【作者】 刘杰；

【导师】 严新平；

【作者基本信息】 武汉理工大学 ， 载运工具运用工程， 2010， 博士

【摘要】 船舶动力设备是船舶的关键部件,对其进行状态监测和故障诊断受到国内外学者和研究机构的广泛关注。在船舶动力设备的故障诊断中,判据知识欠缺-直是制约其发展的一个关键因素,为此非常需要构建一个分布式的船舶动力设备故障诊断的资源环境,在这个资源环境下能共享诊断数据、案例和知识,并利用数据挖掘等技术从中提炼出新的诊断判据。针对目前船舶动力设备故障诊断研究领域的研究工作基本处于相互分离状态的现状,论述了构建船舶动力设备状态监测和故障诊断分布式资源环境的重要意义；分析了构造分布式资源环境的几个关键问题,定义了分布式资源环境中的角色分类和角色功能,论述了资源的分类特征和表述方法；给出了分布式资源环境体系结构和创建步骤,为推动船舶动力设备故障诊断系统真正走向实用建立坚实基础。对状态监测和故障诊断中的仪器设备特征分类进行深入分析,实现了串口类型仪器数据的自动高效采集集成,同时利用文件夹监控方式和消息通信方式实现了自带电脑型仪器的网络化数据集成；在此基础上构建了监测分析实验中心的自动化数据采集系统,经实际运行验证,取得良好效果；提出并实现了基于聚类相似度分析的分析仪器数据格式分析算法,为对加密型仪器设备数据进一步利用奠定了很好的基础；针对船舶移动工况,论述了在状态监测与故障诊断中集成机务维护信息的必要性,利用程序脚本代码自动生成技术方便高效的解决了机务维护信息修改后的集成问题；这些技术的运用,提高了检测信息的集成度,为更好地在分布式资源环境下利用这些数据建立坚实的基础。获取判据知识一直是设备状态监测与故障诊断中的难点。传统的获取方式是依靠不断总结专家的经验,但所形成的诊断知识不一定准确和高效。借助于网络环境,将拥有相同设备的不同公司、用户组织成一个整体,将各自在实际监测诊断中收集的原始数据和形成的诊断判据知识共享,借助于知识挖掘、信息融合等方法,可形成准确度更高的判据知识。文中讨论了在船舶动力设备状态监测和故障诊断领域中应用数据挖掘方法来获得诊断知识的途径；针对数据挖掘中的聚类算法,提出并实现了一种新的谱系图生成算法；分析了传统正态分布方法获取监测数据基线的不足,介绍了用最大熵方法计算判据基线值的过程,通过.net编程语言实现了最大熵算法程序,分析了最大熵方法的应用要求,针对柴油机台架试验数据用最大熵方法和正态分布方法分别计算了油液光谱分析元素浓度绝对值的判据和变化率判据,并对数据进行了分析,得出了最大熵方法挖掘判据基线的前提条件要求；针对目前数据挖掘方法发展变化非常迅速的特点,提出了用反射技术来构建可扩展式数据挖掘应用系统的方法,对分布式资源环境的创建有非常重要的意义。分析了船舶动力设备故障诊断知识的特征,提出了用数据库技术来保存产生式规则知识的体系,研究了相应的存储结构,提出了一种方便灵活的动态知识匹配诊断方法。针对目前故障诊断中,有些知识是模糊性的,还无法表示为规则,只存在相应案例样本的情况,实现用神经网络来保存该类型知识,并编程实现了神经网络的构造、训练、保存、加载和诊断,对分布式资源环境提供了有力的支持。论述了知识服务的概念和意义；提出以远程知识服务的形式来对外开展诊断服务,描述了基于Web Service技术的远程知识服务系统体系的关键技术；解决了分布式资源环境下不同节点的对外服务运作形式问题；以构建远程磨粒图像处理知识服务为例,描述了构造知识服务的过程。介绍了构造的基于分布式资源环境理念的远程船舶故障诊断系统平台,包括设计理念、扩充的接口、运行效果等。更多还原

【Abstract】 The condition monitoring and fault diagnosis for marine power equipment which is the key component of ships has been widely concerned by domestic and foreign scholars and research organizations. However, the shortage of criterion knowledge has been the key obstacle to the development of fault diagnosis for marine power equipment. Thus there is a great request to build a distributed resource environment of fault diagnosis for the marine power equipment, in which the diagnosis data, cases and knowledge can be shared and new criteria for diagnosis can be also refined using the technologies of DM (data mining) and so on.Aiming at the present separation condition of the researches in the field of fault diagnosis for marine power equipment, the significance of building the distributed resource environment of condition monitoring and fault diagnosis for the marine power equipment was discussed. And several key issues about building the distributed resource environment were analyzed. Then the classification and functions of roles in the distributed resource environment were defined. Furthermore the classification features and expression methods of the sources were discussed. Finally the architecture and establishment steps of the distributed resource environment were given, which served to advance the fault diagnosis system for the marine power equipment toward actual practice as a solid foundation.The feature classification of instrument and equipment in the condition monitoring and fault diagnosis was in depth analysis. And the high-efficient automatic acquisition and integration of data acquired by instrument with serial port were realized. Taking advantage of the folder monitoring and message communication, the networked integration of data from instrument with its own computer was also achieved. Based on this, an automatic data acquisition system of experimental center for monitoring and analysis was established, and good effect had been proved by practical running. The data format analytical algorithm for the analytical instrument was put forward and come true based on cluster and similarity analysis, which laid down a good foundation for the further use of enciphered data. Under the shifting condition of ships, it is necessary to integrate the maintenance information in the process of condition monitoring and fault diagnosis. A problem of integration after the maintenance information was modified had been solved conveniently and effectively using the program code automatic generation technology in this article. The integration level of detection information was improved through the use of these technologies, which established a strong foundation to facilitate better use of monitoring data in the distributed resource environment.As we all know, it’s quite difficult to acquire criterion knowledge for the condition monitoring and fault diagnosis in machinery. The traditional approach depends on constant summary of experiences from the experts. However, the acquired knowledge is probably inaccurate and inefficient. With the benefit of network, the different companies and users that have the same equipments can be organized to form an integrated information source in order to share the original data and diagnosis knowledge collected in the actual monitoring diagnosis. And more accurate diagnosis knowledge may be acquired using knowledge mining and information fusion. In the field of the condition monitoring and fault diagnosis for marine power equipment, the approach of obtaining diagnosis knowledge through DM was discussed in this paper. Aiming at the clustering algorithm in DM, a new generation algorithm of pedigree chart was proposed and realized. The shortcoming of applying the traditional normal distribution to acquire the monitoring data’s baseline was analyzed. Maximum entropy method was brought in to calculate the criteria’s baseline value and the algorithm procedures of maximum entropy was also achieved using.net programming language. Requirements for the application of maximum entropy were analyzed. For the data from diesel engine stand test, the absolute value’s criteria and rate of change’s criteria of element concentration in spectrometric oil analysis were computed separately by maximum entropy method and normal distribution method. The precondition requirement of using maximum entropy method to dig out the criterion baseline was gained through data analysis. According to DM’s present characteristic of quick change and rapid development, a method to construct an extendable DM application system based on the reflective technique was proposed, which was of important signification in the foundation of distributed resource environment.Analyzing the features of fault diagnosis knowledge for the marine power equipment, the system which used database technology to store the knowledge of production rules was put forward. The corresponding storage organization was studied. A convenient and flexible method to match and diagnose the knowledge automatically was also proposed. For the present fault diagnosis, considering the fuzziness of some knowledge, which failed to be represented in rules but only had corresponding case samples, so Neural Network (NN) had been brought in to use for storing this class of knowledge. With the help of my own programming, the construction, training, storage, loading and diagnosis of NN had been carried out, which provided the powerful support for the distributed resource environment.In addition, the concept and meaning of knowledge service were discussed. The diagnostic service was provided externally in the form of remote knowledge service. The framework of remote knowledge service system based on web service technology was represented. The operation form of external service for different nodes in the distributed resource environment was solved. Taking the remote knowledge service of wear debris’ image processing for example, the construction procedure of knowledge service was described in detailed.Finally, the system platform of the marine remote fault diagnosis, which was built in this paper based on the idea of distributed resource environment, was introduced including design concept, expansive interface, operation effect, and so on.更多还原