Dissertation > Excellent graduate degree dissertation topics show

Support Vector Data Description fault diagnosis

Author: WangKun
Tutor: LiLingJun
School: Zhengzhou University
Course: Mechanical and Electronic Engineering
Keywords: Support Vector Data Description Empirical Mode Decomposition Single classification Intelligent Fault Diagnosis
CLC: TH165.3
Type: Master's thesis
Year: 2009
Downloads: 30
Quote: 0
Read: Download Dissertation

Abstract

Support Vector Data Description (Support vector data description, SVDD) is different from the traditional pattern classification methods. Traditional pattern classification methods generally require the use of two types of sample (or samples), through two types of samples to determine the decision boundary. Most traditional classification of the degree of balance of the data requirements of a certain extent, its performance is far from ideal when the sample data of one type with little or no. And the SVDD only need a class of samples can establish the classifier and the target sample and non-target samples to distinguish. For some critical equipment is not allowed to the failure, or that the failure rate is very low. Will be the SVDD used in machine fault diagnosis and condition monitoring, will be expected to solve the problems encountered by traditional pattern recognition methods in the absence of fault samples. . SVDD basic theory and algorithms, and calculations using the kernel function instead of the inner product to improve the flexibility of the classifier. Empirical mode decomposition advantages in dealing with non-stationary signals using empirical mode decomposition to decompose the signal, extract the energy of each band as a feature vector and used for training and testing SVDD experiments show that this method can retain the characteristics of the original signal, and achieved good classification effect. When a small amount of fault samples, we can create a traditional classifier can create SVDD classifier. But SVDD using only normal samples, and when failure sample is small and not representative of the failure of typical distribution, the traditional two classifiers and very difficult to achieve better performance. This article studied an improved SVDD - fault samples SVDD fault diagnosis, and by the analysis of the Rolling experimental data show that the method can effectively improve the accuracy of fault diagnosis. 4. Single classification method are three main theories: density estimation, boundary and reconstruction method. And evaluation and comparison of the performance of these methods in the processing of different data sets. Be seen by comparing the density estimation method is the most complete description of the data make, but too many samples may be required; the less boundary method to describe the boundaries of data should be given priority when the number of samples, especially SVDD; final reconstruction method is based on the distribution of the data model defined a distance or reconstruction error, this model can contain extra priori knowledge of the problem.

Related Dissertations

CLC: > Industrial Technology > Machinery and Instrument Industry > Machinery Manufacturing Technology > Flexible manufacturing systems and flexible manufacturing cell > Fault diagnosis and maintenance
© 2012 www.DissertationTopic.Net Mobile