Optimal Control Of A Conveyor-Serviced Production Station With Dynamic Pickup Point Without Considering Delay Waiting

As an important flexible production system,conveyor-serviced production station(CSPS)has been widely used in real production,such as robotic assembly lines and cell production station.Therefore,the study of the optimal control of such type production model has important significance to the improvement of the system's flexibility.The physical model of the CSPS system is usually a model with fixed pickup point,the robot needs to wait for the part to reach the fixed pickup point before the unloading operation can be performed.With the intelligent development of the production line,the robot can unload the part in the working area,which effectively reduces the time of the unloading process and improves the productivity of the system.However,in the existing CSPS with dynamic pickup point,in order to satisfy the system's no aftereffect,the robot may generate invalid delay waiting time after each decision.The delay waiting time caused by this working mode can result in an extended production cycle,which reduces the system's production efficiency.Therefore,this dissertation proposes a working mode that does not consider delay waiting.By converting the delay waiting time into the forward shift of the look-ahead point,the problem of no aftereffect of the system is solved.By using the vacancy of the buffer and the position of the griper as the system state and the look-head as the control variable,a semi-Markov decision process(SMDP)is set up.After modeling,policy iteration algorithm is used to solve the optimal policy and system-related performance values are calculated according to the policy.In addition,due to the complexity of the system,especially the dynamic randomness,it brings great difficulties for modeling.In order to avoid the emergence of problems such as "Modeling Difficult",based on the SMDP model,this dissertation also uses the Q-learning algorithm to find the optimal solution or suboptimal solution.Through simulation,the experimental results are compared with the working mode considering delay waiting time.The results show that both the results obtained by the theoretical algorithm and the results obtained by the Q-learning are superior to the working mode of considering the delay waiting in terms of system performance.Therefore,the validity of the working mode proposed in this dissertation is verified.