基于HMM的退化状态识别和故障预测研究

发布时间：2018-08-02 11:13

【摘要】：现代机械设备的自动化程度和智能化水平越来越先进，它的发展对工业、经济有着深刻影响。随着基于状态维修（CBM）和故障预测与健康管理（PHM）等维修理论和技术的发展，近几年，对状态实时监测技术、状态信息采集和处理技术、故障诊断和预测技术的研究成为热点。在机械设备状态监测和故障诊断领域内，机械设备从正常运行状态到故障状态通常要经过一系列不同的退化状态，如何正确识别设备当前所处的状态，进一步预测设备的发展态势，为维护决策提供依据是一个迫切需要解决的问题。基于上述问题，本文做了以下研究：（1）非平稳信号预处理在对机械设备进行状态监测情况下，由于振动信号较其它信号容易采集，并且对故障较为敏感，能提供设备运行状况的丰富信息，所以把振动信号作为设备退化状态识别的特征信号。振动信号是一种典型的非平稳信号，由于外界干扰，对其进行预处理是后期研究的关键。本文首先介绍了小波包能量阈值去噪法，并对小波包能量阈值去噪方法的小波基的选取进行分析和仿真，通过实验给出小波包能量阈值去噪和谱相减去噪的适用环境。小波包能量阈值去噪适合用于输入信噪比低的信号，谱相减去噪适合用于信噪比高的信号，两种方法可以结合使用对振动信号去噪处理。（2）EMD能量熵特征提取传统的傅里叶变换无法同时兼顾信号在时频两域的全貌和局部化特征，小波分析不具有自适应性，针对两者的不足，论文提出了基于经验模态分解（EMD）的特征提取方法；信息熵是对设备状态不确定程度和复杂程度的描述，，当信源含有的信息波动不稳定、成分比较复杂时，信息熵就越大。论文利用EMD方法把经过去噪后的信号分解成一组固有模态函数（IMF）分量，并提取和计算各IMF能量及能量熵，作为描述退化状态的特征参数。（3）基于HMM退化状态识别和故障预测针对隐马尔科夫模型（HMM）算法中参数设置问题、训练算法容易陷入局部最优的问题，深入研究了HMM的改进算法；针对单一方法进行故障预测存在的缺陷，利用HMM和指数平滑预测相结合的方法进行研究，可以综合两者的优点；最后以液压元件为研究对象，对上述方法进行验证，并和BPNN、SVM的识别效果进行比较，算例结果表明，该方法具有鲁棒性好、分辨灵敏度高和故障预测总体准确率较高的优点。
[Abstract]:The level of automation and intelligence of modern mechanical equipment is more and more advanced, and its development has a profound impact on industry and economy. With the development of maintenance theory and technology such as condition based maintenance (CBM) and fault prediction and health management (PHM), the research on state real-time monitoring technology, state information collection and processing technology, fault diagnosis and prediction technology has become a hot topic in recent years. In the field of mechanical equipment condition monitoring and fault diagnosis, mechanical equipment usually goes through a series of different degenerative states from normal operation state to fault state, how to correctly identify the current state of the equipment, It is an urgent problem to predict the development situation of equipment and provide basis for maintenance decision. Based on the above problems, this paper has done the following research: (1) in the condition of monitoring the condition of mechanical equipment, the vibration signal is easier to collect than other signals, and it is more sensitive to fault. The vibration signal is regarded as the characteristic signal of equipment degenerative state recognition because it can provide abundant information about the equipment running condition. Vibration signal is a kind of typical nonstationary signal. In this paper, the wavelet packet energy threshold denoising method is first introduced, and the selection of wavelet basis for wavelet packet energy threshold denoising method is analyzed and simulated. The suitable environment for wavelet packet energy threshold denoising and spectral phase subtraction is given through experiments. Wavelet packet energy threshold denoising is suitable for input signal with low signal-to-noise ratio (SNR), and spectral phase subtraction is suitable for high SNR signal. The two methods can be combined with vibration signal denoising. (2) EMD energy entropy feature extraction can not simultaneously take into account the full feature and localization feature of the signal in time-frequency domain. Wavelet analysis is not self-adaptive. In view of the shortcomings of the two methods, a feature extraction method based on empirical mode decomposition (EMD) is proposed, and the information entropy is used to describe the degree of uncertainty and complexity of equipment state. When the information contained in the source fluctuates unsteadily and the components are complex, the information entropy increases. In this paper, the de-noised signal is decomposed into a set of inherent mode function (IMF) components by EMD method, and each IMF energy and energy entropy are extracted and calculated. As a characteristic parameter to describe degenerate state. (3) based on HMM degenerate state identification and fault prediction, the training algorithm is prone to fall into the local optimal problem for parameter setting in Hidden Markov Model (HMM) algorithm. This paper deeply studies the improved algorithm of HMM, aiming at the defects of single method in fault prediction, using the combination of HMM and exponential smoothing prediction, it can synthesize the advantages of both. Finally, the hydraulic components are taken as the research object. The method is verified and compared with that of BPNNN SVM. The results show that the proposed method has the advantages of good robustness, high resolution sensitivity and high overall accuracy of fault prediction.
【学位授予单位】：太原科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN911.7

【参考文献】