Research On Fault Diagnosis Method Based On Information Fusion And Application

In the producing industry, in order to make good use of the equipment, people always hope to predict the equipment fault, to nip in the bud, ensure that the equipment runs under the condition of safety, stabilization, long-period and full-load, and to void "over-maintain" or "ill-maintain". So the discipline of equipment fault diagnosis comes into being and attracts large numbers of scientific researcher to research on it, so as to keep it on being developed and perfected.Along with equipment's development towards high-speed, high-power, high-stabilization and large-scale, it's difficult for the traditional method to diagnose the complex equipment fault. In view of this, we introduce the theory of information fusion into the field of fault diagnosis.Firstly, discuss the universality and inevitability of information fusion from the viewpoint of biology and philosophic, and application of information fusion in fault diagnosis proves the feasibility and validity from the angle of information theory. On account of characteristic of fault diagnosis, we construct the generalized frameworks for information fusion fault diagnosis with five basic kinds of fusion process. Based on this, we discuss several methods of information fusion fault diagnosis in detail.1. Information Fusion Fault Diagnosis Based on Neural Network. Firstly we analysis thoroughly characteristic and shortcoming of BP network model. To the problem that it is difficult to construct network, we propose the NSIC(Network Structure Information Criterion) which provides a effectively method of optimizing structure. And in view of much training time and easily stacking into the minimal value locally, we propose a high-speed training algorithm which is very fit for multi-sample training, based on error modified grads-oriented descending. Based on analyzing the existent training algorithm of RBF, propose Self-Adaptive Selection of Clustering Core Algorithm which can self-adaptive select network structure and the best clustering cores, and make the network possess the simplest structure among these that meet classification need. Finally from the viewpoint of information theory, discuss the application principle and the method step of neural network in fault diagnosis, and this method prove practicability and validity by a example.2. information fusion fault diagnosis based on Dempster-Shafer Evidence theory. Present the basis concept and evidential fusion reasoning method in Dempster-Shafer theory, and establish application frameworks of multi-fault-characteristic information fusion applying into general fault diagnosis. Develop a method of construct the basic probability assignment which is fit for fault diagnosis. Finally this method proves high reliability and low uncertainty by multi-characteristic fusing.3. hierarchical information fusion fault diagnosis. With analyzing shortcoming of neural network and D-S theory respectively and their complementary character, we develop a method of hierarchical information fusion fault diagnosis based on neural network and D-S Evidential theory, and discuss the principle and steps in detail. The method prove good veracity and reliability of diagnosis.4. Group Decision theory and application in fault diagnosis. By introducing Group Decision theory into the field of fault diagnosis as a kind of information fusion decision method, we solve the problem of consistency and fusing decision of experts' opinion in multi-expert consultation. Propose Tree-shape Lingual Variable Fuzzy Score Method and data-analyzing method fitted for fault diagnosis. At last the example illustrates applying process and steps of Group Decision theory in the field of fault diagnosis.In order to overcome the shortcoming of traditional method of gearbox performance examining such as long testing-time, great expense, and unfit for modern large-scale production, we propose and develop "Automobile Gearbox Performance Examining System" which is a integrated examining system including examining, on-line monitoring, signal analyzin...