Deep Reinforcement Learning Based Multi-Objective Modeling And Optimization Methods For Quality And Efficiency Improvement

The intelligent processing mode of improving quality and efficiency is an important research for enterprises to adapt to social needs and fierce competition,which can effectively promote the flourishing development of advanced machining technology in automotive manufacturing and other related fields.The selection of process parameters(spindle speed,feed,etc.)has a significant impact on machining performance,and the diversity of parameters combination makes the machine tools have a large space for improvement.How to select reasonable and effective machining parameters is the subject of this research paper.In order to achieve multiple trade-offs of quality,efficiency and performance in the machining process,this paper taking milling TC18 titanium alloy as an example,an optimization model of improving quality and efficiency based on deep reinforcement learning is proposed.The parameters are optimized and modeled by using Double Deep Q-Network(DDQN),which meets the requirements of minimum surface roughness,maximum material removal rate and optimal milling force stability.The results are further evaluated by entropy method.(1)The back-propagation neural network(BPNN)algorithm optimized by DDQN is used to build the prediction model of quality feature(surface roughness)and performance feature(milling force stability).DDQN is used to optimize the important internal parameters(number of hidden layer neurons and learning rate)of BPNN.The reinforcement learning operation strategy for internal parameters optimization is designed,and the influence of parameters selection is analyzed to realize the accurate prediction of the objective function of the multiobjective optimization model.(2)DDQN is used to solve the multi-objective optimization model of machining parameters,and the Pareto frontier solution set satisfying the minimum surface roughness,maximum material removal rate and optimal milling force stability is obtained.The strategy of reinforcement learning for multi-objective optimization is designed,and compared with the common multi-objective algorithm,the effectiveness of the optimization model is verified.Entropy method is used to further analyze the Pareto front solution,and the best processing parameters are obtained for processing guidance.(3)An optimization decision scheme under the efficiency threshold is proposed,in which the secondary objectives of multiple optimization goals are used as constraints to improve the convergence speed of optimization,and different machining parameter solutions are given for different cases of optimization goal focus.This approach can provide cost-effective,flexible and scalable machining solutions for enterprises.