文档介绍:华中科技大学
硕士学位论文
基于机器学习的油液分析系统研究
姓名:段学燕
申请学位级别:硕士
专业:管理科学与工程
指导教师:李世其;朱文革
20060422
摘要
随着主动维修思想的出现,其优势日趋明显,同时,油液分析技术作为其实现的
前提条件,已经得到广泛的应用,人们开发出各样的油液分析综合系统,以适应维修
领域知识专家经验不能满足需要这一现状。本文提出了一个基于机器学习方法的油液
分析系统的模型,运用一种新的增量式的学习方法,解决油液分析系统的知识获取问
题。另一方面,通过决策树方法得到的知识模型,辅助故障源的判断,能够在故障发
生之前提前预防,达到主动维修的目的。
首先,针对实现主动维修的油液分析系统中存在的知识获取,以及故障源挖掘的
问题,结合机器学习方法,提出了一种机器学习方法在油液分析专家系统中的应用模
型,并对模型中的关键部分作了简要概述。
其次,阐述了系统中选取具体的机器学习方法的过程,并对其中的决策树方法辅助
故障源的挖掘过程进行了详细阐述。利用决策树方法学习到的知识模型,结合机器学
习的推理能力,发掘诊断属性中的重要参数,辅助故障源的确定以及维修决策。
接着,本文在决策树增量式学习算法和粗糙集理论的基础上,提出了一种粗
糙集结合决策树增量式的学习方法,使得学习过程在遇到新的故障示例时,不需要重
新计算一次所有结点的信息熵,减小了系统对新示例学习的消耗。并将该方法运用于
系统的知识获取过程,并通过一组油液分析样本数据对该算法进行了验证。
最后,利用 Microsoft SQL Server 以及 Powerbuilder,开发了一个决策树算法应用于
油液分析系统的原型系统。
关键词: 主动维修决策树算法增量式学习粗糙集理论
Abstract
As the development of maintenance, a new kind of maintenance method called
proactive maintenance came out and became more and more effective during the equipment
fault diagnosis. How to define the fault source is the critical problem during the processes of
proactive maintenance. On the other hand, the ability of knowledge acquisition to a expert
system is very important now, and we always use the machine learning method to achieve
the knowledge acquisition automatically. All these problems are discussed in this paper.
A proactive maintenance based on machine learning approach is defined at first to solve
the problems such as knowledge acquisition automatically and help make decision about the
fault source and maintenance.
In all kinds of machine learning approach, the conclude method is more suitable than
the others to the fault diagnosis expert system. With the ability of incremental learning, it can
deal with a new instance without relearning the entire example set. The paper describes an
approach that the incremental induction of decision tree is used in the proactive maintenance.
On the other hand, how to describe the standard of the fault state of equipment is