Research And Realization Of Fault Diagnosis Method For Aluminum Electrolysis Process

Aluminum electrolysis is a complex industrial process, which characters with highly nonlinear, uncertainty, slow time-varying, large time delay, big interference and other negative factors. Fault occurs frequently and the type is varied. The occurrence of faults has significant impact on the whole electrolysis process and various technical indices, which leads to lower output and quality of aluminum, resulting in great waste of energy and big economic loss. However, there is no perfect fault diagnosis system for aluminum electrolysis currently. Therefore, effective fault diagnosis and prediction for aluminum electrolysis will have significance on improving the efficiency of aluminum production, reducing costs, saving energy and realizing safety in production.The fault mechanism and features of aluminum electrolysis has been deeply analyzed in this dissertation. Some subjects are studied in this dissertation viewing from different perspectives and with different ways, such as the fault diagnosis model, data processing, training of fault diagnosis model and optimization algorithm.Aiming at the problems in fault diagnosis of aluminum electrolysis, the research is carried on the theory and application of the fault diagnosis of aluminum electrolysis. The field data was used to test the results of study, and its effectiveness and practicality are verified.Firstly, according to the characteristics of aluminum electrolysis’s anode effect occurs, the fault diagnosis method of aluminum electrolysis based on analytical model is put forward, and the theory of system identification and the technology of fault diagnosis is applied.Recursive extended least squares identification algorithm is used, combining mechanism models with experiments, to establish a dynamic mathematical model of aluminum electrolysis process. Through tracking of multi-parameters model, the change characteristics of parameters is captured to predict anode effect. What else, the influence on the effects of fault diagnosis is studied, under the situations with different model initial conditions and different forgetting factors. The simulation results have verified the effectiveness of the proposed method.Secondly, for the complex features of the aluminum electrolysis production process, with many uncertain factors, nonlinear principal component analysis method is adopted to realize the data dimension reduction, decoupling, and input reduction. Then the principal component of aluminum electrolysis process fault is confirmed, the structure of neural network is simplified. Due to the characteristics of Elman neural network with strong memory, the fault diagnosis method of the aluminum electrolysis anode effect based on wavelet Elman neural network is put forward.Three different fault diagnosis methods are studied, which are based on the Elman neural network, wavelet Elman neural network and improved Elman neural network models. The simulation results indicate that wavelet Elman neural network has characteristics of high prediction accuracy and big advance amount.Thirdly, modular integrated fuzzy neural network multi-fault diagnosis method for aluminum electrolysis is put forward. An aluminum electrolysis fault diagnosis platform is constructed, using a structured, modular design method. The methods for fault diagnosis on two different forms are studied, such as single neural network and modular integrated network. Aluminum electrolysis fault diagnosis model of modular integrated fuzzy neural network adopts two layers network structure, which are fault diagnosis sub-network and decision fusion diagnosis network. It is characterized with the combination of modular and integrated network, the combination of fuzzy and neural network combination, the combination of sub-network and decision network fusion.According to the diagnosis functions of these sub-networks, the input signals are assigned to the sub fault diagnosis network. These sub-networks’relevance and the grades of these signals are considered. So, effectively combination is used to realize modular of sub fault diagnosis network. Elman neural network structure is used in fault diagnosis sub-network, and fuzzy neural network structure is used in decision fusion neural network. The main functions are signal transmission, fuzzy logic operation and anti-fuzzy. The organic combination of fuzzy network and neural network is realized. Particle swarm algorithm is adopted in training decision fusion neural network, to improve the convergence speed and accuracy. Simulation results indicate that comparing to the diagnosis method of single neural network, the modular integrated fuzzy neural network’s multi-fault diagnosis method has great advantages, such as high prediction accuracy, advance amount and can effectively diagnose complex faults.Finally, according to the aluminum electrolysis process technical requirements and control performance indices, a computer control system is designed for aluminum electrolysis process. Distributed control system structure is applied in the system, including the control level, monitor level and management level. System platform of aluminum electrolysis fault diagnosis are established, and two levels fault diagnosis mode for aluminum electrolysis are proposed. Double fuzzy fault diagnosis system is used first level fault diagnosis, to conduct a preliminary judge. Integrated fuzzy neural network diagnosis method is used in the second level fault diagnosis uses, to diagnose multi-fault diagnosis and classify the modes and determine the type of fault. The results show the effectiveness of the proposed method.