英文翻译格式解析

英文资料翻译

系别软件与服务外包学院. 专业通信网络与设备. 班级通信0901 . 学生姓名韩丽司. 学号090969 . 指导教师陈佳.

二○一二年二月

Based on the data fusion of intelligent fault diagnosis system 1. Primed words

Multi sensor data fusion technology was initially mostly used in the military field, but the computer, network and communication technology the rapid development that the application range is expanded greatly. In recent years, many scholars of the data fusion rules and strategy theory to conduct extensive research and improvement. While the artificial intelligence technology research makes the data fusion to improve the knowledge of target decision height, the auxiliary function is greatly strengthened. At the same time, with the industrial technology make a spurt of progress, an intelligent fault diagnosis system of demand in quantity and quality greatly improved. As the intelligent fault diagnosis system for the most basic, the most effective information processing tools, multi sensor data fusion technology development will promote the progress of intelligent fault diagnosis system.

2. Multi sensor data fusion and improved D - S theory

From a military application perspective, data fusion is to make full use of different time and space of the multi sensor information resources according to the time sequence, using computer technology to obtain multiple sensor observation information in certain criteria to be automatic, integrated analysis, control and use, access to the object consistency of interpretation and description, to complete required decision-making and estimation tasks, allowing the system to obtain than its components the better performance of H3. The author uses the present generally agree that the pixel level, feature layer and decision layer three layer fusion structure. The decision level fusion target is to achieve the target situation diagnosis and assessment, applied to the main Bayesian probability reasoning and D S evidence theory. The data fusion method to solve the uncertain information processing problems, D - S method with Dempster - Shafer evidential theory as a foundation, its core is Dempster synthesis rules, for uncertain information expression and synthesis provides natural and robust method. Will force S evidence theory is used for multi sensor fusion, obtained from the sensor related value is the theory of evidence, it can constitute the targets to be recognized patterns of belief function assignment, that each target model hypothesis of credibility, each sensor consists of an evidence of group.

Multi sensor data fusion is through D S united rules to combine several evidence group to form a new integrated evidence group, called the D S association rules with each sensor of confidence function distribution formed by fusion of confidence function distribution, which is target mode decision-making provide comprehensive and accurate information of n ]. In practical application, D - S method requires evidence of independence and evidence combination rule theory support, and the calculation of potential exists the problem of combinatorial explosion, so only the single fusion methods are difficult to obtain ideal fusion effect.

D - S evidence theory has does not require a priori probability advantages; expert

system has a problem domain knowledge; fuzzy system has higher fuzzy language processing; high order neural network has the capacity to be big, approximation ability, fault-tolerant a wide range of features, so the D s evidence theory fusion method with multiple division complementary to improve the D - s method, improve the fusion system for target identification accuracy and reliability, which make the system has strong self learning ability and ability to adapt to their environment.

3. Intelligent fault diagnosis system

Diagnosis system of a failure mode is often caused by multiple fault symptom, and a fault symptom can be caused by multiple failure modes, is many-to-many form. So without a sensor to ensure that at any time to provide complete and reliable information, it is usually in multiple sensor based on integrated diagnosis. In essence, fault diagnosis system is the use of diagnostic object system runs a variety of state information and various kinds of existing knowledge, information processing, finally get on the system operation condition and fault condition of the comprehensive evaluation of n3. Data fusion is typical application system is C3I system, especially in multiple target tracking system. According to C3I system, fault diagnosis information required for access to more diverse, describe diagnostic mathematical model of the object may be greater than the space coordinates and velocity characteristics are more complex, the fault diagnostic object link between ( coupling, backup, transfer ) can be tracked object of coordinated action of the relations to be more close, but can make the diagnosis object is regarded as a sensor through the systematic observation of the particular state space, the fault signal is the space in the specific target signal, the fault diagnosis is based on the signal and the knowledge base to determine the fault alarm.

The fault diagnosis system, very suitable for using the previously described multisensor fusion structure, a pixel layer is layer of data fusion for sensor reflect the direct data; feature layer corresponding to various fault diagnosis methods of data fusion, the results are effective decision; decision fusion for integrated subsystems via the fusion rule of combination made the final the results of fault diagnosis and troubleshooting. The three layer structure corresponding to the fault diagnosis system of monitoring, diagnosis and decision function. In fault diagnosis system Data fusion in the certain degree can make the system to obtain the accurate state estimation, increase the degree of confidence, to reduce ambiguity, improve diagnostic performance, improve the multi sensor information resources utilization. But with the development of new technology, fault diagnosis system is gradually introduced into artificial intelligence technology, the main performance is: the use of neural network local diagnosis; the use of multiple concurrent ES using multiple knowledge in the field of synthetic information; the use of advanced database management technology for decision support system using reasoning; learning, so the automatic adapt to all kinds of trend. In addition, on the basis of data fusion, the fusion levels increase, the data mining and knowledge ( including rules, method and model) fusion.

4．Based on the data fusion of intelligent diagnosis system

From the perspective of multi sensor data fusion, typical application example is the process monitoring and fault diagnosis, and from the perspective of intelligent fault diagnosis system, usually in multi sensor data fusion based on integrated diagnosis. Based on the above on the multi-sensor data fusion technology and intelligent fault diagnosis system are discussed, the following two techniques for organic coupling, based on the establishment of a multi sensor data fusion of intelligent fault diagnosis system structure frame.

4.1 Working principle

The system is composed of input output system, a sensor signal acquisition system, signal pre-processing system, expert system and decision fusion system. When the system works, the first use of multi sensor signal acquisition and signal data were preprocessed ( such as signal filtering, spectrum analysis, wavelet analysis, etc. ) will be processed information and diagnostic system of expert knowledge base ( rules, methods and models of knowledge) according to certain rules, and then each sub-system is the local diagnosis results are parallel fusion for decision fusion system for global diagnosis, the final output diagnosis results and relevant information will be stored in the database and knowledge base for the use of data mining technology for knowledge discovery for the necessary data on reserves.

4.2 Key technolog y

4.2.1 Local diagnosis system

Neural network can realize the complex nonlinear mapping, in the field of fault diagnosis has been widely used export ]. When the system parameters for the diagnosis of more, signs of the large amount of information times, due to the inevitable contradiction between sample and random, if the high dimensional symptom information input at the same time to the same network processing, will make the long training time, the diagnosis of poor results, sometimes evenTo cause the network convergence. Therefore, the human brain in different regions with different information. Different signals are also by the respective neural network diagnosis. So the high dimensional symptom space decomposition into low dimensional symptom space, the process may also be referred to as the local diagnosis. In addition, the neural network system can effectively solve the expert system part of the limitations, so the use of the neural network expert system.

4.2.2 Decision fusion

Using neural network for local diagnosis, from each or several diagnostic parameters can get their diagnostic results, each subsystem is responsible for a fault diagnosis, from different angles, fault diagnosis, decision fusion of these diagnostic results fusion, makes the subsystem is formed between the" consultation", utmost to improve the diagnosis rate. For preprocessing information fusion, inference is more important than numerical computation, should be based on knowledge of the technology of expert system and D - S theory of evidence combination method of fusion.

4.2.3 Data mining and knowledge fusion

System existing operating state to revise the original system knowledge base, can be more quickly, more accurate, more comprehensive fault diagnosis, this is the data mining and knowledge integration issues, data mining techniques in information fusion system will become the necessary part of.

5. The end

Multi sensor data fusion technology and intelligent fault diagnosis system is very practical, and the organic integration of the two can on their respective technology development to promote each other's role. But at present the information fusion system specific fusion rule method based on knowledge fusion technology is still not mature, also remains to be improved, the intelligent diagnosis system need to be improved for AI Technology application. But I believe that with all the technology and the gradual improvement of the practice, continue to accumulate experience, based on the data fusion of intelligent fault diagnosis system will be developed faster and wider application.

基于多传感器数据融合的智能故障诊断系统

1．引言

多传感器数据融合技术最初大多应用于军事领域，但计算机、网络以及通信等先进技术的飞速发展使它的应用范围得到了很大的拓展。近年来，众多学者对数据融合的规则与策略的理论进行了广泛的研究和改进。而人工智能等技术的研究使得数据融合提升到了知识融合的高度，对目标决策的辅助作用大大加强。与此同时，随着工业技术的突飞猛进，智能故障诊断系统的需求在数量上和质量上大大提高了。作为智能故障诊断系统中的最基本、最有效的信息处理工具，多传感器数据融合技术的发展将推动智能故障诊断系统的进步。

2．多传感器数据融合与改进D—S理论

从非军事应用的角度来说，数据融合是指充分利用不同时间与空间的多传感器信息资源，采用计算机技术对按时序获得的多传感器观测信息在一定准则下加以自动分析、综合、支配和使用，获得对被测对象的一致性解释与描述，以完成所需的决策和估计任务，使系统获得比它的各组成部分更优越的性能H3。笔者采用目前普遍认同的像素层、特征层以及决策层的三层融合结构。其中决策级融合的目标是实现对目标态势的诊断和评估，应用到的主要有贝叶斯概率推理和D —S证据理论等方法。这些数据融合方法都必须解决对不确定信息的处理问题，D—S方法以Dempster—Shafer证据理论为基础，其核心是Dempster合成规则，为不确定信息的表达和合成提供了自然而强有力的方法。将胁S证据理论用于多传感器融合时，从传感器获得的相关数值就是该理论中的证据，它可构成待识别目标模式的信度函数分配，表示每一个目标模式假设的可信程度，每一传感器构成一个证据组。

所谓多传感器数据融合就是通过D—S联合规则联合几个证据组形成一个新的综合的证据组，即用D—S联合规则联合每个传感器的信度函数分配形成融合的信度函数分配，从而为目标模式的决策提供综合准确的信息n]。实际应用中，D—S方法要求证据的独立性和证据合成规则的理论支持，而且计算量存在着潜在的组合爆炸问题，所以仅靠这种单一的融合方法难以获得理想的融合效果。

D—S证据理论具有不需要先验概率的优点；专家系统具有问题领域的丰富

知识；模糊系统具有较高的模糊语言处理能力；高阶神经网络具有容量大、逼近能力强、容错范围广的特点，所以将D—s证据理论与多种融合方法的分工互补能够改进D—s方法的不足，提高融合系统中的目标识别的精确性和可靠性，使得系统具有较强的自学习能力以及对外界环境的适应能力。

3．智能故障诊断系统

被诊断系统的一个故障模式往往引起多个故障征兆，而一个故障征兆又可以由多种故障模式引起，是多对多的形式。所以没有一种传感器能够保证在任何时候提供完全可靠的信息，因此通常都是在多传感器的基础上进行综合诊断。本质上，故障诊断系统是利用诊断对象系统运行的各种状态信息和已有的各种知识，进行信息的综合处理，最终得到关于系统运行状况和故障状况的综合评价n3。数据融合现在应用的典型系统是C3I系统，尤其是多目标跟踪系统。比照C 3I 系统，故障诊断所需信息的获取途径要更加多样，描述诊断对象的数学模型可能比空间中坐标和速率等特征要更加复杂，诊断对象的故障之间的联系(耦合、备份、传递等)可能要比跟踪对象之问协调行动的关系要更加紧密，但可以把诊断对象看做是一个通过传感器系统观测的特定状态空间，其故障信号就是该空间中的特定目标信号，故障诊断就是根据信号和知识库确定故障报警。

对于故障诊断系统来讲，很适合采用前面介绍的多传感器融合结构，像素层也就是数据层的融合针对传感器反映的直接数据；特征层对应各种故障诊断方法，对数据融合的结果进行有效的决策；决策层融合综合各个子系统通过融合组合规则做出最终的故障诊断结果和故障对策。这三层结构分别对应于故障诊断系统的监测、诊断和决策功能。故障诊断系统中的

数据融合在一定程度上能够使系统获得精确的状态和状态估计、增加置信程度、降低模糊度、改善诊断性能、提高多传感器信息资源的利用率。但随着新技术的发展，故障诊断系统逐渐引入了人工智能技术，主要表现为：使用神经网络进行局部诊断；使用多个互相协作的ES利用多个领域的知识进行信息综合；使用先进的数据库管理技术为决策级推理提供支持；使用学习系统，以便自动适应各种态势的变化。另外还在数据融合的基础上，提高了融合层次，引入了数据挖掘以及知识(包括规则、方法和模型等)融合。

4．基于多传感器数据融合的智能诊断系统

从多传感器数据融合的角度，典型应用实例就是过程监测和故障诊断，而从智能故障诊断系统的角度讲，通常都是在多传感器数据融合的基础上进行综合诊断。基于以上关于多传感器数据融合技术和智能故障诊断系统的讨论，下面将二项技术进行有机的联结，建立一个基于多传感器数据融合的智能故障诊断系统结构框架。

4．1工作原理

该系统由输入输出系统、多传感器信号采集系统、信号预处理系统、诊断子系统以及决策融合系统构成。系统工作时，先利用多传感器采集信号并对信号的数据进行必要的预处理(例如信号滤波、频谱分析、小波分析等)，将处理好的信息与诊断子系统中的专家知识库(规则、方法与模型等知识)按照一定的规则进行推理，然后将各个子系统的局部诊断结果进行并行融合提供给决策融合系统进行全局诊断，最终输出诊断结果的同时将相关信息存人数据库和知识库为利用数据采掘技术进行知识发现作必要的数据储备。

4．2关键技术

4．2．1局部诊断子系统

神经网络能实现复杂的非线性映射，在故障诊断领域得到了广泛的应用口]。当系统的诊断参数较多，征兆信息量大的时候，由于采取的样本不可避免的存在矛盾性和随机性，若将高维的征兆信息同时输入到同一网络处理，将使得训练时间长，诊断效果差，有时甚至导致网络不收敛。因此，如同人脑中不同的区域处理不同的信息一样。不同的信号也应由各自的神经网络来诊断。这样就将高维的征兆空间分解成为低维的征兆空间，这个过程也可以称为局部诊断。另外，神经网络系统可以有效解决专家系统的部分局限性，因此采用了神经网络专家系统。4．2．2决策融合

用神经网络进行局部诊断后，从每个或几个诊断参数都可以得到各自的诊断结果，每个子系统负责一种故障诊断，从不同的侧面诊断故障，决策融合对这些诊断结果进行融合，使得子系统之间形成“会诊”，最大限度提高确诊率。对于预处理后的信息融合，推理比数值运算更重要，应该采用基于知识的专家系统技术与D—S证据理论相结合的方法进行融合。

4．2．3数据挖掘与知识融合

系统现有的运行状态用来修正系统原有的知识库，可以更迅速、更准确、更全面的进行故障诊断，这就是数据挖掘和知识融合的问题，数据挖掘技术将会成为信息融合系统中的必要组成部分。

5．结束语

多传感器数据融合技术和智能故障诊断系统都具有很强的实用性，而二者的有机结合可以对各自技术的进步发展起到互相促进的作用。但是目前信息融合系统的具体融合规则方法的研究还不成熟，知识融合技术也还有待提高，另外智能诊断系统需要改进对AI技术应用。但是相信随着各项技术的逐步完善以及实践经验的不断积累，基于多传感器数据融合的智能故障诊断系统会得到更快的发展和更广的应用。