基于主元分析的空气压缩机故障诊断研究

发布时间：2019-05-05 07:24

【摘要】：本论文研究的目的旨在寻找一种适合于空气压缩机的、高效可行的故障诊断方法。本文从空气压缩机的工作原理、机械构造、故障类型以及故障机理出发,总结了空气压缩机的主要故障特点,并归纳了空气压缩机故障诊断系统应满足的实现要求。在寻找合适的故障诊断算法时,本文将重点放在如何解决空气压缩机故障检测变量过多和检测变量之间存在高相关性的问题上。为解决这一问题,本文采用了主元分析技术(Principal Components Analysis, PCA)作为检测数据的预处理算法。该算法通过分析数据样本在高维空间内的分布情况,可以找出数据在高维空间内的主要变动方向和趋势,从而提取出包含有原始数据绝大部信息的特征向量来代替高维的、高相关性的原始数据。在以PCA技术作为数据预处理算法的基础上,本文先后提出了基于PCA技术结合径向基函数(Radical Basis Function, RBF)神经网络的空气压缩机故障诊断方法和基于PCA技术结合D-S证据理论的空气压缩机故障诊断方法。基于PCA技术与RBF神经网络的空气压缩机故障诊断方法,是通过建立空气压缩机运行状态的主元模型来处理采集到的庞大、高相关的原始数据集,利用特征提取的方法简化原始数据,并使用简化后的样本数据训练RBF神经网络,最后通过训练好的RBF识别网络实现空气压缩机的故障分类。该方法可以充分发挥PCA技术在数据降维、除相关性上的优势,极大的简化复杂的检测数据。同时,PCA的降维作用也为RBF网络的训练和识别减化了大量的运算过程,从而可以提高神经网络训练和识别的速度,同时神经网络处理数据维数的降低,不仅避免了RBF网络在训练过程中由于处理数据维数过高可能发生崩溃的危险,而且还提高了神经网络的分辨率。基于PCA技术与D-S证据理论的空气压缩机故障诊断方法,是一种基于信息融合思想的故障诊断方法,该方法从空气压缩机不同运行状态的角度(即不同证据)进行观察,通过分析检测数据在各证据下呈现的特征信息对压缩机的运行状态进行判断,最后以D-S组合规则将各证据下的判别结果融合成一个综合的结果,从而实现空气压缩机运行状态的最终判别。该方法可以更加全面分析检测数据的信息,具有较快的处理速度和抗干扰能力强的特点,可以实现高精确度的故障分离和判别。
[Abstract]:The purpose of this paper is to find an efficient and feasible fault diagnosis method for air compressor. Based on the working principle, mechanical structure, fault type and fault mechanism of the air compressor, the main fault characteristics of the air compressor are summarized, and the realization requirements of the fault diagnosis system for the air compressor are summarized. In order to find a suitable fault diagnosis algorithm, this paper focuses on how to solve the problem of high correlation between the air compressor fault detection variables and the detection variables. In order to solve this problem, the principal component analysis (Principal Components Analysis, PCA) technique is used as the preprocessing algorithm of the detection data in this paper. By analyzing the distribution of data samples in the high-dimensional space, the algorithm can find out the main changing directions and trends of the data in the high-dimensional space, and then extract the feature vectors which contain most of the information of the original data to replace the high-dimensional ones. Highly correlated raw data. On the basis of the PCA technology as the data preprocessing algorithm, this paper proposes the combination of the radial basis function (Radical Basis Function, and the PCA technology. RBF) neural network based air compressor fault diagnosis method and air compressor fault diagnosis method based on PCA technology combined with DES evidence theory. The fault diagnosis method of air compressor based on PCA technology and RBF neural network is to process the large and highly correlated original data set by establishing the principal component model of the running state of air compressor. The method of feature extraction is used to simplify the original data, and the simplified sample data is used to train the RBF neural network. Finally, the trained RBF recognition network is used to realize the fault classification of the air compressor. This method can make full use of the advantages of PCA technology in data dimensionality reduction and correlation, and greatly simplify the complex detection data. At the same time, the dimensionality reduction of PCA reduces a lot of operation process for the training and recognition of RBF network, which can improve the speed of training and recognition of neural network, and reduce the dimension of data processed by neural network at the same time. The RBF network not only avoids the possibility of collapse due to the high dimension of processing data, but also improves the resolution of neural network in the process of training. The fault diagnosis method of air compressor based on PCA technology and D S evidence theory is a fault diagnosis method based on the idea of information fusion. The method is observed from the angle of different operating state of air compressor (that is, different evidence), and the fault diagnosis method is based on the idea of information fusion. By analyzing the characteristic information of the test data under each evidence, the running state of the compressor is judged. Finally, the discriminant results under each evidence are fused into a comprehensive result by the combination rule of DES. Thus, the final discrimination of the running state of the air compressor can be realized. This method can analyze the information of the detected data more comprehensively, has the characteristics of fast processing speed and strong anti-jamming ability, and can realize high-precision fault separation and discrimination.
【学位授予单位】：长春工业大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TH165.3;TH45

【参考文献】