Blast Furnace Fault Diagnosis Based On Fuzzy Neural Network

Iron and steel industry is a pillar industry of national economy, and the blast furnaceprocess is an important iron and steel smelting process. It takes place in a closed environmentand there occurs complex physical and chemical reactions, which will cause large fluctuations.In case the blast furnace fault happens, there will cause huge economic losses. To make surethe stability state of the blast furnace, it will be very meaningful for fault diagnosis of blastfurnace.The state within the blast furnace, can neither be directly observed nor directly detected bythe instrument. It is only can be inferred through a kind of diagnostic model established bychanges of parameters detected outside the blast furnace.In this paper, we take use of a kind ofexpert system based on fuzzy neural Petri net for fault diagnosis. Considering that neuralnetwork has advantage of adaptive, self-learning and fuzzy control of high logical reasoningand that can express the fuzzy semantic, we also establish a kind of diagnostic model based onfuzzy neural Petri net. For the blast furnace has such characteristic that we take use of it intoresearch on aura forecast analysis principle. In this paper, attempt of the three aspects is asfollows:(1) Five major faults and their corresponding signs have been analyzed and discussed.Considering the principle of the blast furnace process and the extraction of the characteristicquantities, we select a set of characteristic suitable for the blast furnace fault diagnosis. Afterthe characteristic having been analyzed, we can have the fault such as nodulation, hearthaccumulation, pipeline failures and furnace heated diagnosed.(2)Considering the advantage of both fuzzy Petri net and neural network, we construct a setof expert diagnosed system based on fuzzy neural Petri net. The thinking of the“Pre-consulation, consulation, referral diagnosis” and classification diagnosis thinking hasbeen referred. To make the diagnosis more accurate, we take use of a set of stratified markwhile reasoning of fuzzy neural Petri net.(3)We take use of MFC of VC++6.0to design a set of simulation program corresponds toprevious diagnostic system. To do so, we can prove the feasibility of the diagnostic model. Forsome problems in the diagnostic system, some aspects that need to be improved have been putforward.