Globally Optimized Production by Co-operating Production Agents Based on Bellmans Principle

摘要

The production of items is usually separated into a sequence of processing steps from raw materials to the finished product. Each of the processing steps is executed by dedicated machines where the output of one machine is the input of the next machine. The total effort of all processes can be drastically reduced and the resulting quality of the end product be maximized by exploiting the mutual dependencies of the individual process steps. The concepts of task-driven, intelligent production agents are extended to account for this global optimization task, maintaining the autonomous decision of the individual agent about the optimal process parameters. This can be reached by supplying the local production agent with information about the effect of some of its output on the efforts of the subsequent processes and with information about the actual input to be processed. When the process agent knows the efforts related to its own parameters required to transform the input into some output states, the overall effort can be minimized. Stochastic process influences turn the optimization into a Markov decision process where Bellmans equation can be applied to yield on average the best total result at lowest effort. The encountered exponential complexity when solving Bellmans equation via Dynamic Programming is relieved by Approximate Dynamic Programming. By looking upon one single process, as a process chain with discrete, repetitive steps with different process parameter values, the same optimization concept can be applied to control the individual process. Agents using this optimization scheme require special capabilities: output state estimation, state transformation function representation, Bellman optimization and assessment function representation (assigning effort to process output). The concepts, the architecture, the required components and the methods will be presented in this paper.