US 12,455,556 B2
Device and method for scheduling a set of jobs for a plurality of machines
Ayal Taitler, Haifa (IL); Christian Daniel, Leonberg (DE); Dotan Di Castro, Haifa (IL); Felix Milo Richter, Heidenheim (DE); Joel Oren, Tel Aviv (IL); Maksym Lefarov, Stuttgart (DE); Nima Manafzadeh Dizbin, Istanbul (TR); and Zohar Feldman, Haifa (IL)
Assigned to ROBERT BOSCH GMBH, Stuttgart (DE)
Filed by Robert Bosch GmbH, Stuttgart (DE)
Filed on Feb. 19, 2021, as Appl. No. 17/179,702.
Claims priority of application No. 102020204351.5 (DE), filed on Apr. 3, 2020.
Prior Publication US 2021/0312280 A1, Oct. 7, 2021
Int. Cl. G05B 19/418 (2006.01); G06F 18/21 (2023.01); G06F 18/243 (2023.01); G06N 3/04 (2023.01); G06N 3/08 (2023.01); G06N 5/01 (2023.01)
CPC G05B 19/41865 (2013.01) [G06F 18/217 (2023.01); G06F 18/24323 (2023.01); G06N 3/04 (2013.01); G06N 3/08 (2013.01); G06N 5/01 (2023.01); G05B 2219/32335 (2013.01)] 9 Claims
OG exemplary drawing
 
1. A method for scheduling a set of jobs on a plurality of machines, each job being defined by at least one feature which characterizes a processing time of a job, the method comprising:
when any of the plurality of machines is a free machine:
selecting a job from of the set of jobs to be carried out by the free machine,
scheduling the selected job on the free machine, wherein each of the plurality of machines is a cutter for cutting or a gun drill for drilling in a manufacturing system,
wherein the job is selected by:
receiving as input, by a Graph Neural Network, the set of jobs and a current state of each machine of at least one of the plurality of machines, the at least one of the plurality of machines including the free machine,
outputting, by the Graph Neural Network, rewards for each job launched on the at least one of the plurality of machines, which states are inputted to the Graph Neural Network, and
selecting the job for the free machine depending on the Graph Neural Network outputted rewards to carry out the selected job at a minimized total completion time, wherein a parametrization θ of the Graph Neural Network is optimized by deep Q-Network learning, and wherein for selecting a subsequent job, a Monte-Carlo Tree Search is applied, wherein the Monte-Carlo Tree Search builds iteratively a search tree starting from the current state of each of the plurality of machines and the set of jobs, wherein for expending the search tree, the outputted reward of the Graph Neural Network is used as a search heuristic, wherein the subsequent job involves a physical performance of a task by the free machine, and wherein depending on the search tree the subsequent job is selected; and
controlling the free machine to perform the physical performance of cutting or drilling in the subsequent job using a generated control signal
performing the physical performance of the subsequent job by the free machine, wherein the free machine includes a robot, wherein a control signal is generated in accordance with the subsequent job, and wherein the performing includes applying the control signal for causing the robot to execute the physical performance of the subsequent job.