Research On Fault Diagnosis Based On Integrated Wavelet Neural Network In Aluminum Electrolysis

The process of aluminum electrolysis is a non-linearity, multi-couplings, time-variable and large time-delay industrial process. Due to the process of aluminum electrolysis is complex, a variety of faults may occur. If some kind of fault is in the event, it may have a great impact on the whole electrolytic series, which will cause the current efficiency to drop, affect various technical indicators of the electrolysis series, reduce the yield and quality of aluminum, and destroy stable power supply of the whole. Therefore, timely and exactly fault diagnosis is of great significance in ensuring smooth and safe production of the electrolytic series and energy conservation.Due to its nonlinear characteristics in aluminum electrolysis, it is difficult to establish exact mechanism model for analysis. However, by the analysis of various characteristic volumes in the process of aluminum electrolysis, it is feasible to identify the model through artificial intelligent algorithm by analyzing relevant variable which could reflect the conditions of aluminum electrolysis. For instance, some researchers have put rough set, Back-Propagation neural network, Elman neural network and wavelet neural network in use for fault diagnosis. Although certain results have got, these models can be used only for single fault. When they are used for compound fault, these models are not convergence sometimes and it is easy to fall into local minimum. Besides, it can be easily omitting input which reflects feature information of fault in aluminum electrolysis, leading to low accuracy and real-time which is not permitted in practical engineering. Because of this, this paper puts forward a new fault diagnosis method and a corresponding dealing way for the problem consisted in president fault diagnosis system. The author's jobs are listed as following:Firstly, this paper has outlined the development status of aluminum electrolysis both domestic and international, also illustrated the significance of fault diagnosis in aluminum electrolysis and discussed the application and problems of existing fault diagnosis method. Besides, the necessary of improving existing fault diagnosis method is depicted.Secondly, the author has described faults diagnosis method in control system for theoretical study. Faced with various faults in aluminum electrolysis, the author has made detailed analysis of the causes and prevention methods. Besides, characteristics of faults diagnosis are extracted and methods in this paper are presented.Thirdly, the possibilities and necessities of neural network in fault diagnosis are discussed. The author has expounded the combination of wavelet analysis and neural network and discussed common methods in anode effect prediction, also pointed out the deficiencies of current anode effect prediction method.Fourthly, according to the mechanism of anode effect, fault diagnosis model is established by using Elman neural network and wavelet Elman neural network respectively. Basic Elman neural network model is modified in order to achieve the desired effect of fault diagnosis. Simulation results show that wavelet Elman neural network has promoted accuracy and real-time of fault diagnosis and has a better accuracy and reliability. Fifthly, according to the specific features of aluminum electrolysis, the author has presented faults diagnosis method based on integrated wavelet neural network, and designed human interface. Simulation results show that this method can be effective in multi-fault diagnosis, which has solved staggering problem in single fault diagnosis.Sixthly, conclusion and analysis of the fault diagnosis technology and then imagine the next research job.