基于粗糙集理论和人工神经网络的滚动轴承故障诊断

发布时间：2018-02-28 16:48

本文关键词： 滚动轴承故障诊断经验模态分解小波分析小波包分析粗糙集人工神经网络　出处：《西南交通大学》2012年硕士论文　论文类型：学位论文

【摘要】：随着现代机械设备日趋大型化、精密化和自动化,有效的机械故障诊断作为系统可靠和安全运行的保障,具有非常重要的价值。滚动轴承因摩擦力小、装配方便等优点成为机械设备中最常用的零部件,其稳定性直接影响设备的性能。对滚动轴承实施状态监测与故障诊断,对避免经济损失和重大事故发生有重大意义。在滚动轴承故障诊断中,振动分析是最常用和有效的方法。掌握滚动轴承振动机理,确定振动信号测定方式,模拟滚动轴承内圈、外圈故障后,构建故障诊断实验系统。本文采集并处理振动信号,从时域、频域和时频域分析并提取反映滚动轴承运作状态的特征向量。其中,基于经验的模态分解根据自身时间尺度特征分解信号,具有高信噪比和自适应性,在非线性、非平稳信号的分析和处理上占有很大优势；小波分析采用变化的窗函数实现局部化的频率分析,具有多分辨特性,广泛应用于信号的降噪和压缩,从小波能量的角度出发,可以挑选感兴趣的分层进行分析；小波包分析是小波分析的延伸与拓展,实现高频和低频的同步分解,提高时频分辨率,更具应用价值。粗糙集理论是人工智能领域中处理不完备、不精确信息的软计算方法,在知识挖掘、决策分析等领域有着广泛的应用。在滚动轴承故障诊断中,保证诊断精度不变的情况下,粗糙集能有效地减少特征维数,保留核属性,减小计算量和不确定因素的影响,降低故障诊断系统的复杂度与规模。人工神经网络模拟人脑结构和功能,是强大的信息处理系统,具有高度自适应性、并行处理方式、自我学习和归纳的能力。通过学习和训练,神经网络由故障症状推断故障产生原因,实现滚动轴承故障诊断和模式识别。本文采用三种方法对比实现轴承故障诊断。第一,将归一化的特征向量导入训练好的神经网络,实现滚动轴承故障诊断;第二,建立粗糙集分类器,通过自学习实现滚动轴承状态分类；第三,将粗糙集作为前端数据预处理器,实现数据离散、属性约简和决策规则的生成,优化的特征参量作为神经网络的输入。结果表明,粗糙集和神经网络相结合的故障诊断系统准确率和效率明显提高。本文的重点是实现敏感特征向量的有效提取,灵活运用粗糙集理论预处理特征,消除冗余信息,防止信息爆炸,结合神经网络容错和泛化能力强的优势,有效地实现滚动轴承故障诊断。同时,粗糙集理论作为全新的特征降维技术,在智能化故障诊断领域都有着广泛的应用和发展。
[Abstract]:With modern mechanical equipment becoming larger and larger, precision and automation, effective mechanical fault diagnosis as a guarantee of reliable and safe operation of the system, has a very important value. The advantages of convenient assembly have become the most commonly used parts in mechanical equipment, and its stability directly affects the performance of the equipment. It is of great significance to avoid economic losses and serious accidents to implement condition monitoring and fault diagnosis for rolling bearings. In the fault diagnosis of rolling bearing, vibration analysis is the most common and effective method. After mastering the vibration mechanism of rolling bearing, determining the measuring method of vibration signal, simulating the fault of inner ring and outer ring of rolling bearing, In this paper, the vibration signals are collected and processed, and the characteristic vectors reflecting the operation state of rolling bearings are analyzed and extracted from time domain, frequency domain and time frequency domain. Among them, the empirical mode decomposition decomposes the signal according to its own time scale characteristic, has the high signal-to-noise ratio and the adaptability, and has the very big superiority in the non-linear, the non-stationary signal analysis and the processing; Wavelet analysis uses the variable window function to realize localized frequency analysis, which has multi-resolution characteristic, and is widely used in signal denoising and compression. From the angle of wave energy, we can select the layers of interest for analysis. Wavelet packet analysis is an extension and extension of wavelet analysis. It can realize synchronous decomposition of high frequency and low frequency and improve time-frequency resolution. Rough set theory is a soft computing method for dealing with incomplete and imprecise information in the field of artificial intelligence. It is widely used in the fields of knowledge mining, decision analysis and so on. Rough set can effectively reduce the feature dimension, preserve kernel attributes, reduce the influence of computation and uncertainty, and reduce the complexity and scale of fault diagnosis system. Artificial neural network simulates the structure and function of human brain. It is a powerful information processing system with high adaptability, parallel processing, self-learning and inductive ability. Neural network infer the cause of fault from the fault symptom, and realize fault diagnosis and pattern recognition of rolling bearing. This paper uses three methods to realize bearing fault diagnosis. First, the normalized eigenvector is introduced into the trained neural network to realize the rolling bearing fault diagnosis; second, the rough set classifier is established. The status classification of rolling bearing is realized by self-learning. Thirdly, rough set is used as front-end data preprocessor to realize data discretization, attribute reduction and decision rule generation, and optimized characteristic parameters are used as input of neural network. The accuracy and efficiency of the fault diagnosis system based on rough set and neural network are improved obviously. The emphasis of this paper is to realize the effective extraction of sensitive feature vectors, to flexibly use rough set theory to preprocess features, to eliminate redundant information, to prevent information explosion, and to combine the advantages of neural network with strong fault tolerance and generalization ability. At the same time, as a new feature dimension reduction technology, rough set theory has been widely used and developed in the field of intelligent fault diagnosis.
【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH133.33;TH165.3

【参考文献】