小波神经网络在滚动轴承故障诊断中的应用研究

发布时间：2018-10-15 20:44

【摘要】：轴承是机械设备中最常用的部件,且应用广泛,尤其是滚动轴承在应用中起着越来越重要的作用,同理在轴承的各种故障中,滚动轴承故障占有的比例也越来越大。而随着工业4.0概念的提出,我国重大技术机械设备必将迎来大的发展,而机械设备的稳定性、准确性、可靠性和各项经济指标,都与轴承有着密切。对轴承故障判别不准有可能造成轴承浪费或者严重生产事故,因此,有必要对滚动轴承的故障诊断。对滚动轴承的故障诊断主要包含信号测取、特征提取、状态识别、诊断分析、决策干预五个环节,其中特征提取和状态识别是轴承故障诊断的核心。目前对滚动轴承的故障诊断方法主要有传统的诊断方法和现代诊断算法两类。如:冲击脉冲法、小波分析法、神经网络、模糊理论、专家系统和粒子群算法等。小波变换具有较好的时频局部分析能力,神经网络具有处理复杂多模式及进行联想、推测和记忆的功能,正是基于此提出本文的研究内容。首先利用小波变换对轴承振动信号进行滤波,然后对传统的软硬阈值函数进行优化,设计全新的阈值函数,对滚动轴承进行小波阈值滤波。由于不同的特征值和同一特征值的不同样本之间相差较大,需要对其进行归一化处理,然后作为神经网络的输入,利用构造的神经网络对输入的特征进行识别,诊断结果显示,文中构造的小波阈值函数能够有效地去除噪声信号,具有较高的识别正确率。
[Abstract]:Bearing is the most commonly used part in mechanical equipment, and it is widely used, especially the rolling bearing plays more and more important role in the application. Similarly, in all kinds of bearing faults, the proportion of rolling bearing fault occupies more and more. With the development of the concept of industry 4.0, the important technical machinery and equipment in our country will be greatly developed, and the stability, accuracy, reliability and various economic indexes of the machinery and equipment are closely related to the bearing. Therefore, it is necessary to diagnose the fault of rolling bearing. The fault diagnosis of rolling bearing mainly includes five steps: signal measurement, feature extraction, status recognition, diagnosis analysis and decision intervention. Feature extraction and status recognition are the core of bearing fault diagnosis. At present, there are two kinds of fault diagnosis methods for rolling bearing: traditional diagnosis method and modern diagnosis algorithm. For example: impulse method, wavelet analysis, neural network, fuzzy theory, expert system and particle swarm optimization. Wavelet transform has better time-frequency local analysis ability, neural network has the function of dealing with complex multi-mode and associating, speculating and memorizing. Based on this, the research content of this paper is put forward. Firstly, the bearing vibration signal is filtered by wavelet transform, then the traditional soft and hard threshold function is optimized, a new threshold function is designed, and the rolling bearing is filtered by wavelet threshold. Due to the large difference between different eigenvalues and different samples of the same eigenvalue, it is necessary to normalize them. Then as the input of the neural network, the input features are identified by using the constructed neural network, and the diagnosis results show that, The wavelet threshold function constructed in this paper can effectively remove the noise signal and has a high recognition accuracy.
【学位授予单位】：河南工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH133.33

【相似文献】