基于全矢极限学习机的轴承故障预测研究

发布时间：2018-04-02 04:31

本文选题：全矢谱　切入点：Hilbert解调　出处：《郑州大学》2017年硕士论文

【摘要】：当前机械设备向着更加高效化、集成化和大型化的方向发展,因此如何保证其安全稳定的运行具有重大研究意义。随着计算机技术和传感器技术的发展,故障预测理论和技术也随之发展完善,并为上述问题提供一种较好的解决途径。但是传统的信息采集方式是用单一传感器来采集单通道信息来进行分析,因为信息不够全面,可靠性并不是很好。而全矢谱技术则将互为正交的两个通道的信号加以融合,能更全面准确反映出设备运行状态。预测算法方面,一直是人工智能研究的热点,而且近年出现了一系列新的算法,比如极限学习机,具备良好的精度和可靠性,并且具有计算量小,速度快等特点。本文将全矢谱技术与神经网络和极限学习机相结合,并以滚动轴承为对象进行故障预测研究,主要研究内容如下几个方面:(1)采用全矢Hilbert解调方法提取滚动轴承的故障特征频率,通过仿真和实例分析,本方法可以有效将特征频率提取出来,为神经网络训练样本的提取奠定理论基础;(2)提出了基于全矢模糊神经网络的故障预测模型,并进行实例验证,结果表明,因为模糊神经网络自身的缺陷,预测结果稳定性不佳,容易造成误诊;(3)提出了基于全矢小波神经网络的故障预测模型,并进行实例验证,结果表明,利用小波基函数对神经网络进行改造,可以提高神经网络在滚动轴承故障预测中的效果。但是小波神经网络依然没有解决传统神经网络的故有缺陷,依然采用梯度下降的学习方法,因此容易陷入局部极值而影响预测模型效果;(4)针对传统神经网络的不足,提出全矢极限学习机的轴承故障预测模型,通过实例验证表明,它可以很好的拟合期望输出,误差度小,对滚动轴承故障的预测表现了很好的稳定性和精确性。极限学习机不像传统神经网络那样需要设置大量参数,只需要设置隐含层神经元数即可,因此其计算量小,运算速度快,具有很好的实时性能和实用价值。
[Abstract]:At present, mechanical equipment is developing towards higher efficiency, integration and large scale, so it is of great significance to study how to ensure its safe and stable operation.With the development of computer technology and sensor technology, the theory and technology of fault prediction are improved, and a better way to solve the above problems is provided.But the traditional way of information acquisition is to use a single sensor to collect single channel information for analysis, because the information is not comprehensive, the reliability is not very good.The full-vector spectrum technology combines the signals of the two channels which are orthogonal to each other and can reflect the running state of the equipment more comprehensively and accurately.Prediction algorithm has always been a hot topic in artificial intelligence, and in recent years, a series of new algorithms have emerged, such as extreme learning machine, with good accuracy and reliability, and has the characteristics of small computation, fast speed, and so on.In this paper, the full-vector spectrum technique is combined with neural network and ultimate learning machine, and the fault prediction of rolling bearing is studied. The main research contents are as follows: 1) the fault characteristic frequency of rolling bearing is extracted by full-vector Hilbert demodulation method.Through simulation and example analysis, this method can effectively extract the feature frequency, lay a theoretical foundation for neural network training sample extraction. (2) A fault prediction model based on full-vector fuzzy neural network is proposed and verified by an example.The results show that because of the defects of the fuzzy neural network, the stability of the prediction results is not good, and it is easy to cause misdiagnosis. (3) A fault prediction model based on the full-vector wavelet neural network is proposed and verified by an example.The improvement of neural network by wavelet basis function can improve the effect of neural network in rolling bearing fault prediction.However, wavelet neural network still has not solved the shortcomings of traditional neural network, and still adopts gradient descent learning method, so it is easy to fall into local extremum and affect the prediction model effect.A bearing fault prediction model of full-vector learning machine is put forward. The example shows that it can fit the expected output well, and the error degree is small, and the prediction of rolling bearing fault shows good stability and accuracy.Extreme learning machine does not need to set a large number of parameters as the traditional neural network, only need to set the number of hidden layer neurons, so it has a small amount of calculation, fast operation, good real-time performance and practical value.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TH133.3

【相似文献】