Intelligent Optimization Method For Dynamics Of Multi-pass Roll-die Forming Electromechanical Syestem

Variable section roll-die forming equipment is a complex mechanical and electrical equipment for sheet metal gradual forming process with variable section.The processed products have many kinds and excellent load performance,so they have a wide application prospect.Therefore,it is of great significance to analyze and optimize the acceleration of the five-pass variable section roll-die forming machine,reduce the peak acceleration,and improve the stability of equipment operation,so as to improve the product quality.Thesis uses the reinforcement learning intelligent optimization method to optimize the roll-die forming machine.Firstly,based on the energy conservation principle,the potential energy,kinetic energy,electric field energy,magnetic energy,mechanical loss energy,and electromagnetic loss energy are calculated in the electromechanical system of roll-die forming machine,and the set of differential equations of dynamics of rolldie forming machine is obtained.Based on the kinetic equations of the roll-die forming machine,the optimization objective function and the parameters to be optimized are determined.Finally,the set of dynamics equations of the roll-die forming electromechanical system is defined as the environment,the parameters to be optimized are defined as the action,the absolute peak value of acceleration is defined as the state,and the reward function is obtained according to the optimization objective function to establish the reinforcement learning model of the dynamics of the roll-die forming electromechanical system.For the optimization problem of such high-dimensional and continuous action space as the optimization of electromechanical system dynamics,thesis adopts the DDPG algorithm to optimize the dynamic model of roll-die forming machine.Optimization was performed according to the DDPG algorithm and the algorithm obtained local optimization results.Therefore,the DDPG algorithm is improved: 1.only data with reward values greater than 0 are stored in the experience pool.2.dynamic reduction method is adopted for action disturbance.Through these two improvements,the peak acceleration obtained by the improved DDPG algorithm is about 28.13% lower than that without optimization.In order to verify the effectiveness of the reinforcement learning method in the field of electromechanical system dynamics optimization,thesis also compares the optimization results of the PSO optimization algorithm with those of the improved DDPG algorithm,and the difference between the acceleration peaks after optimization is less than 1%.This shows that reinforcement learning can be used in the field of electromechanical system dynamics optimization with good results.Thesis not only provides a theoretical basis for the high-volume industrial production application of the five-pass variable section roll-die forming machine but also provides a new method for the optimization of electromechanical system dynamics.