【作者】 王飞；

【导师】 岳东；

【作者基本信息】 华中科技大学 ， 控制理论与控制工程， 2013， 硕士

【摘要】 随着现代机械设备朝着大型化、智能化发展，设备结构变得越来越复杂，对设备进行故障诊断也越发困难。在机器学习和数据挖掘技术的推动下，基于机器自学习的故障诊断系统正飞速发展。智能化的故障诊断系统克服了原有诊断方式的缺陷，能够及时准确发现机械设备隐藏的故障信息，从而快速自动监测诊断设备故障，提高了诊断效率，减少了因诊断不及时不准确造成的损失。智能化的故障诊断系统中，诊断规则如何获取是一个关键。本文采用数据挖掘中应用十分广泛成熟的决策树技术作为诊断规则提取的主要技术。首先对已有的决策树方法进行了分析和研究，发现基于粗糙集和变精度粗糙集理论的决策树构造方法具有较好的分类效果，然而也存在分类精度不够高，决策树节点属性选择困难，对噪声数据抑制能力差等特点，由此提出了基于变精度粗糙集的决策树构造改进算法。在Weka机器学习平台上试验比较，结果表明本文提出的算法在分类精度、决策树复杂度、抑制噪声能力方面都得到一定提升，证明了算法的有效性。在所提改进方法的基础上，以选煤厂机械设备为对象，设计实现了一套故障诊断与分析系统。系统根据已经测得的能够反映设备运行状况的振动历史数据，利用本文提出的决策树构造算法得到设备诊断规则，再将规则应用得到实时诊断系统中来，从而达到对设备实时运行状况进行分析诊断的目的。更多还原

【Abstract】 With the development of modern mechanical equipments in large-scale and intelligent, the structure of the mechanical equipments is more and more complicated, the fault diagnosis of equipment is becoming more and more difficult. Driven by machine learning and data mining, the fault diagnosis system based on machine self-learning has been developing rapidly. The intelligent fault diagnosis system overcomes the defects of the original diagnosis, and can find the hidden faults of mechanical equipment timely and accurately, thus monitoring and diagnosing the equipment’s fault automatically and fast, improving the diagnostic efficiency, and reducing the loss caused by the inaccurate diagnosis.In the intelligent fault diagnosis system, how to get the diagnosis rules is a key. This paper uses the decision tree technology as the main technique extracting diagnosis rules which are widely used in data mining. By analyzing and researching the existing decision tree method, I found that the decision tree Constructor method based on rough sets and variable precision rough set theory has a better effect on classifying. However there are also disadvantages, for example, the classification accuracy is not high, it’s difficult to select the decision tree node attribute, the capability of rejecting the noisy data is poor etc. Thus a improved decision trees algorithm based on variable precision rough set is proposed. Through testing it on a machine learning platform called Weka, the results show that the proposed algorithm are improved in classification accuracy, complexity and noise resistant ability. So we prove the effectiveness of the proposed algorithm.On the basis of the improved method, we designed and implemented a set of fault diagnosis and analysis system, which is targeted at the mechanical equipment in coal preparation plant. According to the history data of vibration which can reflect the running state of the equipments, we can get the diagnosis rules by using the proposed algorithm. Then the rules apply to the real-time diagnosis system, so that we can analyze and diagnose the real-time conditions of the equipments.更多还原