Because of non-deterministic polynomial time hardness of job-shop scheduling problem (JSP), approximate optimization based on meta-heuristics has been actively discussed. Considering position of planners in production sites, it is desirable to develop a method in which their know-how is respected. An approach for meeting this requirement is to set the schedule generated by a planner as the initial solution and then gradually improve the solution by repeating a search in its neighborhood so that he/she can follow and thoroughly examine the improved solution. For this reason, this research is focused on scheduling using simulated annealing (SA). Because SA has a disadvantage that good solutions cannot be obtained efficiently if the initial solution has not been given appropriately, methods for solving this problem have been proposed for JSPs aiming at minimizing makespan. In high-mix low-volume manufacturing, it is also important to minimize production lead time to reduce work-in-process inventory. This research takes up production lead time defined as the time between the starting and the finishing times of a job considering strong constraint on places for putting works-in-process in production of large equipment, and deals with development of an efficient method using SA for JSPs aiming at minimizing the average value of the production lead times. Two methods of neighborhood limitation in SA for reducing the evaluation value were developed by focusing on waiting time of operations. It was proven that using one of the proposed methods in SA with appropriate probabilities is effective to JSPs of a certain size by numerical examples.
This study proposes a greedy-based approximation algorithm to determine the route of a land leveler, which is a type of agricultural machine for leveling and flattening the surfaces of fields. First, we describe the definition and the formulation of the land leveler problem (LLP) and show the exact solution based on dynamic programming for solving the LLP. Second, we propose the approximation algorithm and compare three visit functions which are used in the algorithm. Third, the comparison between the approximation algorithm and the exact solution is shown. Fourth, we apply the approximation algorithm to measurement data of a real field. Finally, we verify the validity of the movement constraint which is assumed for computing the route.
Because robotic experiments are often expensive in time and/or money, it is a common idea to use simulations instead of real robot experiments to have a robot acquire a motion through reinforcement learning. However, simulation models inevitably have some modeling errors, because of which the solution can be an inappropriate one for the real robot. As a solution to this problem, additional learning processes will be performed using the real robot in many studies, but for some robots and tasks, it will be difficult or infeasible. Therefore, learning methods that can find a robust solution without real robot experiments are desirable. This paper proposes a novel method to update the cost function so that the minimization of the cost will lead to a robust solution, only using simulations. As the method only modifies the cost, the convergence to a solution will not be a problem, unlike the existing method that is based on a similar idea. The validity of the idea is tested by simulations.
In this paper, we propose a data-driven design of feedforward controllers so that closed-loop characteristics of two-degree-of-freedom control systems can be theoretically guaranteed. The controller is designed by finite-time markov parameter matching. We first show an upper bound of the error between a desired closed-loop performance and that achieved by the designed controller. Based on this error analysis, an iterative algorithm, where data collection and the error bound estimation are iteratively performed, is proposed for designing a feedforward controller with a desired error precision. The effectiveness of the proposed algorithm is shown through a numerical simulation.