Research On Fault Diagnosis Method For Rotating Machinery Using Variable Predictive Model Based Class Discriminate

With the development of science and technology, fault diagnosis technology has gradually become one of the core support technology to guarantee reliable and safe operation for rotating machinery equipment, and plays a decisive role in the modern production. Therefore, the research on new technology and new method for rotating machinery fault diagnosis has important theoretical and practical significance.The essence of the rotating machinery fault diagnosis technology is a problem of pattern recognition. After extracting appropriate sensitive fault feature, different methods of pattern recognition are applied. But their precision and stability of fault diagnosis have greatly difference. Some pattern recognition methods, such as artificial neural network(ANN), support vector machine(SVM) and so on, are generally used. However, these methods exist respective limitations, and they don’t make good use of the inner relations among characteristic variables. In fact, the features extracted by modern signal processing method tend to have certain mutual inherent relationship, and for different system or state, the inherent relationship between each other has obvious different mathematical expression.Variable predictive model based class discriminate(VPMCD) is a new method of pattern recognition. VPMCD method can make full use of the intrinsic relationship between features to establish mathematical variable predictive model(VPM) for class discrimination. Supported by National Natural Science Foundation(No. 51175158), this dissertation has taken deeply and systematically research on the key theory of VPMCD method and application to rotating machinery fault diagnosis with small samples and multiple class. The main research contents and innovation points are as follows:(1) The basic principle and specific algorithm of VPMCD pattern recognition method is studied, the characteristics of VPMCD is summarized, and VPMCD method is compared with ANN and SVM, the comparison results show that VPMCD method has many obvious advantages in classification performance and computing speed.(2) Aiming at the shortcomings of estimation approach for VPM’s parameters in original VPMCD method, the weight least square parameter estimation method is introducted to replace least squares parameter estimation method to improve original VPMCD. The simulation results show that the higher model fitting precision can be obtained by the improved VPMCD under less training samples.(3) According to the specific problems of rotating machinery fault diagnosis, combined with the latest modern signal processing technology, a variety of feature extraction methods are put forward: LMD energy moment feature extraction method, improved improved ITD feature extraction method, fuzzy entropy feature extraction method with LCD, LCD-SVD feature extraction method and multi-scale higher order singular spectrum feature extraction method. Various models of fault diagnosis based on improved VPMCD are proposed with the above feature extraction approaches. The applications have verified that the proposed models can be effectively used for rotating machinery fault diagnosis.(4) VPMCD method uses a single model as VPM and has inadequate information utilization in the process of model type selection. In view of the problem, GA-VPMCD method for classification is proposed in this paper. Firstly, Re-substitution or cross validation method is used by combining with model testing. The models with highest classification accuracy and highest fitting precision are selected as the weak VPMs. Then, model fusion is adopted and the genetic algorithm is applied to fuse predictive values of various weak VPMs to achieve classification. Combined with the order envelope analysis technique, GA-VPMCD method is applied into intelligent fault diagnosis of rolling bearing under variable speed condition. Combined with the multi-scale higher order singular spectrum analysis, GA-VPMCD method is applied into the rotor fault diagnosis. The experimental results show that these proposed methods can effectively improve the fault diagnosis accuracy and stability.(5) In view of feature selection, ANN- MIV-VPMCD method for classification is proposed with combination artificial neural network and mean impact value(MIV) and VPMCD. Furthermore, the fault diagnosis model based on LCD-SVD technique and ANN-MIV-VPMCD method is proposed and is applied to the rolling bearing fault diagnosis. The experimental results verify the effectiveness and superiority of the proposed method.(6) In the most cases, there only exist normal samples or lack of typical fault samples during rotating machinery fault diagnosis. Aiming at this problem, OC-VPMCD method for novelty detection is put forward in this paper and the OC-VPMCD method is applied to the novelty detection of rotating machinery. The experimental results show that OC-VPMCD method can be effectively applied to the novelty detection of rotating machinery.