基于云计算的组合方法在电机故障诊断中的研究

发布时间：2018-06-25 11:40

本文选题：等谱流形学习算法 + 降维　；参考：《兰州理工大学》2017年硕士论文

【摘要】：当前随着社会经济和科学技术的不断发展,各类电机出现在工业生产和人们的日常工作中并且它所起到的作用也越来越大。而电机的故障一旦发生,轻则会影响人们的生产生活,重则会危害人的生命安全以及造成严重的经济损失。因此,为了满足工业自动化对电机的高品质需求,研究应用于诊断电机故障的方法在现代生活中具有重大意义。随着科技的不断发展,电机诊断更需要准确性和快速性,也就导致在电机故障诊断领域存在一些技术难题。例如高维故障数据的特征提取不精确导致诊断精度低的难题、单一诊断方法的局限性、较低的运算效率等问题。基于上述情况,本文主要研究了等谱流形学习算法、狼群算法、组合诊断模型以及云计算来解决问题,并最后通过美国凯斯西储大学的数据作为实例进行仿真验证。具体内容概况如下:本文首先针对电机故障的高维数据在提取特征集时不精确导致其诊断精度低的问题,引入了等谱流形学习算法进行降维处理,即采用此算法对经过主成分分析降维处理后的数据进行二次降维,此算法通过其修正后的稀疏重构权矩阵构建邻接图,使得经降维后同类样本更聚集,不同类样本更疏散,有效实现了高维数据的去冗降维,最后将其和主成分分析进行比较分析,其效果显著。然后针对诊断精度易受其等谱流形学习算法、最小二乘支持向量机参数影响的问题,通过狼群算法对其中参数进行优化。在优化最小二乘支持向量机的参数组合时使用Fisher准则函数作为所选参数的优劣标准,而在优化等谱流形学习算法的参数时使用的适应度函数为最近邻分类法的识别率,利用优化的参数建立最优的诊断模型。仿真实验表明该模型有很好的诊断结果,并在优化参数时将狼群算法和粒子群优化算法进行了比较分析,其诊断效果显著。最后针对电机故障征兆的多样性和单一诊断方法的局限性等问题,本文通过采用组合诊断模型进行解决。本文将最小二乘支持向量机、模糊神经网络以及RBF神经网络相结合并按照最小化诊断误差平方和来形成最优的组合模型,通过实验得到的结果可以看出组合诊断模型能够对单一方法带来的缺陷进行弥补,并与这三种单一诊断方法相比,具有更高的故障识别率,鲁棒性也更好。最后利用云平台技术解决了组合复杂模型带来的运行时间长等问题,有效地提高了故障诊断的效率。
[Abstract]:At present, with the development of social economy and science and technology, all kinds of motors appear in industrial production and people's daily work, and it plays a more and more important role. Once the fault of motor occurs, light will affect people's production and life, heavy will endanger the safety of human life and cause serious economic losses. Therefore, in order to meet the high quality demand of motor in industrial automation, it is of great significance to study the method of fault diagnosis of motor in modern life. With the development of science and technology, motor diagnosis needs more accuracy and rapidity, which leads to some technical problems in the field of motor fault diagnosis. For example, the inaccuracy of feature extraction of high-dimensional fault data leads to the difficulty of low diagnostic accuracy, the limitation of single diagnosis method, and the low computational efficiency. Based on the above situation, this paper mainly studies isospectral manifold learning algorithm, wolf swarm algorithm, combined diagnosis model and cloud computing to solve the problem, and finally through the case of Western Reserve University data as an example to verify the problem. The specific contents are as follows: firstly, aiming at the problem of low diagnostic accuracy caused by the imprecision of the high dimensional data of motor fault extraction, the isospectral manifold learning algorithm is introduced to reduce the dimension. That is to say, this algorithm is used to reduce the dimension of the data after dimension reduction by principal component analysis (PCA). The algorithm constructs the adjacent graph by its modified sparse reconstruction weight matrix, which makes the similar samples gather more after dimensionality reduction, and the different classes of samples are more evacuated. The dimensionality reduction of high dimensional data is realized effectively, and the results are compared with that of principal component analysis (PCA). Then, aiming at the problem that the diagnosis accuracy is easily affected by the parameters of the isospectral manifold learning algorithm and the least squares support vector machine, the parameters are optimized by the wolf swarm algorithm. Fisher criterion function is used to optimize the parameter combination of least squares support vector machine, while the fitness function used in optimizing the parameters of isospectral manifold learning algorithm is the recognition rate of nearest neighbor classification. The optimal diagnostic model is established by using the optimized parameters. The simulation results show that the model has good diagnostic results, and the wolf swarm optimization algorithm and particle swarm optimization algorithm are compared with each other in the optimization of parameters, and the diagnosis effect is remarkable. Finally, aiming at the diversity of motor fault symptoms and the limitation of single diagnosis method, the combined diagnosis model is adopted to solve the problems. In this paper, the least square support vector machine, fuzzy neural network and RBF neural network are combined to form the optimal combination model according to minimizing the sum of diagnostic error squared. The experimental results show that the combined diagnosis model can make up for the defects brought by the single method, and compared with the three single diagnosis methods, it has higher fault identification rate and better robustness. Finally, the long running time caused by the combined complex model is solved by using cloud platform technology, and the efficiency of fault diagnosis is improved effectively.
【学位授予单位】：兰州理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM307

【参考文献】