Recently, there has been a surge of interest and research on a phenomenon popularly called "the bullwhip effect". This phenomenon states that a variance in customer demand can increase as we proceed upstream in the supply chain. The bullwhip effect has been recognized in many diverse markets. The phenomenon itself is not new and several studies have speculated on it; however, most of them have focused on demonstrating its existence, identifying its possible causes, and providing methods for reducing its impact. In particular, Lee et al. identified the main cause of the bullwhip effect: demand signal processing based on local optimization using forecasting methodology, non-zero replenishment lead-time, and batch ordering. On the other hand, Chen et al. and Levi et al. have succeeded in quantifying the bullwhip effect. However, these studies mainly focus on determining the impact of demand forecasting and lead-time, and the impact of batch ordering remains unanswered. Since it can be considered that batch ordering may have a large impact on the bullwhip effect, we first extend their supply chain model by introducing a periodic review system with a fixed review cycle period. Adopting this system, we build a batch ordering factor into our model and derive new quantifications for the bullwhip effect. Our quantifications include the previous one. To evaluate these quantifications, we also examined the accuracy and adequacy of our quantification by simulations. The results show the effect of a larger impact of the review cycle period.

View full abstract

In the traditional assembly line, usually each worker is assigned to a particular fixed task and the greatest work content determines the production rate. For this line, assigning a worker to balance the line is studied in previous research. However, it is inflexible. In this kind of line, when an unbalance between workers' speeds exists, the slowest worker will delay overall production. As a result, the production rate of the particular line will decrease. For solving this problem, a "Self-Balancing Production Line" was introduced. The utilization of the mentioned method is reported in at least two commercial environments: apparel manufacturing and distribution warehousing. In this type of production line, each worker is assigned task dynamically, and when the last worker completes an item, he/she walks back and takes over the next item of his/her predecessor, who relinquishes it, walks back, takes over the next item of his/her predecessor, and so on until the first worker walks back and starts a new item. Since faster workers are assigned more time in processing an item, and slower workers spend less time, a balance is created. For this line, it is found that production achieves the maximum production rate if the workers are sequenced from slowest to fastest. Additionally, other conditions for three workers were found numerically by simulation. However, this condition is extremely limited, and it is necessary to conduct further research for three or more workers to find the maximum production rate. In this paper, in spite of the traditional sequence of assigning workers from slowest to fastest in the production line, other conditions that can reach the same self-balancing effect are found and analyzed.

View full abstract

The total loss of fresh agricultural products amounts to 20-60% of all the products harvested in any country, requiring an effective harvesting-delivering plan in the agro-industry. This paper constructs a basic supply chain model for harvesting and delivering fresh agricultural products to multiple markets. This is accomplished by formulating the plant growing process and the loss process of fresh products in mathematical forms. The model deals with periodical harvesting under periodical flowering to maximize a level of demand satisfied constantly in each market every period. The demand level satisfied in each market is assumed to be proportional to the market size. The periodical flowering and growing process is formulated in a quasi-concave function to express the maximum amount of harvestable fresh products, provided that any requirements for harvesting fresh products should be satisfied through the earliest possible flowering. The loss process is identified in a monotone non-increasing loss function expressing the amount of available fresh products whose quality is good enough to consume. After any amount of fresh products is consumed in any market, the remaining fresh products available in the next period is calculated by considering the continual loss process. Periodical multiple deliveries to each market are also formulated to express the maximum amount of available fresh products, provided that any requirements for satisfying demand should be satisfied through the earliest possible delivery. An optimization algorithm is proposed for finding an optimal harvesting pattern to maximize the demand level satisfied constantly every period. The derivation process of the proposed algorithm is described in detail, and the proof is given to show that the proposed algorithm is optimal. Numerical examples are demonstrated to show how the optimal harvesting pattern varies with changes in the interval of periodical flowering and delivery lead-time.

View full abstract

The effectiveness of feed-forward neural networks has been recognized widely. The learning iteration in neural networks is largely influenced by parameter values such as momentum terms (α), learning rates (η), adaptive steepness factors (ε), and width of distribution of initial connection weight (wd). However, in most studies, parameter values are empirically determined. The Genetic Algorithm (GA) has been applied to obtain the approximate value for an optimization problem and has attracted attention as a practical search algorithm. The Parameter-free Genetic Algorithm (PfGA), which does not require the setting of parameter values, was recently suggested. The purpose of this study was to apply PfGA to a parameter decision problem in neural networks and to consider the effectiveness of PfGA by comparing it with existing methods, such as simple GA and random search. As the result of a computer simulation, the search time for the proposed method (PfGA) was shorter than that for the simple GA and for the random search. Each parameter (α, η, ε, and wd) was distributed uniformly in a random search. Distributions of each parameter had a peak in the PfGA and simple GA. The peak of the distribution was higher in the PfGA than in the simple GA. Moreover, the PfGA obtained better parameter values for the neural network than the other methods. These results indicate that the PfGA is better than the existing methods for parameter-related decision problems in neural networks.

View full abstract

There is a new advanced method, called Design Method-II, for flexible mixed-product lines, which follows the pioneering method, Design Method-I. The word "flexible" means product-mix flexibility. A flexible mixed-product line is capable of executing various production plans. Each production plan represents the number of products from each product type to be assembled within a time specific period. This design method consists of a new algorithm imitating the fission of living beings, created to solve design problems where search trees branch out widely. The algorithm called "fissiparous algorithm", could bean evolutionary algorithm. Conventional evolutionary algorithms change chromosomes randomly or according to probabilistic rules, but the fissiparous algorithm does not change chromosomes in this manner. This algorithm changes the chromosomes of each individual in the population by one power, accelerating the evolution. In Design Method-II a chromosome is a string of the order numbers discriminating tasks (elemental works). The design method looks for a solution to the design problem by assigning tasks in the chromosome to each workstation and constructing "station conditions." Design Method-II combines a method for constructing station conditions with the fissiparous algorithm while Design Method-I does not construct station conditions. Station condition is a set of production factors capable of performing the tasks assigned to each workstation. This design method requires the following data and information: (1) The product types to be assembled on the same line. (2) The various production plans to be executed. (3) The precedence diagram of each product type and the combined precedence diagram. The diagram contains the kinds of subjects (e.g. worker, robot, equipment or combination thereof), tools capable of performing each task and performance times. A station's ability to perform a task determines the task performance time. This design method aims to minimize the number of workstations to execute each production plan within a time specific period. We confirmed the validity of this method by solving problems.

View full abstract

QFD (Quality Function Deployment) is widely used to reflect the customers' needs in product design. Engineering design evaluation is characterized by imprecise importance and satisfaction levels of customer requirement, which are more fuzzy variables rather than they are crisp variables. Fung et al. proposed that the cost for making technical improvements was represented by fuzzy variables. Bovea et al. suggested fuzzy variables approach, based on the HoQ (House of Quality) matrix in the QFD methodology, provided a more quantitative method for evaluating the imprecision of the customer requirement. However, most authors have avoided the extended division for weighing the fuzzy sets using linguistic variables. Vanegas et al. proposed NFWA (New Fuzzy Weighted Average) as a new method of operating on fuzzy numbers to determine a fuzzy weighted average. The resulting fuzzy number is less imprecise and more realistic. In this paper, the weight of customer requirements and the relationship strength between customer requirements and quality characteristics are represented by linguistic variables, which are expressed by trapezoidal membership functions. We also extend the NFWA to treat not only triangular membership function but also trapezoidal membership function. In addition, we define unsolved technical problems by considering the relationship between the specified technical attribute and corresponding functions or quality characteristics, and we execute technical innovation by applying TRIZ (a Russian acronym for Theory of Inventive Problem Solving), which was developed to assist engineers in finding innovative solutions to technical problems in development processes.

View full abstract

Simultaneous approaches to fuzzy clustering and quantification of categorical data have been applied to knowledge discovery. In the approach, a data set is partitioned considering similarities of samples or variables and the feature values derived in each cluster effectively reveal the local characteristics of non-linearly distributed data. This paper proposes a simultaneous application of homogeneity analysis and fuzzy clustering that can deal with missing values in the iterative algorithm based on the "alternating least squares" technique. We discuss the differences between the proposed method and the other quantification methods through a numerical experiment, followed by an analysis of questionnaire data about information technology.

View full abstract

This paper aims to enhance the effectiveness of the process of improving manufacturing quality of a product in a multi-stage manufacturing system. The manufacturing quality to be improved is measured by the expected value of Taguchi's quadric quality loss function, and improving it is equivalent to reducing the variation of the final quality characteristic of the product, inspected at the end of the manufacturing system. The dispersion to be decreased is caused by many different noise sources throughout the multiple production stages, each of which carries out a certain manufacturing operation on the product and its result is assumed to be captured by an intermediate quality characteristic. Stabilizing the dispersion requires conducting robustness enhancement and/or noise source control at some of the production stages. This paper proposes a decision support approach to determine which means should be applied to which production stages. The proposed approach first estimates the cause and effect relationships among the multiple intermediate characteristics and the final one through engineering knowledge and observational data analysis. It then evaluates the expected variation reduction effect of each means at every production stage according to the causal relationships among the characteristics and their variance covariance matrix. Further, the resultant estimations are visualized in order to support the decision maker. An example semiconductor manufacturing system confirms the effectiveness of the proposed approach in directing quality improvement efforts.

View full abstract

In this study, methods to raise the efficiency of human movement via transportation in general are considered. First, we systematically express the transfer of humans and objects separately, and discuss more efficient methods of traveling by auxiliary methods, such as bicycles, taxis, and other vehicles. An accurate forecast method of market demand for transportation and also a method that provides an efficient supply to the market are necessary to increase the efficiency of transportation using auxiliary methods. In this paper, a multivariate linear analysis model used to demonstrate the forecasting of market demand for auxiliary methods is proposed, and a demonstration verifies the accuracy of this model. Moreover, this paper shows that the multivariate linear analysis model depends on time factors such as time of the day and day of the week, environmental factors such as weather, direction and speed of wind, temperature, and humidity, regional factors such as residential versus industrial or commercial locations, and cases of sudden or single occurrences (such as traffic accidents and traffic jams, and other factors which are festivals, opening and closing of shopping stores, entrance examinations, graduation ceremonies, school entrance ceremonies, academic meetings, etc.). In order to show the accuracy of the multivariate linear analysis model, these factors are considered using taxi transportation records of O city as examples to verify this model. The term "buturyuu" in Japanese means "materials flow", which is very popular. But, the term "jinryu" means people flow, and is a new word.

View full abstract