CN102637019B

Movatterモバイル変換

Info

Publication number: CN102637019B
Application number: CN201110035484.XA
Authority: CN
Inventors: 潘炼; 钦小平; 刘晓鸣; 王薇; 陈城; 罗杰; 彭鑫; 李柯; 严文; 徐辉; 卢伟
Original assignee: Wuhan University of Science and Technology WHUST
Current assignee: Wuhan University of Science and Technology WHUST
Priority date: 2011-02-10
Filing date: 2011-02-10
Publication date: 2014-04-02
Anticipated expiration: 2031-02-10
Also published as: CN102637019A

Abstract

Translated fromChinese

本发明涉及一种工业生产过程智能集成故障诊断方法，其特征在于，包括以下步骤：工业生产过程数据采集；根据采集到的信号进行对象特征的分析和处理；按照智能集成方法，结合专家知识，进行高炉故障诊断分析，对故障进行识别，寻找故障原因，进行故障准确定位和诊断决策，有效地调节生产过程，从而使工业生产过程能正常进行。其中智能集成诊断方法包含以下步骤：贝叶斯网络模型的建立；FTA和FMEA模型的综合分析和处理；神经网络专家系故障诊断分析和处理。同时还涉及到采用一种工业生产过程智能集成故障诊断装置，用以实现所提供的故障诊断方法。根据本发明方法，融合各种信息，在复杂情况下进行推理，能有效地对工业生产过程故障进行综合诊断，提高了故障诊断系统的集成性、智能性、准确性和有效性，保障生产过程的顺利进行。

The invention relates to an intelligent integrated fault diagnosis method for industrial production process, which is characterized in that it comprises the following steps: collecting industrial production process data; analyzing and processing object characteristics according to the collected signals; according to the intelligent integration method, combined with expert knowledge, Carry out blast furnace fault diagnosis and analysis, identify the fault, find out the cause of the fault, carry out accurate fault location and diagnosis decision-making, and effectively adjust the production process, so that the industrial production process can go on normally. The intelligent integrated diagnosis method includes the following steps: the establishment of the Bayesian network model; the comprehensive analysis and processing of the FTA and FMEA models; the fault diagnosis analysis and processing of the neural network expert system. At the same time, it also involves adopting an intelligent integrated fault diagnosis device for industrial production process to realize the provided fault diagnosis method. According to the method of the present invention, all kinds of information are fused, reasoning is carried out in complex situations, and the faults in the industrial production process can be effectively diagnosed comprehensively, the integration, intelligence, accuracy and effectiveness of the fault diagnosis system are improved, and the production process is guaranteed. went smoothly.

Description

Translated fromChinese

一种工业生产过程智能集成故障诊断方法及装置An intelligent integrated fault diagnosis method and device for industrial production process

技术领域technical field

本发明涉及测控技术中故障诊断技术，尤其涉及故障诊断中的智能集成诊断方法，并且更具体地涉及工业生产过程智能集成故障诊断方法及实现装置。The invention relates to fault diagnosis technology in measurement and control technology, in particular to an intelligent integrated fault diagnosis method in fault diagnosis, and more particularly to an intelligent integrated fault diagnosis method and a realization device in an industrial production process.

背景技术Background technique

随着现代工业和科学技术的发展，由于产品质量改善、产量提高、节约能源、防止环境污染的需要，对于连续、大批量的现代化生产过程迫切需要建立监控系统对生产过程控制系统进行故障诊断，实时检测出系统发生的故障，并对故障原因、故障频率和故障危害程度进行分析、判断，得出结论，采取必要的措施，防止灾难性事故的发生，保障生产的稳定运行，从而切实提高工业企业的经济效益。With the development of modern industry and science and technology, due to the needs of improving product quality, increasing output, saving energy, and preventing environmental pollution, it is urgent to establish a monitoring system for continuous and large-scale modern production processes to diagnose faults in the production process control system. Real-time detection of system failures, analysis and judgment on the causes, frequency and hazards of failures, drawing conclusions, and taking necessary measures to prevent catastrophic accidents and ensure the stable operation of production, thereby effectively improving industrial economic benefits of the enterprise.

现有的故障诊断方法主要有：基于规则的诊断、基于实例的诊断、模糊诊断等，各种诊断方法在诊断知识获取、诊断结论的可靠性，以及解释能力上具有各自的特点。Existing fault diagnosis methods mainly include: rule-based diagnosis, case-based diagnosis, fuzzy diagnosis, etc. Each diagnosis method has its own characteristics in the acquisition of diagnostic knowledge, the reliability of diagnostic conclusions, and the ability to explain.

基于规则的诊断方法是通过专家诊断经验的积累而建立的。这些经验由规则形式描述，将征兆与潜在的故障联系起来该诊断方法。专利文献：一种建立网络故障诊断规则库的方法(CN100393048C)、一种故障诊断方法及系统(CN101848477A)和基于分类式规则的空调系统故障诊断方法(CN1967077A)，通过建立网络故障诊断规则库、知识库进行故障诊断。该方法存在着知识获取困难，知识间上、下文敏感，不确定推理及自适应能力差等问题。在诊断方面，存在的问题是对生产系统的依赖性强，即每一种新生产系统都需要一组新规则，而积累这些规则是需要很长时间的，知识成了经验的封装。在故障诊断之前，必须要有相当多的经验规则，否则无法进行诊断。如果诊断经验不多，不仅会出现新的情况，还会发生漏诊现象。The rule-based diagnostic method is established through the accumulation of expert diagnostic experience. These experiences are described in a rule-based form, linking symptoms to potential failures as a diagnostic method. Patent literature: A method for establishing a network fault diagnosis rule base (CN100393048C), a fault diagnosis method and system (CN101848477A), and a classification rule-based air conditioning system fault diagnosis method (CN1967077A), by establishing a network fault diagnosis rule base, Knowledge base for troubleshooting. This method has problems such as difficulty in knowledge acquisition, sensitivity between context and context, uncertain reasoning and poor self-adaptive ability. In terms of diagnosis, the problem is that it is highly dependent on the production system, that is, each new production system requires a new set of rules, and it takes a long time to accumulate these rules, and knowledge becomes the encapsulation of experience. There must be quite a few rules of thumb before troubleshooting, otherwise the diagnosis cannot be made. If there is not much diagnostic experience, not only new situations will appear, but also missed diagnosis will occur.

基于实例的诊断是一种使用过去的经验实例指导解决新问题的方法，该诊断方法的关键，是如何建立一个有效的实例索引机制与实例组织方式。专利文献：一种基于PDA故障诊断系统及方法(CN101387582A)，主要针对飞行系统采用典型案例法进行故障分析，该方法能搜集到的诊断实例是有限的，不可能覆盖所有解空间，搜索时也可能会漏掉最优解，当出现异常征兆时，由于找不到最佳匹配，可能造成误诊或漏诊，产生严重后果。此外，故障实例之间的一致性维护困难，改写新实例需更多的知识。这样建立起来的故障诊断系统抗干扰能力弱，安全性难以保障。Case-based diagnosis is a method that uses past experience and examples to guide and solve new problems. The key to this diagnosis method is how to establish an effective instance index mechanism and instance organization. Patent Document: A PDA-based fault diagnosis system and method (CN101387582A), which mainly uses the typical case method for fault analysis of the flight system. The diagnostic examples that can be collected by this method are limited, and it is impossible to cover all solution spaces. The optimal solution may be missed. When an abnormal symptom occurs, because the best match cannot be found, it may cause misdiagnosis or missed diagnosis, resulting in serious consequences. In addition, it is difficult to maintain consistency between faulty instances, and rewriting new instances requires more knowledge. The fault diagnosis system established in this way has weak anti-interference ability and difficult to guarantee the safety.

基于模糊理论的诊断方法，由于模糊语言变量接近自然语言，知识的表示可读性强，模糊推理逻辑严谨，类似人类思维过程，易于解释。但模糊诊断知识获取困难，尤其是故障与征兆的模糊关系较难确定，且系统的诊断能力依赖模糊知识库，学习能力差，容易发生误诊或漏诊。The diagnostic method based on fuzzy theory, because the fuzzy language variables are close to natural language, the representation of knowledge is highly readable, the logic of fuzzy reasoning is rigorous, similar to the human thinking process, and easy to explain. However, it is difficult to obtain fuzzy diagnostic knowledge, especially the fuzzy relationship between faults and symptoms is difficult to determine, and the diagnostic ability of the system depends on the fuzzy knowledge base, and the learning ability is poor, which is prone to misdiagnosis or missed diagnosis.

由于目前各种故障诊断方法各自存在着局限性，对于工业生产过程这种复杂对象的故障诊断，如用单一的知识表示方法，有时很难完整地表示对象的故障诊断知识领域，而集成多种知识表示方法则能更好地表示对象的故障诊断领域知识。基于贝叶斯网络模型、FTA故障树分析模型、FMEA故障模式影响分析模型、神经网络模型、专家系统的集成诊断方法能综合各种诊断方法的特点，克服各诊断方法的局限性，从而提高诊断系统的智能性和诊断效率，集成型的故障诊断系统还能有效地处理真值维护、结论解释、机器学习等。智能集成故障诊断技术是故障诊断领域的一个重要发展方向。Due to the limitations of various fault diagnosis methods at present, for the fault diagnosis of complex objects such as industrial production processes, if a single knowledge representation method is used, sometimes it is difficult to completely represent the fault diagnosis knowledge domain of the object, and the integration of multiple The knowledge representation method can better represent the fault diagnosis domain knowledge of the object. The integrated diagnosis method based on Bayesian network model, FTA fault tree analysis model, FMEA failure mode impact analysis model, neural network model and expert system can synthesize the characteristics of various diagnosis methods, overcome the limitations of each diagnosis method, and improve diagnosis. The intelligence and diagnostic efficiency of the system, the integrated fault diagnosis system can also effectively handle truth value maintenance, conclusion interpretation, machine learning, etc. Intelligent integrated fault diagnosis technology is an important development direction in the field of fault diagnosis.

目前故障诊断的装置多采用常规仪表和设备以及传统的工业控制器，这些传统常规的模拟信号诊断装置功能结构单一，可靠性、互换性差，维护困难，影响了故障诊断的效果。本发明采用全数字化的FCS现场总线控制系统进行工业生产过程的故障诊断，由于该系统是-个分布式的网络控制系统，功能综合、可靠性高、互换性好、抗干扰能力强、维护容易、安装使用费用低等特点，能有效地提高故障诊断效率，系统达到较好的诊断控制水平。At present, the fault diagnosis devices mostly use conventional instruments and equipment and traditional industrial controllers. These traditional analog signal diagnosis devices have single function structure, poor reliability and interchangeability, and difficult maintenance, which affects the effect of fault diagnosis. The present invention adopts the all-digital FCS fieldbus control system for fault diagnosis in the industrial production process. Since the system is a distributed network control system, it has comprehensive functions, high reliability, good interchangeability, strong anti-interference ability, and easy maintenance. Easy installation and low cost of installation and use can effectively improve the efficiency of fault diagnosis, and the system can achieve a better level of diagnosis and control.

该方法具有重要的应用价值，它能更有效地减少故障误报或漏报的现象，及时采取措施进行相应地调整，保障工业生产过程安全有效地运行。This method has important application value. It can more effectively reduce the phenomenon of false alarm or missing alarm, and take timely measures to adjust accordingly, so as to ensure the safe and effective operation of the industrial production process.

发明内容Contents of the invention

本发明要解决的技术问题是，针对工业生产过程的复杂性、不确定性等特点，及现有诊断技术和方法存在的局限性，用准确、可靠的方式进行工业生产过程故障诊断，为正常的工业生产过程提供有效地保障。The technical problem to be solved by the present invention is to carry out fault diagnosis of industrial production process in an accurate and reliable manner in view of the characteristics of the complexity and uncertainty of the industrial production process, and the limitations of existing diagnostic techniques and methods. Provide effective protection for the industrial production process.

为实现上述目的，本发明提供了一种工业生产过程故障诊断方法，其特征在于包括以下步骤：To achieve the above object, the invention provides a method for fault diagnosis of industrial production process, which is characterized in that it comprises the following steps:

生产过程故障诊断数据的检测、信号采集等工作；Detection of fault diagnosis data in the production process, signal acquisition, etc.;

根据采集到的信号进行对象特征的分析和处理；Analyze and process object characteristics according to the collected signals;

按照智能集成方法，根据对象特征，进行生产过程故障诊断分析，对故障进行识别，寻找故障原因，进行故障准确定位，并进行诊断决策，有效地进行系统调节，从而使工业生产过程能顺利进行。According to the intelligent integration method, according to the characteristics of the object, the fault diagnosis and analysis of the production process are carried out, the fault is identified, the cause of the fault is found, the fault is accurately located, and the diagnosis decision is made, and the system is adjusted effectively, so that the industrial production process can be carried out smoothly.

其中智能集成诊断方法包含以下步骤：Wherein the intelligent integrated diagnosis method includes the following steps:

贝叶斯网络模型的建立；The establishment of Bayesian network model;

FTA(故障树分析)和FMEA(故障模式影响分析)模型的综合分析和处理；Comprehensive analysis and processing of FTA (Fault Tree Analysis) and FMEA (Failure Mode Effect Analysis) models;

神经网络模型故障诊断分析和处理；Neural network model fault diagnosis analysis and processing;

专家系统故障诊断分析和处理。Expert system fault diagnosis analysis and processing.

智能集成故障诊断技术方案：整个网络诊断系统由贝叶斯网络模型、FTA和FMEA故障影响分析模型和神经网络模型构成，用以完成模型的建立、故障分析和逻辑推理过程。这里引入贝叶斯网络模型是为了解决工业生产过程的不确定因素，提高工业生产过程故障诊断的准确性；采用FTA和FMEA技术，可以帮助系统在故障诊断早期确定导致软件失效的模块，缩小故障定位的范围。此外，贝叶斯网络技术在推理机制和系统状态上与FTA和FMEA有很大的相似性，同时还可以提高它们的描述能力，通过表达不确定、随机事件之间的概率关系获得更多有益的结论，为故障诊断和定位提供依据。Intelligent integrated fault diagnosis technology solution: the entire network diagnosis system is composed of Bayesian network model, FTA and FMEA fault impact analysis model and neural network model to complete the process of model establishment, fault analysis and logical reasoning. The Bayesian network model is introduced here to solve the uncertain factors in the industrial production process and improve the accuracy of fault diagnosis in the industrial production process; the use of FTA and FMEA technology can help the system determine the module that causes software failure in the early stage of fault diagnosis and reduce the fault The range of targeting. In addition, Bayesian network technology has great similarities with FTA and FMEA in the reasoning mechanism and system state, and can also improve their description ability, and gain more benefits by expressing the probability relationship between uncertain and random events The conclusion provides basis for fault diagnosis and location.

这里FMEA故障模式影响分析法是分析系统中每一个模块、组建所有可能产生的故障模式及其对系统造成的所有可能影响的一种归纳方法；FTA故障树分析法是用于表明系统中哪些模块有故障、外部事件或者它们的组合导致系统发生故障的逻辑关系。单独使用FTA和FMEA都各有弊端，只利用FMEA会加大工作量，而且容易遗漏故障模式和故障影响，只利用FTA又容易遗漏造成顶事件发生的底事件。因此，为避免单独使用FTA和FMEA所带来的弊端，将FTA和FMEA分别与贝叶斯网络技术结合，化为贝叶斯网络结构模型进行故障诊断，可以提高故障诊断效率。诊断模型首先通过基于贝叶斯网络的FTA和FMEA综合分析技术对故障现象进行分析，将故障初步定位到某一模块，然后运用神经网络专家系统诊断技术实现故障进一步准确定位，完成故障诊断过程。Here, the FMEA failure mode effect analysis method is an inductive method for analyzing each module in the system, forming all possible failure modes and all possible impacts on the system; the FTA fault tree analysis method is used to indicate which modules in the system A logical relationship in which a fault, an external event, or a combination thereof causes the system to fail. Using FTA and FMEA alone has its own disadvantages. Only using FMEA will increase the workload, and it is easy to miss the failure mode and failure effect. Only using FTA is easy to miss the bottom event that caused the top event. Therefore, in order to avoid the disadvantages caused by using FTA and FMEA alone, combining FTA and FMEA with Bayesian network technology and transforming it into a Bayesian network structure model for fault diagnosis can improve the efficiency of fault diagnosis. The diagnosis model first analyzes the fault phenomenon through the comprehensive analysis technology of FTA and FMEA based on Bayesian network, and initially locates the fault to a certain module, and then uses the neural network expert system diagnosis technology to realize the further accurate positioning of the fault and complete the fault diagnosis process.

我们采用神经网络通过对初级知识进行学习来处理过程的底层故障，专家系统则根据知识库储存的目标级知识和中间级知识对过程的高层故障进行诊断。We use the neural network to deal with the low-level faults of the process by learning the primary knowledge, and the expert system diagnoses the high-level faults of the process according to the target-level knowledge and intermediate-level knowledge stored in the knowledge base.

整个集成系统充分利用模型内故障诊断方法各自的优点进行互补，从而有效地克服了它们各自存在的不足，具有较好的的诊断效率和可靠性。The entire integrated system makes full use of the advantages of the fault diagnosis methods in the model to complement each other, thereby effectively overcoming their respective shortcomings, and has better diagnostic efficiency and reliability.

本发明采用一种工业生产过程故障诊断的装置，该装置现场总线控制系统FCS是一个网络化控制系统，用于实现智能集成故障诊断方法，它包括：The present invention adopts a kind of device of industrial production process fault diagnosis, and the field bus control system FCS of this device is a networked control system, is used for realizing intelligent integrated fault diagnosis method, and it comprises:

FD现场总线仪表和设备，用于故障诊断数据信号的输入、输出、运算、控制和通信，并提供功能块，以便在现场总线上构成控制回路；FD fieldbus instruments and equipment are used for the input, output, calculation, control and communication of fault diagnosis data signals, and provide function blocks to form a control loop on the fieldbus;

FBI现场总线接口，用于下接FNET现场总线网络、上接SNET监控网络，从而实现FNET现场总线网络和SNET监控网络的相互连接；The FBI field bus interface is used to connect the FNET field bus network and the SNET monitoring network, so as to realize the mutual connection between the FNET field bus network and the SNET monitoring network;

FNET现场总线网络，用于下接FD现场总线仪表和设备、上接FBI现场总线接口，进行现场信号和控制信号的相互交换；The FNET field bus network is used to connect FD field bus instruments and equipment, connect to FBI field bus interface, and exchange field signals and control signals;

SNET监控网络，用于联接FBI现场总线接口、OS操作员计算机、ES工程师控制机、SCS故障诊断监控计算机和CG计算机网关，进行信号传输；SNET monitoring network is used to connect FBI field bus interface, OS operator computer, ES engineer control computer, SCS fault diagnosis monitoring computer and CG computer gateway for signal transmission;

OS操作员计算机，用于现场工艺操作员对生产过程进行监视、操作和管理，进行人机对话；OS operator computer, used for on-site process operators to monitor, operate and manage the production process, and conduct man-machine dialogue;

ES工程师控制机，用于现场工程师对FCS进行系统生成和故障诊断维护，提供控制工程师进行控制回路组态编程和特殊应用软件的编制；ES engineer control machine is used for field engineers to perform system generation and fault diagnosis and maintenance for FCS, and provides control engineers for control loop configuration programming and special application software compilation;

SCS故障诊断监控计算机，用于故障诊断模型的建立、数据库和知识库的建立和更新维护、对生产过程进行故障诊断、预报和分析，保证安全生产；SCS fault diagnosis monitoring computer is used for establishment of fault diagnosis model, establishment, update and maintenance of database and knowledge base, fault diagnosis, forecast and analysis of production process, to ensure safe production;

CG计算机网关，用于连接监控网络和生产管理网络，实现它们之间的相互通信。The CG computer gateway is used to connect the monitoring network and the production management network to realize the mutual communication between them.

在故障诊断装置现场总线控制系统FCS运行时，首先通过FD现场总线仪表和设备采集故障诊断所必须的诊断信息和故障征兆，通过FBI现场总线接口和SNET监控网络将信息上传到故障诊断监控计算机，由其根据诊断知识，利用智能集成故障诊断模型进行推理，根据存储在网络权值上的底层故障知识完成对检测数据的故障分析，以故障征兆作为结果的输出；推理机接收到网络诊断系统的底层故障诊断结果后，结合知识库中的目标级知识，根据推理策略进行推理诊断，得到高层故障诊断的预测结论，如果预测故障被确认，由初级知识指导推理机从预测结论出发，利用目标级知识根据推理策略进行推理诊断，若推理得出的故障征兆与网络诊断模型的故障征兆一致，则高层故障被确认并被送到解释处理系统，解释处理系统根据初级知识指出故障可能出现的原因和故障发生的部位，并给出处理故障的相应方法，通过工程师控制机ES来处理相应的调整策略；如果不一致或遇到新的故障征兆而在知识库中无法找到匹配规则，则说明过程可能发生疑难故障，则启动推理机中的疑难故障诊断方法；最后，诊断结果由解释系统输出至用户接口。如果经过疑难故障诊断方法的诊断结果与实际检测结果不符，说明知识库现有的知识不能满足诊断的要求，需要补充新的知识，转入新信息获取程序，补充新的信息后重新启动诊断程序，以便在新增信息支持下达到正确诊断的目的，从而保证系统对疑难故障诊断的正确率。When the field bus control system FCS of the fault diagnosis device is running, the diagnostic information and fault symptoms necessary for fault diagnosis are firstly collected through the FD field bus instrument and equipment, and the information is uploaded to the fault diagnosis monitoring computer through the FBI field bus interface and SNET monitoring network. According to the diagnostic knowledge, it uses the intelligent integrated fault diagnosis model to perform inference, completes the fault analysis of the detection data according to the underlying fault knowledge stored in the network weight, and takes the fault symptom as the output of the result; the inference engine receives the network diagnosis system After the low-level fault diagnosis results, combined with the target-level knowledge in the knowledge base, reasoning and diagnosis are carried out according to the reasoning strategy, and the prediction conclusion of the high-level fault diagnosis is obtained. The knowledge is reasoned and diagnosed according to the reasoning strategy. If the reasoned fault symptom is consistent with the fault symptom of the network diagnosis model, the high-level fault is confirmed and sent to the interpretation processing system. The interpretation processing system points out the possible cause and The position where the fault occurs, and the corresponding method to deal with the fault is given, and the corresponding adjustment strategy is handled by the engineer control machine ES; if there is an inconsistency or a new fault symptom is encountered and the matching rule cannot be found in the knowledge base, it means that the process may occur If there is a difficult fault, start the difficult fault diagnosis method in the inference engine; finally, the diagnosis result is output to the user interface by the interpretation system. If the diagnosis result of the difficult fault diagnosis method is inconsistent with the actual detection result, it means that the existing knowledge in the knowledge base cannot meet the diagnosis requirements, and new knowledge needs to be supplemented, and the new information acquisition program is transferred to, and the diagnostic program is restarted after adding new information. , in order to achieve the purpose of correct diagnosis with the support of new information, so as to ensure the correct rate of system diagnosis of difficult faults.

根据本发明的方法，可以对工业生产过程中间状况进行逻辑推理，判断工业生产过程异常状况和变化发展规律，能更有效地对工业生产动态过程的故障信号进行有效的处理，分析判断故障信号与原因之间的相互关系，避免出现故障的错判和漏判现象，提高系统的智能性和诊断效率，保障工业生产过程稳定地正常生产。According to the method of the present invention, it is possible to carry out logical reasoning on the intermediate state of the industrial production process, judge the abnormal state of the industrial production process and the law of change and development, and more effectively process the fault signal in the dynamic process of industrial production, analyze and judge the fault signal and The interrelationship between causes can avoid misjudgment and missed judgment of failure, improve the intelligence and diagnosis efficiency of the system, and ensure the stable and normal production of the industrial production process.

通过结合以下附图，阅读本发明实施方式的详细描述后，本发明的其他特征、特点和优点将会变得更加清楚。Other characteristics, characteristics and advantages of the present invention will become more clear after reading the detailed description of the embodiments of the present invention in conjunction with the following drawings.

附图说明Description of drawings

图1是根据本发明的工业生产过程智能集成故障诊断方法示意流程图；Fig. 1 is a schematic flow chart of an intelligent integrated fault diagnosis method for an industrial production process according to the present invention;

图2是获得图1工业生产过程智能集成故障诊断方法中贝叶斯网络模型建立流程图；Fig. 2 is the flow chart of establishing the Bayesian network model in the intelligent integrated fault diagnosis method for the industrial production process in Fig. 1;

图3是获得图1工业生产过程智能集成故障诊断方法中FTA和FMEA模型的综合诊断分析和处理的流程图；Fig. 3 is the flowchart that obtains the comprehensive diagnosis analysis and processing of FTA and FMEA model in the intelligent integrated fault diagnosis method of Fig. 1 industrial production process;

图4是获得图1工业生产过程智能集成故障诊断方法中神经网络模型故障诊断分析和处理流程图；Fig. 4 is to obtain the neural network model fault diagnosis analysis and processing flowchart in the intelligent integrated fault diagnosis method of the industrial production process of Fig. 1;

图5是获得图1工业生产过程智能集成故障诊断方法中专家系统故障诊断分析和处理流程图；Fig. 5 is to obtain the expert system fault diagnosis analysis and processing flowchart in the intelligent integrated fault diagnosis method of the industrial production process of Fig. 1;

图6示意性示出一种工业生产过程智能集成故障诊断装置——现场总线控制系统FCS，图1到图5中所示的各种方法可在该系统中实现。Fig. 6 schematically shows a field bus control system FCS, an intelligent integrated fault diagnosis device for industrial production process, in which various methods shown in Fig. 1 to Fig. 5 can be implemented.

具体实施方式Detailed ways

下面将结合附图对本发明的具体实施方式进行详细说明。Specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings.

图1是根据本发明的工业生产过程智能集成故障诊断方法示意流程图。Fig. 1 is a schematic flowchart of an intelligent integrated fault diagnosis method for an industrial production process according to the present invention.

首先，在步骤101确定被测工业生产过程对象，在步骤102对被测工业生产过程对象进行信号采集，获取相关信息；在步骤103对工业生产过程对象特征信息进行处理、状态进行识别；在步骤105结合步骤104知识库和数据库信息处理系统所提供的过程状态参数，按照智能集成型故障诊断方法对数据进行分析和处理；在步骤106对故障进行识别；在步骤107对故障进行准确判断和定位；在步骤108根据解释处理系统进行诊断决策，对系统进行有效地调节。First, determine the measured industrial production process object instep 101, carry out signal acquisition to the measured industrial production process object instep 102, obtain relevant information; process the characteristic information of the industrial production process object instep 103, and identify the state; 105 Combine the process state parameters provided by the knowledge base and database information processing system instep 104, analyze and process the data according to the intelligent integrated fault diagnosis method; identify the fault instep 106; accurately judge and locate the fault instep 107 ; Instep 108, the diagnostic decision is made according to the interpretation processing system, and the system is effectively adjusted.

图2是获得图1工业生产过程智能集成故障诊断方法中贝叶斯网络模型的建立流程图。Fig. 2 is a flow chart of establishing a Bayesian network model in the intelligent integrated fault diagnosis method for the industrial production process in Fig. 1 .

在步骤201，模型建立开始。Atstep 201, model building begins.

在步骤202，是根据过程参数变量的因果关系，确定网络的拓扑结构，变量可选取离散型或连续分布型。Instep 202, the topological structure of the network is determined according to the causal relationship of process parameter variables, and the variables can be discrete or continuous.

在步骤203，为贝叶斯网络赋值。对于根节点，要确定其先验概率；对于其它节点，则要确定条件概率。先验概率和条件概率的确定，可以根据专家信息和试验信息获得。Instep 203, assign values to the Bayesian network. For the root node, the prior probability should be determined; for other nodes, the conditional probability should be determined. The determination of prior probability and conditional probability can be obtained according to expert information and experimental information.

在步骤204，确定节点之间的因果联系，设计条件概率分布表，它是基于上一级节点不同状态的概率分布。确定根节点的先验概率和其余节点的条件概率后，就可以根据贝叶斯理论确定所有节点的无条件先验概率，在此基础上进行统计推断。Instep 204, the causal connection between nodes is determined, and a conditional probability distribution table is designed, which is based on the probability distribution of different states of upper-level nodes. After determining the prior probability of the root node and the conditional probabilities of the remaining nodes, the unconditional prior probability of all nodes can be determined according to Bayesian theory, and statistical inference can be carried out on this basis.

在步骤205，确定严酷程度的概率分布表。对于严酷程度的概率分布，要根据试验情况及对系统认识的不断深入，加以修正，以反映系统的实际情况。Instep 205, a probability distribution table of severity levels is determined. For the probability distribution of severity, it should be revised according to the test situation and the continuous deepening of the understanding of the system to reflect the actual situation of the system.

在步骤206，是信念传播和推断。这一过程主要是融合新的试验信息和专家信息，改进网络结构节点和概率分布，并进行最后的统计推断。Atstep 206, is belief propagation and inference. This process is mainly to integrate new experimental information and expert information, improve network structure nodes and probability distribution, and make final statistical inference.

在步骤207，结束。Instep 207, end.

通过信念传播，不断调整更节点的先验概率、条件概率和严酷度概率，实现贝叶斯网络的学习功能，这是进行统计推断的基础。Through belief propagation, the prior probability, conditional probability and severity probability of more nodes are constantly adjusted to realize the learning function of the Bayesian network, which is the basis for statistical inference.

确定了新的先验概率、条件概率和严酷度概率分布表后，就可利用贝叶斯网络进行统计推断。After determining the new prior probability, conditional probability and severity probability distribution table, the Bayesian network can be used for statistical inference.

图3是获得图1工业生产过程智能集成故障诊断方法中FTA和FMEA模型的综合诊断分析和处理的流程图。Fig. 3 is a flow chart of obtaining the comprehensive diagnosis analysis and processing of the FTA and FMEA models in the intelligent integrated fault diagnosis method of the industrial production process in Fig. 1 .

在步骤301，首先采集关键疑难故障信息作为顶事件进行FTA分析，形成故障树。Instep 301, the key difficult fault information is firstly collected as the top event for FTA analysis to form a fault tree.

在步骤302，通过顶事件找到中间事件。Instep 302, the middle event is found through the top event.

在步骤303，确定底事件。Instep 303, the bottom event is determined.

在步骤304，根据故障树所表达的逻辑关系在贝叶斯网络中的形式表达方式，将该故障树转化为贝叶斯网络的FTA拓扑模型，确定重要底事件。Instep 304, according to the formal expression of the logical relationship expressed by the fault tree in the Bayesian network, the fault tree is transformed into an FTA topology model of the Bayesian network, and important bottom events are determined.

在转化后的贝叶斯网络模型中，根据已假定的各底事件先验概率分布，利用故障树中某事件已发生的信息，分析各事件发生的条件概率，并选择发生故障概率最大的底事件作为重要底事件。In the transformed Bayesian network model, according to the assumed prior probability distribution of each bottom event, the conditional probability of each event is analyzed by using the information that an event has occurred in the fault tree, and the bottom with the highest failure probability is selected. events as important events.

在步骤305，运行贝叶斯网络的FMEA模型。Instep 305, the FMEA model of the Bayesian network is run.

在步骤306，围绕重要底事件展开FMEA分析，给出严酷度等级定义表，并分析故障原因、故障模式、严酷度和故障影响，将FMEA分析结果转化为CFE贝叶斯网络模型。其中，FMEA的故障原因、故障模式和故障影响分别对应于CFE模型中的原因层、故障层和影响层。Instep 306, FMEA analysis is carried out around important bottom events, a severity level definition table is given, and failure causes, failure modes, severity and failure effects are analyzed, and FMEA analysis results are converted into a CFE Bayesian network model. Among them, the fault cause, fault mode and fault effect of FMEA correspond to the cause layer, fault layer and effect layer in the CFE model respectively.

在步骤307，判断故障线索是否存在，如果存在则流程转到步骤308，否则流程转到步骤310Instep 307, it is judged whether the fault clue exists, if there is, then the process proceeds to step 308, otherwise the process proceeds to step 310

在步骤308，根据故障树分析所得的各故障模式发生的概率，利用步骤306的CFE贝叶斯网络模型和系统故障间的因果性和层次性，计算出各故障模式的发生概率以及最可能的故障模式组合发生的概率，确定最有可能发生故障的模块，将故障定位到模块并提出改进措施。Instep 308, according to the probability of occurrence of each failure mode obtained by fault tree analysis, the probability of occurrence of each failure mode and the most likely The probability of occurrence of failure mode combinations, determine the most likely failure module, locate the failure to the module and propose improvement measures.

在步骤309，运用神经网络专家系统诊断技术进行故障定位分析。Instep 309, fault location analysis is performed using neural network expert system diagnosis technology.

在步骤310，补充底事件的潜在影响，建立新的故障树模式分析，重复以上分析过程，流程转到步骤301。Instep 310 , the potential impact of the bottom event is supplemented, a new fault tree mode analysis is established, and the above analysis process is repeated, and the process goes to step 301 .

在步骤311，流程结束，实现故障定位，完成故障诊断过程。Instep 311, the process ends, and the fault location is realized, and the fault diagnosis process is completed.

对于工业生产过程，整个系统通常会被划分成不同层次多个部分，包括：原因层分析、故障层分析、影响层分析。一般来说，需要把系统低一层次FMEA(Failure Modes and EffectsAnalysis)的分析结果综合到高一层次结构，才能得到总体上的FMEA分析结果，在各个层次的分析结果之间，存在着一定的因果关系，即：低一层次系统的故障模式对高一层次的影响，就是高一层次系统的故障模式；而低一层次系统导致该故障影响的故障模式，则是高一层次系统该故障撞式的故障原因，由此上推至整个系统。通过这种迭代关系，可以将低一层次系统的分析结果纳入到高一层次系统的分析之中。For the industrial production process, the entire system is usually divided into multiple parts at different levels, including: cause level analysis, failure level analysis, and impact level analysis. Generally speaking, it is necessary to integrate the analysis results of FMEA (Failure Modes and Effects Analysis) at the lower level of the system to the higher level structure in order to obtain the overall FMEA analysis results. There is a certain cause and effect between the analysis results at each level relationship, namely: the influence of the failure mode of the lower-level system on the higher-level system is the failure mode of the higher-level system; The cause of the fault is pushed up to the whole system. Through this iterative relationship, the analysis results of the lower-level system can be incorporated into the analysis of the higher-level system.

通过贝叶斯网络的故障分析系统确定了变量间的相互关系，使对复杂系统的分析成为可能，用贝叶斯网络进行复杂系统的FMEA研究，能融合各种来源信息，在不确定条件下进行推理，提高了可信度。The fault analysis system of Bayesian network determines the relationship between variables, which makes the analysis of complex systems possible. Using Bayesian network for FMEA research of complex systems can integrate information from various sources. Under uncertain conditions Inferences are made to increase credibility.

围绕重要底事件展开FMEA分析，给出严酷度等级定义表，并分析故障原因、故障模式、严酷度和故障影响，将FMEA分析结果转化为CFE贝叶斯网络模型。其中，FMEA的故障原因、故障模式和故障影响分别对应于CFE模型中的原因层、故障层和影响层。Carry out FMEA analysis around important bottom events, give the severity level definition table, and analyze the cause of failure, failure mode, severity and failure impact, and transform the FMEA analysis results into a CFE Bayesian network model. Among them, the fault cause, fault mode and fault effect of FMEA correspond to the cause layer, fault layer and effect layer in the CFE model respectively.

CFE型贝叶斯网络的诊断决策是结合原因层各节点的先验概率和所有故障层节点在其母节点给定情况下的条件概率，依据全概率公式推理得出故障层各节点的先验概率值，当给定征兆信息时，通过贝叶斯公式进一步修正该先验概率值，得到后验概率值，从而可以确定某一节点故障发生的可能性。The diagnosis decision of CFE-type Bayesian network is to combine the prior probability of each node in the cause layer and the conditional probability of all fault layer nodes under the given conditions of their parent nodes, and derive the prior probability of each node in the fault layer according to the total probability formula. Probability value, when the symptom information is given, the prior probability value is further corrected by the Bayesian formula to obtain the posterior probability value, so that the possibility of a certain node failure can be determined.

图4是获得图1工业生产过程智能集成故障诊断方法中神经网络模型故障诊断分析和处理流程图。Fig. 4 is a flow chart of the neural network model fault diagnosis analysis and processing in the intelligent integrated fault diagnosis method of the industrial production process in Fig. 1 .

在步骤401，进行系统初始化。Instep 401, system initialization is performed.

在步骤402，给定系统输入值和输出期望目标值。Instep 402, the system input values and output desired target values are given.

在步骤403，根据输入和目标值判断系统是否学习完成，如果没有，则流程转至步骤404，如果完成，则流程进入工作状态，转至步骤408。Instep 403, it is judged whether the learning of the system is completed according to the input and the target value. If not, the process goes to step 404. If it is completed, the process enters the working state and goes to step 408.

在步骤404，计算隐含层和输出层各单元值。Instep 404, the values of each unit of the hidden layer and the output layer are calculated.

在步骤405，计算目标值和实际输出值的偏差E。Instep 405, the deviation E between the target value and the actual output value is calculated.

在步骤406，判断偏差E是否满足工艺要求，如果没有满足，则流程转至步骤410，如果满足，则流程转至步骤407。Instep 406, it is judged whether the deviation E satisfies the process requirement, if not, the flow goes to step 410, and if it does, the flow goes to step 407.

在步骤407，判断所有单元的偏差e是否满足工艺要求，如果没有满足，则流程转至步骤410，如果满足，则流程转至步骤408。Instep 407, it is judged whether the deviation e of all units meets the process requirement, if not, the process goes to step 410, and if it is satisfied, the process goes to step 408.

在步骤408，求输出向量。Instep 408, the output vector is calculated.

在步骤409，启动专家系统给出诊断结果。Instep 409, start the expert system to give the diagnosis result.

在步骤410，计算隐含层单元的误差。Instep 410, the errors of the hidden layer units are calculated.

在步骤411，调整学习率。Instep 411, the learning rate is adjusted.

在步骤412，调整中间层到输出层的连接权值和输出层单元的阈值。Instep 412, the connection weights from the intermediate layer to the output layer and the thresholds of the output layer units are adjusted.

在步骤413，调整输入层到中间层的连接权值和中间层单元的阈值。Instep 413, the connection weights from the input layer to the intermediate layer and the thresholds of the intermediate layer units are adjusted.

在步骤414，学习完成。Atstep 414, learning is complete.

该神经网络方法是先进行学习，学习完成后，启动专家系统，再给出诊断结果，其学习是有指导的学习，即人为给出输入和输出样本。The neural network method is to learn first, after the learning is completed, the expert system is started, and then the diagnosis result is given. The learning is guided learning, that is, the input and output samples are given artificially.

图5是获得图1工业生产过程智能集成故障诊断方法中专家系统故障诊断分析和处理流程图。Fig. 5 is a flow chart of expert system fault diagnosis analysis and processing in the intelligent integrated fault diagnosis method of the industrial production process in Fig. 1 .

在步骤501，启动专家系统。Instep 501, the expert system is started.

在步骤502，运行专家先验知识库。Instep 502, the expert prior knowledge base is run.

在步骤503，搜索案例库进行推理，确定故障模块，得到基于案例推理的诊断结果，如对诊断结果不满意或没有结果，再搜索规则库。Instep 503, search the case base for reasoning, determine the fault module, and obtain the diagnosis result based on case reasoning, if the diagnosis result is not satisfied or there is no result, search the rule base again.

在步骤504，搜索规则库，得到基于规则推理的诊断结果；如仍对诊断结果不满意或没有结果，再进行模型推理，得到基于模型推理的诊断结果。Instep 504, the rule base is searched to obtain a diagnosis result based on rule reasoning; if the diagnosis result is still not satisfied or there is no result, model reasoning is performed again to obtain a diagnosis result based on model reasoning.

在步骤505，基于模型进行推理。Instep 505, reasoning is performed based on the model.

在步骤506，对不同推理方法得到的诊断结果进行综合分析。Instep 506, a comprehensive analysis is performed on the diagnosis results obtained by different reasoning methods.

在步骤507，运行解释处理系统，对推理得到的确认结果进行解释。Instep 507, run the interpretation processing system to explain the confirmation result obtained by reasoning.

在步骤508，诊断结果输出。Instep 508, the diagnosis result is output.

在步骤509，结束。Instep 509, end.

601OS表示操作员计算机，602ES表示工程师控制机，603SCS表示故障诊断监控计算机，604CG表示计算机网关，605SNET表示监控网络，606FBI表示现场总线接口，607FNET现场总线网络，608表示FD现场总线仪表和设备，609表示工业生产过程。601OS means operator computer, 602ES means engineer control machine, 603SCS means fault diagnosis monitoring computer, 604CG means computer gateway, 605SNET means monitoring network, 606FBI means field bus interface, 607FNET field bus network, 608 means FD field bus instrument and equipment, 609 Represents an industrial production process.

OS操作员计算机601，用于现场工艺操作员对生产过程进行监视、操作和管理，进行人机对话。OS operator computer 601 is used for on-site process operators to monitor, operate and manage the production process, and conduct man-machine dialogue.

ES工程师控制机602，用于现场工程师对FCS进行系统生成和故障诊断维护，提供控制工程师进行控制回路组态编程和特殊应用软件的编制。ESengineer control machine 602 is used for field engineers to perform system generation and fault diagnosis and maintenance for FCS, and provides control engineers for control loop configuration programming and compilation of special application software.

SCS故障诊断监控计算机603，用于故障诊断模型的建立、数据库和知识库的建立和更新维护、对生产过程进行故障诊断、预报和分析，保证安全生产。The SCS faultdiagnosis monitoring computer 603 is used for establishment of fault diagnosis model, establishment and update maintenance of database and knowledge base, fault diagnosis, forecast and analysis for production process, to ensure safe production.

CG计算机网关604，用于连接监控网络605和生产管理网络，实现它们之间的相互通信。TheCG computer gateway 604 is used to connect themonitoring network 605 and the production management network to realize mutual communication between them.

SNET监控网络605，用于联接FBI现场总线接口和设备608、OS操作员计算机601、ES工程师控制机602、SCS故障诊断监控计算机603和CG计算机网关604，进行信号传输。SNET monitoring network 605 is used to connect FBI field bus interface andequipment 608,OS operator computer 601, ESengineer control computer 602, SCS faultdiagnosis monitoring computer 603 andCG computer gateway 604 for signal transmission.

FBI现场总线接口606，用于下接FNET现场总线网络607、上接SNET监控网络605，从而实现FNET现场总线网络607和SNET监控网络605的相互连接。The FBIfield bus interface 606 is used to connect the FNETfield bus network 607 downward and theSNET monitoring network 605 upward, so as to realize the mutual connection between the FNETfield bus network 607 and theSNET monitoring network 605 .

FNET现场总线网络607，用于下接FD现场总线仪表和设备608、上接FBI现场总线接口606，进行现场信号和控制信号的相互交换。The FNETfield bus network 607 is used to connect to the FD field bus instrument andequipment 608 and connect to the FBIfield bus interface 606 to exchange field signals and control signals.

FD现场总线仪表和设备608，包含有：传感器、变送器、执行器、仪表电源、电源适配器和安全栅等，用于故障诊断的数据信号的输入、输出、运算，故障的调节、控制和通信，并提供功能块，以便在现场总线上构成控制回路。FD field bus instruments andequipment 608, including: sensors, transmitters, actuators, instrument power supplies, power adapters and safety barriers, etc., used for input, output, calculation of data signals for fault diagnosis, fault adjustment, control and Communication, and provide function blocks to form a control loop on the fieldbus.

工业生产过程609是被诊断对象，故障诊断系统对该对象实施故障分析。Theindustrial production process 609 is the object to be diagnosed, and the fault diagnosis system performs fault analysis on the object.

技术方案：在故障诊断装置现场总线控制系统FCS运行时，首先通过FD现场总线仪表和设备608采集工业生产过程故障诊断对象609所必须的诊断信息和故障征兆，通过FNET现场总线607、FBI现场总线接口606和SNET监控网络605将信息上传到操作员计算机601、工程师控制机602和故障诊断监控计算机603，并通过CG计算机网关604，进行监控网络和生产管理网络之间相互通信。由FCS系统根据诊断知识，利用智能集成故障诊断模型进行推理，根据存储在网络权值上的底层故障知识完成对检测数据的故障分析，以故障征兆作为结果的输出；推理机接收到网络诊断系统的底层故障诊断结果后，结合知识库中的目标级知识，根据推理策略进行推理诊断，得到高层故障诊断的预测结论，如果预测故障被确认，由初级知识指导推理机从预测结论出发，利用目标级知识根据推理策略进行推理诊断，若推理得出的故障征兆与网络诊断模型的故障征兆一致，则高层故障被确认并被送到解释处理系统，解释处理系统根据初级知识指出故障可能出现的原因和故障发生的部位，并给出处理故障的相应方法，通过工程师控制机ES602来处理相应的调整策略；如果不一致或遇到新的故障征兆而在知识库中无法找到匹配规则，则说明过程可能发生疑难故障，则启动推理机中的疑难故障诊断方法；最后，诊断结果由解释系统输出至用户接口。如果经过疑难故障诊断方法的诊断结果与实际检测结果不符，说明知识库现有的知识不能满足诊断的要求，需要补充新的知识，转入新信息获取程序，补充新的信息后重新启动诊断程序，以便在新增信息支持下达到正确诊断的目的，从而保证系统对疑难故障诊断的正确率。Technical solution: When the field bus control system FCS of the fault diagnosis device is running, first collect the necessary diagnostic information and fault symptoms of thefault diagnosis object 609 in the industrial production process through the FD field bus instrument andequipment 608, and then use theFNET field bus 607 and FBIfield bus Interface 606 andSNET monitoring network 605 upload information tooperator computer 601,engineer control machine 602 and faultdiagnosis monitoring computer 603, and communicate with each other between monitoring network and production management network throughCG computer gateway 604. Based on the diagnosis knowledge, the FCS system uses the intelligent integrated fault diagnosis model to perform inference, completes the fault analysis of the detection data according to the underlying fault knowledge stored in the network weight, and takes the fault symptom as the output of the result; the reasoning machine receives the network diagnosis system After the low-level fault diagnosis results, combined with the target-level knowledge in the knowledge base, reasoning and diagnosis are carried out according to the reasoning strategy, and the prediction conclusion of the high-level fault diagnosis is obtained. High-level knowledge conducts inference diagnosis according to the inference strategy. If the fault symptom obtained by reasoning is consistent with the fault symptom of the network diagnosis model, the high-level fault is confirmed and sent to the interpretation processing system, which points out the possible cause of the fault based on the primary knowledge. and the location where the fault occurred, and give the corresponding method to deal with the fault, and handle the corresponding adjustment strategy through the engineer control machine ES602; if it is inconsistent or encounters a new fault symptom and cannot find a matching rule in the knowledge base, it means that the process may When a difficult fault occurs, the difficult fault diagnosis method in the inference engine is started; finally, the diagnosis result is output by the interpretation system to the user interface. If the diagnosis result of the difficult fault diagnosis method is inconsistent with the actual detection result, it means that the existing knowledge in the knowledge base cannot meet the diagnosis requirements, and new knowledge needs to be supplemented, and the new information acquisition program is transferred to, and the diagnostic program is restarted after adding new information. , in order to achieve the purpose of correct diagnosis with the support of new information, so as to ensure the correct rate of system diagnosis of difficult faults.

如果不同推理方法得到相似的诊断结果，则诊断结果比较可信；如果不同推理方法得的诊断结果相差较大，则需要对诊断结果进行具体分析，这时可以选择基于模型的诊断结果，或可信度较大的诊断结果作为诊断结论。If different reasoning methods obtain similar diagnostic results, the diagnostic results are more credible; if the diagnostic results obtained by different reasoning methods are quite different, specific analysis of the diagnostic results is required. In this case, model-based diagnostic results can be selected, or The more reliable diagnostic results were used as the diagnostic conclusions.

在图1到图5所示的流程图的基础上，结合本系统技术人员无需创造性的工作即可开发出更多应用软件，进行生产过程故障诊断。On the basis of the flowcharts shown in Figures 1 to 5, combined with this system, technicians can develop more application software without creative work, and carry out fault diagnosis in the production process.

智能集成诊断方法的有益效果和优点：系统集成度高，能充分发挥各种诊断方法的优势，克服各自方法存在的不足，并能最大限度地找到问题的解决方法，对故障进行准确定位，避免故障的错判和漏判现象，具有较高的诊断效率和确诊率。The beneficial effects and advantages of the intelligent integrated diagnosis method: the system is highly integrated, can give full play to the advantages of various diagnosis methods, overcome the shortcomings of each method, and can find solutions to problems to the greatest extent, accurately locate faults, and avoid Fault misjudgment and missed judgment phenomenon, with high diagnostic efficiency and diagnosis rate.