This paper formulates flowshop scheduling problems where the processing time of each job at each machine is represented by an interval. Because the processing time is given as an interval, various scheduling criteria such as the makespan and the total flow time are also calculated as intervals. In this paper, we first describe how the completion time of each job can be calculated as an interval in flowshop scheduling problems using interval processing time. Then we define the interval makespan, which is used as an interval objective function in flowshop scheduling problems. For minimizing the interval makespan, we use order relations between intervals, which have been utilized in mathematical programming problems with interval coefficients. Based on the order relations between intervals, the minimization problem of the interval makespan is reformulated as a multi-objective problem with two objectives : to minimize the upper limit and the center of the interval makespan. By computer simulations, we demonstrate that a multi-objective genetic algorithm can efficiently find nondominated solutions of this two-objective problem. Then other scheduling criteria such as the total flow time, the total tardiness and the maximum tardiness are extended to interval objective functions in the same manner as the interval makespan. In such an extension, we explain that the minimization problem of the number of late jobs should be handled in a different manner from the other scheduling criteria in flowshop scheduling problems with interval processing time. Finally, we introduce the concept of dynamic rescheduling. In dynamic rescheduling, the current schedule is always reconsidered for rescheduling whenever the processing of a job at a machine is finished (i.e., whenever the actual processing time is known).

View full abstract

In manufacturing processes, many robots are used to operate, assemble, and handle the parts and materials. However, the recognition ability of these robots is not highly flexible for different shapes of parts. Recently, as the life cycle of products becomes shorter, robots should be required to handle more different shapes of parts. To correspond to this requirement, recognition ability by image processing should be continuously increased. In this paper, we propose an assembly robot system with high recognition ability attained by combining three-dimensional computer-aided design information with image processing. This system is composed of two robots, six cameras, and four computers. One camera (upper camera) is set at the upper side, and the other three cameras (side cameras) at the contour of the recognition area. On the robot, one camera is set at the hand (hand camera). The recognition ability of this system can be explained by the following three functions. First, the type and direction of the parts are estimated by matching the height, which is estimated by the side cameras and images of parts to the computer-aided design information which is given a priori, since the parts are not only in singular state but also in contacting, leaning, and stacked states in the recognition area. Second, the direction of gripping is determined by the simulation of the interference between the parts and the hand of a robot using the computer-aided design information and images input from the hand camera. Finally, the position of a part being assembled by a robot and that of an assembled part are determined by the centers of the parts using two side cameras to assure the accuracy of assembly. In this system, a part can be handed from one robot to another, to change its direction and hold it.

View full abstract

Under the globalization process, products and components should correspond to the many standards for input and output sources of power, information and specifications. In this study, we propose a standardization method to correspond to many standards and specifications for the purpose of satisfying customers' various requests. Firstly, a corresponding index to the kind of standard for the input and output sources is defined for the product. Secondly, an influence index to humans or the environment is defined for the origin sources of the unit. Thirdly, the correlation index between the output sources of one unit and the input sources of an other unit is defined based on the entropy theory. Fourthly, an index for parts and subunits is defined by specifications and functions. The standardization of a product or component is determined using these indexes as follows. Firstly, increase the corresponding index of input and output of the product, because the product should correspond to the power and information sources of the many countries which cannot change their specifications. Secondly, as the influence to humans and the environment should be decreased on a common basis around the world, the influence index should be increased to the highest level. Thirdly, from the manufacturing point of view, the correlation index among components should be increased, because assembly work can be simplified and made more efficient.

View full abstract

The objective of this study is to select a limited number of securities and decide the portfolio of the selected securities which satisfies the decision makers. Mixed-integer portfolio selection is solved using a genetic algorithm. It is difficult to generate individuals who choose N securities out of n securities based on the 0-1 encoding of a genetic algorithm. Therefore, encoding is employed to set an allele to the number of 0-1 variables and the chromosome length to the number of choices. The portfolio of selected securities is decided based on a fuzzy-logic concept which realizes the quasi-satisfaction of a decision maker for the expected rate of return and risk ; in other words, the aspiration level of a decision maker. It is shown, using an example which chooses 4,5 or 6 securities out of 15,that the mixed-integer fuzzy portfolio selection using a genetic algorithm is an effective method.

View full abstract

In this paper, we propose a method for making route decisions in a dynamic traffic network. Many factors exist for route decisions, P. Mirchandani et al.(1985) and A. Wijeratne et al.(1993) studied two-objective and multi-objective problems. But A. Wijeratne's study shows sets of candidate routes ; therefore requiring drivers to decide the route on their own. Furthermore, they did not consider the dynamic variable conditions of traffic. These days, since the introduction of vehicle information control systems, drivers can receive information on the traffic conditions in a network. In dynamic traffic networks, route decisions should consider traffic congestion. We propose an algorithm which presents the path that avoids traffic congestion using dynamic traffic information from vehicle information controls systems, and examine its efficiency.

View full abstract

A reinforcement-learning-based approach is proposed for analyzing a market selection problem involving many agents who compete with each other in several markets. In our market selection problem, each agent is supposed to choose one market from several to sell his product. The market price of his product is determined by the supply and demand balance in the market. Thus, the profit the agent obtains from the market depends on the total amount of other agents' products that are sold in that market. In this paper, we first formulate our market selection problem as an iterated n-person non-cooperative game. Then we show how a Q-learning-based method can be applied to our market selection problem. The Q-learning-based method is applicable to our market selection problem even if each agent does not know the demand function of each market, which determines the market price. Finally, we demonstrate the effectiveness of the Q-learning-based method by computer simulations. In the computer simulations, we compare the Q-learning-based method with a tailored strategy where each agent chooses a market with the maximum expected profit when other agents choose the same markets as previous trial.

View full abstract

This paper describes a method to decide the inventory levels of semifinished and finished products so as to realize the target service rate by shipping products within a specified time, which we call "acceptable response time, " under the situation that demand follows normal distribution. We analyze the substitution relation between the acceptable response time and the inventory levels of semifinished and finished products. As a result, inventory levels required according to the length of the acceptable response time are obtained. The validity of our method is also demonstrated through simulation experiments with the conditions given by analysis.

View full abstract