Applying Reinforcement Learning to Plan Manufacturing Material Handling Part 2: Experimentation and Results

被引:2
作者
Govindaiah, Swetha [1 ]
Petty, Mikel D. [1 ]
机构
[1] Univ Alabama Huntsville, Comp Sci, Huntsville, AL 35899 USA
来源
PROCEEDINGS OF THE 2019 ANNUAL ACM SOUTHEAST CONFERENCE (ACMSE 2019) | 2019年
关键词
Material handling; machine learning; reinforcement learning; planning; multi-objective learning;
D O I
10.1145/3299815.3314427
中图分类号
TP301 [理论、方法];
学科分类号
081202 ;
摘要
Applying machine learning to improve the efficiency of complex manufacturing processes, particularly logistics and material handling, can be a challenging problem. The interconnectedness of the multiple components that compose such processes and the typically large number of variables required to specify procedures and plans within those processes combine to make it very difficult to map the details of real-world manufacturing processes to an abstract mathematical representation suitable for machine learning methods. In this paper, we report on the application of machine learning methods, in particular reinforcement learning, to generate increasingly efficient plans for material handling to satisfy temporally varying product demands in a representative manufacturing facility. The essential steps in the research included defining a formal representation of a realistically complex material handling plan, defining a set of suitable two-stage plan change operators as reinforcement learning actions, implementing a simulation-based multi-objective reward function that considers multiple components of material handling costs, and abstracting the many possible material handling plans into a state set small enough to enable reinforcement learning. Extensive experimentation with multiple starting plans showed that the reinforcement learning process could consistently reduce the material handling plans' costs over time. This work may be one of the first applications of reinforcement learning with a multiobjective reward function to a realistically complex material handling process. This paper first provides an explanation of how the material handling plans and rewards were abstracted into a manageable state set. It then details the various initial plans and experimental trials used to test the plans. Finally, it reports the results of those experimental trials, including the plan change policies learned and the reductions in material handling costs achieved.
引用
收藏
页码:16 / 23
页数:8
相关论文
共 7 条
  • [1] Bellman R.E., 1957, DYNAMIC PROGRAMMING
  • [2] Cormen T. H., Introduction to Algorithms, V2nd
  • [3] Govindaiah S., 2019, P 2019 SIM INN WORKS
  • [4] Applying Reinforcement Learning to Plan Manufacturing Material Handling Part 1: Background and Formal Problem Specification
    Govindaiah, Swetha
    Petty, Mikel D.
    [J]. PROCEEDINGS OF THE 2019 ANNUAL ACM SOUTHEAST CONFERENCE (ACMSE 2019), 2019, : 168 - 171
  • [5] Sutton R. S., 1998, Reinforcement Learning: An Introduction, V2
  • [6] Sutton RS, 2018, ADAPT COMPUT MACH LE, P1
  • [7] Empirical evaluation methods for multiobjective reinforcement learning algorithms
    Vamplew, Peter
    Dazeley, Richard
    Berry, Adam
    Issabekov, Rustam
    Dekker, Evan
    [J]. MACHINE LEARNING, 2011, 84 (1-2) : 51 - 80