Research On Metallurgic Fan Machinery Intelligent Diagnosis System

Metallurgic fan machinery is the pivotal device during the iron and steel production. Because working in the tough environment of high temperature, heavy pressure, and especially heavy metallic dusts, metallurgic fan machinery is easy to get out of order. With the application of artificial neural networks, information fusion, clustering analysis, computational fluid dynamics, etc, the intelligent diagnosis methods, the integrated dynamic neural networks model, the child neural networks algorithm and the decision fusion algorithm in the integrated study system, and the blade wearing fault simulation method of metallurgic fan machinery are studied in this thesis. The intelligent diagnosis system based on integrated dynamic neural networks is built. The main contents of this thesis are summed up as following:1. The fault characteristics of metallurgic fan machinery are analyzed; the current methods for metallurgic fan machinery faults diagnosis are reviewed. On the basis of the research of faults diagnosis mechanism of neural network, the merits and the deficiencies of BPNN and RBFNN are analyzed through the three aspects of study ability, diagnosis ability and antinoise ability of neural networks. It is shown the single structure of neural network has the bad performance on concurrent fault diagnosis.2. The intelligent diagnosis method based on integrated dynamic neural networks is proposed with the combination of information fusion and neural network, thereby the integrated dynamic neural networks intelligent diagnosis model of metallurgic fan machinery is built. It is proved with information entropy method that the results are more accurate and efficient than single-sensor neural network diagnosis results. In terms of the abstract of the intelligent diagnosis model, the integrated study system composed of the classing unit and fusion unit is built.3. The classing algorithm and the fusion algorithm are researched. With the adoption of the RBFNN in the classing unit, the nonlinear mapping ability and the diagnosis speed are improved through dynamic diagnosis process. In the fusion unit of the integrated study system, it shows that if appropriate weight coefficients are adopted to fuse the outputs of the multiple children neural networks, the system diagnosis accuracy can be improved. But it is incapable to finish decisional fusion task with unsupervised algorithm because of lacking prior knowledge.4. The method of using unsupervised clustering analysis to realize decisional fusion is proposed. Aimed at the deficiency of traditional HKM algorithm and FKM algorithm, the method of combining fuzzy theory and immune network theory is proposed, and the clustering fusion algorithm based on fuzzy immune network is built. It is verified that this algorithm can not only classify samples automatically but also has good recognizing ability for the samples with fuzzy borders. The validity of the algorithm is proved by the field diagnosis examples.5. To solve the difficulties of getting blade wearing samples of metallurgic fan machinery, the flow field simulation method based on FLUENT and ADAMS/Vibration is proposed. With the analysis and calculation of metallurgic fan machinery flow fluid, the precise 3D model is built. The moving traces of different particles and the blade wearing status are simulated and the relative faults samples are acquired.6. With the combination of the research products and the engineering projects, the remote condition monitor and intelligent diagnosis system of metallurgic fan machinery is developed. This system can solve the difficulties of acquirement of diagnosis knowledge, improve the tolerant ability and diagnosis accuracy, and enhance the real-time ability in conventional faults diagnosis system. The project reaches the international advanced level and is awarded as the third science and technology progress prize of Hubei province in 2007 year.