The Research Of Fault Diagnosis System For 350kA Pre-baked Aluminum Reduction Cell

The large aluminum reduction is a very complex process, and its inner state is difficult to be judged and adjusted in time. The study of supervising the cell working state and its fault diagnosis is paid on much more attention. This paper made the spectrum analysis of the cell resistance, set up the BP neural network model for the fault diagnosis of the aluminum reduction cell and designed the fault diagnosis software system.On the bases of the analysis of the fault mechanisms of the aluminum cell, the paper pointed out the reason for the high and low frequency noises in the cell resistance signal. And three methods were used, i.e., Fast Fourier Transformation (FFT), Power Spectrum Density (PSD) estimation based on period gram and PSD estimation based on AR model, to analyze the spectrum of the cell resistance. The experimental results showed the main frequency domain of the cell resistance are 0~0.1Hz and there was a stable low frequency signal caused by alumina content change and polar distance regulation in the electrolysis operation;When the large fluctuation of fluid aluminum, there was an obvious wave crest in higher frequency;when abnormal anode, there were two obvious wave crests in higher frequency.Aiming at the three different states of 350kA large pre-bake aluminum reduction cell, the fault samples which took the frequency and spectrum energy as eigenvectors were collected and its corresponding BP neural network model was set up. The influences of the number of hiding layer node, learning rate etc were discussed during the course of samples training. By comprehensive analysis and comparing, a three-layer BP neural network was adopted. Input-layer contained ten nerve cells (the proportion of every 0.01Hz band energy in the whole energy between 0~0.09Hz and the normalization whole energy) and output-layer contained three nerve cells (normal state, fluctuation of fluid aluminum, abnormal state of anode). The model was tested by using the other real data of cell resistance, and the result indicated that the fault diagnosis system could reduce the training error to the aimvalue quickly and the cell faults could be diagnosed exactly and quickly.Programming combining with Visual C++, SQL Server 2000, MATLAB, the fault diagnosis software system was designed, which included three main modules: monitoring module, signal processing module and diagnosis module. The program realized displaying real time curve and history curve of cell signals, querying history report, spectrum analysis of cell resistance and fault diagnosis for the cell state.