The Intelligent Method Research Of Aluminum Electrolysis Fault Diagnosis

Aluminum Electrolysis industry is an important process industry, aluminum reduction cell is the core equipment in the production of aluminum electrolysis. Its operating conditions will directly affect the aluminum production operational efficiency. However, in the practical production, the aluminum electrolysis cell has a large volume and quality, which runs under the complex physical and chemical changes and the cell is subject to the strong electric field, magnetic field, high corrosion and external tough working conditions. A series of negative factors makes aluminum electrolysis a nonlinear time-varying system characteristic of high noise, multivariables, tight coupling, large time delay and large amount of data. Due to its actual special status in aluminum electrolysis industrial production, it is difficult to establish exact mechanism model for analysis. However, when single neural network is adopted to do the fault diagnosis, the network requires quite a few fault samples.Yet it’s no easy a task to determine the characteristic vectors of a variety of faults. What’s more, the network structure which is suitable for multiple fault diagnosis is difficult to determine. Even if it can be attained, the large number of nodes over the network and the huge structure will bring about the tough works in training.According to the fault characteristics of the aluminum electrolysis, a multi-fault diagnosis method of aluminum electrolysis which is based on integrated modular fuzzy neural network is proposed. Considering the shortages of a single network applied in multi-fault diagnosis, a multi-fault diagnosis platform with two layers of sub-network and decision fusion network is constructed in multi-fault diagnosis of aluminum electrolysis, combining fuzzy logic and neural network by the application of the concept of modular integration. Mixed Particle Swarm Optimization algorithm is adopted in the thesis so that the convergence speed and accuracy of the network can be increased to some extent. Simulation results show that the proposed method can improve the accuracy rate of fault prediction and give the prediction advance. The author’s jobs are listed as following:Firstly, this thesis discussed some of the background of the aluminum electrolysis process fault diagnosis, the development of aluminum electrolysis home and abroad in this stage, and clarified the necessity for fault diagnosis.Secondly, this thesis expounded the basic principle of fault diagnosis technology and some of the commonly diagnostic method in general, some of the deficiencies of single fault diagnosis network in this stage and proposed a multi-fault diagnosis method based on modular integrated fuzzy neural network aluminum electrolysis.Thirdly, this thesis stated an establishment of sub-fault diagnosis network method, which adopted Elman neural network as sub-fault diagnosis network which the number of neurons in the hidden layer was determined by genetic algorithm. This method result an optimization neuron model. And then this thesis introduced a new method to establish sub-fault diagnostic network. This method adopts mixed extension neural network based on the Matter Element Theory. And this thesis has a simple introduction and comparison.Fourthly, this thesis stated a method to establish the decision fusion neural network which adopted fuzzy neural network as a The decision fusion neural network to calculate the different input variables separately. This method fully extracts the information contained in the different data to achieve MA fuzzy reasoning to realize the fault diagnosis and decision-making.Fifthly, this thesis stated the defects in the network learning through ordinary particle swarm Optimization algorithm. And according to these shortcomings this thesis proposed an improved optimization algorithm, which is a fusion of SA algorithm, the PSO algorithm and gradient descent method, so that the convergence rate and accuracy of the network are increased to some extent.Sixthly, this thesis adopted MS VC++and ACCESS database to establish an aluminum electrolysis process control system host computer system, which is mainly directed against the fault diagnosis section.Seventhly, this thesis summarized and made a further analysis of multi-fault diagnosis method based on modular integrated fuzzy neural network of Aluminum electrolysis. And this thesis prospect of the future work.