Research And Application Of Fault Diagnosis Of Ultra-low-energy Aluminum Reduction Cell Based On Data-driven

With the scale of modern aluminum electrolysis industry becoming large and the cost of investment increasing, aluminum smelter factories pursue high security and high efficiency. Due to the complex production environment and the role of various uncertain factors, the fault occurs frequently in the aluminum reduction cell. If the abnormal conditions in the cell cannot be diagnosed and the control strategy be adjusted instantly and accurately, safety accidents would occur easily which leads to tragedy. However, the parameters are non-linear, strong correlated and dynamic, thus it is difficult to diagnose the abnormal state. At the same time, the frequent fluctuations due to the successful application of ultra-low-energy aluminum reduction cell and the obvious unsteady heterogeneous phenomenon make the accurate diagnosis much more difficult. On the basis of analyzing and summarizing the domestic and foreign researches, a method for the diagnosis of ultra-low-energy aluminum reduction cell based on data-driven is proposed. Through a variety of data processing and analysis methods, the inherent laws of ultra-low-energy aluminum reduction cell are mined to monitor and diagnose the fault and to further trace the root causes. Therefore, the decision of adjustment of the control strategy is provided to make the aluminum reduction cell run smoothly, reliably and safely.In this paper, the main research contents include the following aspects.Firstly, we propose method a diagnostic method for aluminum reduction cells based on optimize relative principal component analysis. An Effective principal of determining the relative weight is put forward,which takes advantage of relative principal component analysis reducing dimensions. In the method,genetic algorithm is used to optimize the fitness function about false alarm rate. The diversification of that sample project in principal component space and residual space is observed to acquire the best relative transforming matrix, so the false alarm rate of Hotelling’s T~2(T~2)test and Squared Prediction Error(SPE) would be reduced to the least.Secondly, a diagnostic model for status of aluminum reduction cells based on optimizing relative transformation matrix in feature subspace is proposed. In the method, firstly, kernel function is introduced to map the input space into a subspace via nonlinear mapping. Secondly,bacterial foraging optimization algorithm(BFO),, which is simply constructed and not easy to fall into the local minimum, is used to optimize relative transformation matrix for utmostly maintaining the features distribution of original data. By monitoring comprehensive indexes φthat consist of SPE and T~2 statistics, the proposed method is successfully applied to fault detection in aluminum electrolysis process.Thirdly, a model which is used to diagnose the status of aluminum reduction cells based on dynamic kernel relative principal component analysis method is proposed. In consideration of the dynamic characteristic of the parameters, the augmented matrix is built through the observation of variables on the current time which is extended by the past observation. Then the augmented matrix is transformed by using the optimal relative matrix, so the kernel principal component analysis can be used to diagnose the fault and trace the cause.Finally, the system which is used to monitor and diagnose the cell, is designed and developed based on dynamic kernel relative principal component analysis method. The system is developed based on MATLAB and C#, and can run in the environment of Windows2003 and above. It can achieve real-time monitoring and fault diagnosis for aluminum reduction cell, and provide decision basis for adjustment of control strategy. Through the experiment of 300 KA operating aluminum smelter in Chongqing Tiantai Aluminum Corporation Ltd., the system can make accurate and instant diagnosis of the abnormal situations in the groove, and so achieve the desired results.In conclusion, this paper provides an effective way to achieve accurate and instant diagnosis for the status in ultra-low-energy aluminum reduction cell.