology,a LDG model reasoning machine is developed,which can utilize the LDG models,perform cross-model casual reasoning and generate whole HAZOP report automatically.The LDG reasoning algorithms are detailed in the paper, including the introductions of the algorithm candidates,the comparison among algorithms,the modification of the original algorithms,the algorithm of cross-model reasoning using multi-model and the algorithm of sorting reasoning results.The reasoning machine is modularized, consisting of one core reasoning algorithm module and 4 appendix algorithm modules,which are all discussed in the paper.The framework of the reasoning machine can cope with various reasoning demands,such as the complicated conditional reasoning,the interaction with abnormal chemicals,the utilization of quantitative information about the process, the influences of control loops and safety equipments with in the process and so on.The reasoning machine’s performance is tested,and the results show that the reasoning machine is practicable in most scenario.The advantages of LDG model methodology is proved by comparison of the reasoning results deduced by LDG model,SDG model and human experts.3) To improve the data preparation’s efficiency,this paper presented a framework of integrating LDG model system with external CAD/CAM systems.The framework has a modularized structure,consist of 5 modules.The different CAD/CAM systems are packaged in a uniformed interface,so that LDG model system can access the data without considering the original data form.A CAD system--Intergraph SmartPlant P&ID is integrated with LDG model system in ths paper,as the implementation of the framework.Each modules of the framework are detailed and the relative data manipulating algorithms,such as data getting,data sorting,data fitting,are also discussed.The performance of data processing is tested.The whole workflow of performing automatic HAZOP analysis,from Piping and Instrument Diagram(P&ID) to HAZOP report,performed by the integrated LDG expert system is presented at the end of paper.更多还原