Research On Diagnosis System For Aluminum Reduction Cell Based On Genetic Neural Network

Aluminium reduction cell is the main production equipment. Whether the status of the aluminium reduction cell is normal, not only related to the economic and technical indicators, but also affect the life of aluminium cell and day-to-day production. Aluminium reduction cell is a non-linear, multi-coupling, and the time-varying time-delay system of industrial process. In the electrolytic process, it formed a complex multi-Change the status of the cell features, such as anode effects, anode disease, cold cell, heat cell and unstable cell, etc. It is difficult to be detected, and some faults will result in huge economic losses and the impact on production safety. Therefore, this paper mainly investigated the fault diagnosis of the aluminium reduction cell and designed of the software system.In this paper, the principle of electrolytic production and its fault modes was introduced and analysis. Firstly, the author set up the BP neural network model for the fault diagnosis of the aluminum reduction cell. Because it is hard to conform the initial value of the BP neural network, and it is easy to local minimum, slow convergence defects and shortcomings. Thus, the paper used genetic algorithm to optimize the BP neural network initial value. It can effectively accelerate the learning speed and enhanced recognition of the generalization ability. The fault diagnosis algorithm is programmed by the MATLAB, and the author used the COM Builder to compiler it to COM components. Thus, it can be used by Visual C++.Secondly, the author designed and realized the fault diagnosis software system with Viusal C++ and SQL Server 2000, which can be divided into sub-system fault diagnosis and data acquisition systems. The system mainly composes of many modules, such as real-time data display and dynamic curve modules, historical data query module, printing report forms module, online help module, etc. Then, the author made the software installation package.Finally, the BP neural network which has not been optimized reach the appointed accuracy after 3131 training, when run-time is 388.282413s, The BP neural network which has been optimized reach the appointed accuracy after 2571 training, when run-time is 221.996945s.The training times reduced 18% and the run-time reduced 43%. According to 30 test samples, the accuracy of fault diagnosis is about 80%. The system made a result for the fault diagnosis under the test environment and test project. The result is basic right, and the author gives the user interface and test results in the end.