【作者】 陈刚；

【导师】 唐昊；

【作者基本信息】 合肥工业大学 ， 模式识别与智能系统， 2020， 硕士

【摘要】 在现实工业领域的生产加工场景中,有一类由工人利用外骨骼机器人的动力辅助对生产线传送带上的工件进行搬运并分类存放的生产加工模型,称之为外骨骼人机生产线系统。外骨骼人机生产线系统作为第四次工业革命中一类具有代表性的智能生产加工模型,研究其建模与优化问题,具有重要的现实意义。本文主要考虑一类面向多品种工件生产加工的单站点外骨骼人机生产线系统人机助力的建模与优化控制问题。该系统中,多品种工件在传送带上随机到达搬运站点,工人在站点利用外骨骼机器人的动力辅助对生产线传送带上的工件进行搬运并分类存放,同时为保护产线工人因累积疲劳后带来肌体损伤等问题,引入了疲劳过度后强制休息一定时间的保护机制。系统的优化目标是通过选择外骨骼机器人最优的助力权重策略,优化与提升外骨骼机器人与工人的生理疲劳以及生产线生产加工效率之间的智能共融水平。论文首先考虑了外骨骼机器人外接电源驱动下的人机生产线系统,以外骨骼机器人的搬运阶段类型、产线工人的实时疲劳等级以及工件的品种类型作为系统的联合状态,外骨骼机器人的助力权重系数为控制决策变量,无穷时段期望代价最小为优化目标,将其助力权重系数分配的控制问题建模为半马尔可夫决策过程(SMDP)模型。并给出了该系统各搬运阶段下,产线工人肌肉疲劳评价的数学模型,该模型的建立为引入的基于模拟退火的Q-学习算法提供了模拟系统实际运行的数学基础。在仿真计算中,给出了 Q-学习算法的优化曲线,并分析讨论了不同策略以及各品种工件到达率参数变化下对系统性能的影响,仿真结果验证了所建模型的合理性与优化学习算法的有效性。为了满足外骨骼机器人在系统中全方位灵活移动的实际需求,进一步考虑了外骨骼机器人电池驱动下的人机生产线系统,系统自带电池驱动下灵活移动,但有限的电能容量使更换电池频率成为了影响系统性能的另一要素。因此本文进一步加入了电池剩余容量作为系统优化控制的联合状态,并给出了电池消耗模型和更换电池过程的肌肉疲劳评价模型。建立完整强化学习模型后,利用基于模拟退火的Q-学习算法求解,获得系统在平均准则下的最优助力权重系数策略。实验结果表明,通过我们建立的模型和给出的优化算法,系统可以得到最优或者次优助力控制策略。更多还原

【Abstract】 In real industrial production and processing scenarios,there is a production and processing model in which workers use the power assistance of exoskeleton robot to carry and classify and store the workpiece on the conveyor belt of production line,which is called the exoskeleton man-machine production line system.As a representative intelligent production and processing model in the fourth industrial revolution,it is of great practical significance to study the modeling and optimization of the exoskeleton man-machine production line system.This thesis mainly considers the modeling and optimal control of human-computer assist in a single site exoskeleton human-computer production line system for the production and processing of multiple kinds of workpiece.In this system,many kinds of artifacts at random to carry on the conveyor belt site,workers at the site using the exoskeleton robot dynamic auxiliary production line conveyor belt workpiece handling and classified storage,at the same time to protect the production line workers due to accumulated fatigue or even the body injury problems,the introduction of mandatory rest time after excessive fatigue protection mechanism.The optimization goal of the system is to coordinate and optimize the level of intelligent integration between the exoskeleton robot and the physiological fatigue of the workers and the relative productivity of the production line by selecting the optimal power weighting strategy of the exoskeleton robot.Thesis first consider the exoskeleton robot machine production line system driven by an external power source,outside the skeleton robot handling phase type,production line workers of real-time fatigue level and the variety types of the artifacts as a system of the United States,in the aid of exoskeleton robot weight coefficient to control the decision variables,the infinite time expected cost minimum as the optimization goal,its weight distribution coefficient of power control problem is modeled as a markov decision process(SMDP)model.The mathematical model of muscle fatigue evaluation of production line workers in each handling stage of the system is presented.The establishment of the model provides a mathematical basis for the actual operation of the simulation system based on the Q-learning algorithm based on simulated annealing.In the simulation,the optimization curve of the Q-learning algorithm is given,and the influence of different strategies and different kinds of workpiece arrival rate parameters on the system performance is analyzed and discussed.The simulation results verify the rationality of the model and the effectiveness of the optimization learning algorithm.In order to meet the practical needs of the exoskeleton robot to move flexibly in the system,the man-machine production line system driven by the exoskeleton robot battery is further considered.The system comes with its own battery for flexible mobility,but the limited power capacity makes battery replacement frequency another factor affecting system performance.Therefore,in this thesis,the residual capacity of battery is further added as the joint state of system optimization control,and the battery consumption model and the muscle fatigue evaluation model of battery replacement process are presented.After the complete reinforcement learning model was established,the optimal power contribution coefficient of the system under the average criterion was obtained by using the Q-learning algorithm based on simulated annealing.The experimental results show that the optimal or sub-optimal power assist control strategy can be obtained by the model and optimization algorithm.更多还原