提升机故障智能诊断理论及应用

发布时间：2018-04-21 23:30

本文选题：希尔伯特-黄变换 + 核方法　；参考：《中国矿业大学》2013年博士论文

【摘要】：机械设备在线监测是企业安全生产、产品质量保证的关键。一方面机械设备结构、运行状态复杂难以建立准确的数学模型，另一方面设备运行状态数据量大，非线性度高、噪声干扰强、不确定等特性使得故障诊断比较困难。本论文借鉴机器学习、故障诊断、人工智能等理论和应用成果，对复杂机械设备的智能故障检测、诊断进行了深入研究，主要内容有： (1)复杂非线性、动态信号处理以及故障统计量构造研究。该方法使用希尔伯特-黄变换（Hilbert-Huang Transform，HHT）振动信号分解到感兴趣的子频带；然后使用HHT把子频带信号分解为多个内蕴模式函数(Intrinsic Mode Function, IMF)，根据IMF系数的邻域相关性去噪，基于信号能量准则消除虚假IMF；提出基于数据依赖KICA（Data Dependent Kernel Component Analysisn, DDKICA）获取描述过程特征的内蕴信息，，给出经验特征空间的DDKICA模型选择准则；最后根据抽取的时频域特征分布使用支持向量描述(Support Vector Data Description, SVDD)构造新的统计量、确定置信度进行故障监控。研究表明该方法能够及时发现异常情况。 (2)基于多尺度理论的振动信号去噪和故障特征提取。分析了形态梯度小波的多尺度特性及其特点，使用形态梯度小波对振动信号进行多尺度分解，对各层的细节系数进行软阈值降噪处理，然后进行信号重构；对降噪后的信号采用S-变换进行多分辨率时频分析，从S变换谱图中提取故障特征。仿真和实例证明该方法能有效提取故障特征，适合在线监测和诊断。 (3)先进机器学习理论在提升机故障监控研究和应用。针对具有冗余、异构（heterogenous）和多尺度特性的高维数据集，本文提出多核正交局部鉴别分析和全局保持（Multiple Kernel Orthogonal Locality Discriminative Analysis with GlobalityPreserving, MKOLDAGP）维数约简算法。该方法不仅保证了低维特征空间与原始数据空间具有相似的几何结构，具有更好的鉴别特性，而且使得数据局部聚类概率密度近似服从高斯分布。最后给出基于GMM的故障监测和故障统计量，较好地克服了现有因非线性、非高斯特性而导致高斯混合模型（Gaussian Mixture Model,GMM）的故障监测性能下降问题。仿真实验表明了本算法可以有效抽取数据特征，有较强的故障检测能力。 (4)不平衡数据集的v-NSVDD多分类研究。分析了多类支持向量数据描述(support vector data description,SVDD)算法存在的问题，提出一种新的不平衡数据v-NSVDD多分类算法。该方法基于不同类别样本间隔最大原理，较好地克服噪声和在野点的影响，提高了分类模型的泛化性能；通过样本加权的方法解决了不平衡类别样本预测精度低的问题，并在理论上给出了根据类别样本数量设置样本加权系数的方法。为实现多分类器拒判，防止因每个分类器的核函数参数不同而影响判决结果的准确性和可靠性，本文给出基于相对距离和K-NN规则相结合的多分类方法。使用Benchmark数据集。进行仿真实验，结果表明本算法能够获得较低的分类误差，能够有效处理样本不平衡问题。
[Abstract]:The on - line monitoring of mechanical equipment is the key to enterprise safety production and product quality assurance . On the one hand , it is difficult to establish an accurate mathematical model on the structure of mechanical equipment and operation state . On the other hand , it is difficult to establish an accurate mathematical model on the other hand , such as machine learning , fault diagnosis , artificial intelligence and so on .

( 1 ) The structure of complex nonlinear , dynamic signal processing and fault statistics is studied . Hilbert - Huang Transform ( HHT ) vibration signal is used to decompose the Hilbert - Huang Transform ( HHT ) vibration signal to the sub - band of interest ;
then using the HHT to decompose the subband signals into a plurality of intrinsic mode functions ( IMF ) , de - noising based on the neighborhood correlation of the IMF coefficients , and eliminating the false IMF based on the signal energy criterion ;
The data dependent Kernel Component ( DDKICA ) is proposed to obtain the intrinsic information describing the process characteristics , and the DDKICA model selection criterion of the empirical feature space is given .
Finally , based on the extracted time - domain feature distribution , a new statistic is constructed using Support Vector Data Description ( SVDD ) , and the confidence is determined to be fault - monitored . The research shows that the method can detect the abnormal condition in time .

( 2 ) The vibration signal de - noising and fault feature extraction based on the multi - scale theory are analyzed . The multi - scale characteristics and characteristics of the morphological gradient wavelet are analyzed , the multi - scale decomposition of the vibration signals is carried out by using the morphological gradient wavelet , the detail coefficients of each layer are soft - threshold denoising and then the signal reconstruction is carried out ;
The multi - resolution time - frequency analysis of the signal after noise reduction is carried out by using S - transform . The fault feature is extracted from the S - transform spectrum diagram . Simulation and examples show that the method can effectively extract fault features and is suitable for on - line monitoring and diagnosis .

( 3 ) The research and application of advanced machine learning theory in the fault monitoring of hoist . Aiming at the high - dimensional data set with redundant , heterogeneous and multi - scale characteristics , this paper puts forward multi - core orthogonal partial differential analysis and global preserving ( Multiple Kernel Orthogonal Locality Analysis with Globality Analysis , MKOLDAGP ) dimension reduction algorithm .

( 4 ) v - nSVDD multi - classification research of unbalanced data set . The problem of support vector data description ( SVDD ) algorithm is analyzed . A new multi - classification algorithm for unbalanced data v - NSVDD is presented .
In this paper , the problem of low prediction accuracy of unbalanced class samples is solved by means of sample weighted method , and the method of setting sample weighting coefficients according to the number of class samples is given . In order to realize multi - classifier rejection , it is possible to prevent the accuracy and reliability of the decision result due to different kernel function parameters of each classifier . The paper gives a multi - classification method based on relative distance and K - NN rule . The simulation experiment is carried out using Benchmark data set . The results show that the algorithm can obtain lower classification error and can effectively deal with the problem of sample imbalance .

【学位授予单位】：中国矿业大学
【学位级别】：博士
【学位授予年份】：2013
【分类号】：TH165.3

【参考文献】