Study On Molten Pool Level Control Of Twin-roll Casting Based On Neural Network

The twin-roll casting process combines the traditional casting and rolling processes so that the liquid metal is cast into a 1?6mm sheet in a very short time.Compared with the traditional process,twin-roll casting technology has the advantages of short process,low cost,high production efficiency and less environmental pollution.At present,the commercial twin-roll casting production line has been established in some developed countries,but the development of this technology is seriously limited due to the problems of testing equipment and controlling model precision.Twin-roll casting technology in our country starts late,and there is a big gap between our country and foreign countries in control model and integrated control level.The fluctuation of the liquid level will have a direct influence on the subsequent condensation and rolling of the metal material.In order to produce high quality sheet metal and maintain the stability of the liquid level,it is necessary to precisely control the molten pool level.The research object of this topic is the twin-roll caster developed by the magnesium alloy laboratory of University of Science and Technology Liaoning,adopting neural network control method to solve the difficult problem of self-adaptive control of molten level,which is not easy to realize by traditional control method.Aiming at the problems of neural network control,a multi-dimensional PSO algorithm is further proposed to adjust the parameters of network controller.In order to improve the search accuracy of algorithm,a PSO algorithm based on variable inertia weight is used to improve the performance of the controller.In this paper,firstly the twin-roll casting technology is introduced,the difficulties and significance of level control in molten pool are analyzed,and the research of level control and the development of related technology are briefly introduced.The mathematical model of level system is established and the influence of process parameters on liquid level is analyzed,then the control strategy based on BP-PID neural network is used to control the level.Aiming at the problems existing in BP control,a RBF neural network control strategy based on Kmeans algorithm is proposed.Compared with the BP-PID control strategy,the control effect is better,but there are still some problems,such as larger overshoot,delayed response and so on.Finally,PSO algorithm with variable inertia weight is used to optimize the performance of the Kmeans-RBF controller,the design method and realization method of the controller are introduced in detail.The experimental results show that the optimal control strategy proposed in this paper has higher accuracy and stability.