This paper proposes a method to visualize worker behavior by utilizing machine learning from data obtained using wearable devices and videos. In conventional IE, worker behavior is observed using a stopwatch or video camera, and the amount of time required for each motion is measured. On the other hand, it has been reported that human motion estimation using machine learning is highly accurate when skeletal data is extracted from videos, and motion estimation is performed using a learning model of a graph neural network series. However, when applied to manufacturing sites, various problems have hindered remove widespread use, such as the loss of skeletal data obtained at the manufacturing site due to the positional relationship with equipment and other workers.

In order to solve the problem mentioned above, this paper proposes a method to collect position and motion data from smart tags and skeletal data from video analysis, and to estimate workers behavior using machine learning. The smart tag was jointly developed by the research group and a company as a simple wearable device for workers. One of the features of this paper providing proof that the combination of video analysis and wearable devices can solve problems that are difficult to solve otherwise.

This paper clarifies the effectiveness of the proposed motion estimation method in the case where posture estimation accuracy is reduced by obstacles such as machinery and equipment, work-in-progress, and other workers, and where skeletal data is often missing. For this purpose, an experiment is conducted assuming cell production and remove the accuracies of several motion estimation models are compared.

Fuel-cell vehicles that use hydrogen as the fuel are one of the next-generation vehicles in the world. However, an inefficient hydrogen supply chain network (HSCN) hinders the spread of fuel-cell vehicles, and the main cause of inefficiency is the danger of transporting super-cold liquefied hydrogen or highly compressed hydrogen gas. In order to solve this problem, organic hydride is currently being developed. This technology transports hydrogen as methylcyclohexane, and this enables the use of the same infrastructure as petrol. Due to the novelty of organic hydride, there's no research regarding a HSCN involving organic hydride. In this research, a HSCN with three ways of obtaining hydrogen and three ways of transporting is modeled as Mixed Integer Linear Programming. Hydrogen is obtained through domestic production, importation, or production at onsite hydrogen fueling stations (HFSs), and transported as liquid, gas, or organic hydride. In order to test the effectiveness of organic hydride, three scenarios are compared for 25 different sized instances, and a sensitivity analysis is conducted to determine the changes in operational cost affected by a budget crunch. From the results, it is possible to say that organic hydride makes a HSCN more effective when ports, dehydrogenation plants, and HFSs are sufficiently close. Additionally, the more choices involved in the HSCN, the more effective the HSCN can be designed, regardless of the size of the problem. Furthermore, the actual decisions of what to use in a HSCN are decided based on the relative positions between each infrastructure, capacity, level of demand, and budget. In addition, when the amount of budget decreases, operational cost contrarily increases due to the impossibility of constructing an onsite HFS, and the longer traveling distance required due to the reduction of the number of possible construction locations.

The household electricity consumption data contains information on the lifestyles of each household unit, which is useful for marketing of consumer durables and services for households. In general, however, only the main power consumption can be metered using a single smart meter for each residence, and the operating status of home appliances is unknown from the companies' viewpoint. Therefore, there have been several attempts to estimate the electricity consumption of each home appliance using dis-aggregation technology. However, there have been cost problems in practical use for exact estimation, such as the need for additional sensors. In this study, the authors formulate a problem specialized for estimating the operation and non-operation of home appliances using observed main energy data as household attribute information for marketing purposes. They then propose a state estimation model in a snapshot using the assumption of normal distributions for the electricity consumption of home appliances. Simulation experiments in a virtual housing environment show that the proposed model has an effective rate of correct answers in practical use. Additionally, a relationship between the proposed method and the prior probability distribution of device states that serve as input for the estimation is shown. The proposed model is a low-cost method for estimating household attributes that do not require additional sensors, and is expected to be used as basic information for marketing strategies.

Since September 2021, because of the COVID-19 pandemic, the Japanese government has heavily limited business practices in the restaurant industry. While many restaurants operate in accordance with the guidelines for preventing the spread of COVID-19, the details of operation during business hours are left to each restaurant. In particular, as social distancing significantly contributes to reducing the infection risk, a common strategy is that an operator restricts the available seats in advance and then allocates customers. However, the effectiveness of seat allocation for reducing the infection risk is not always the same due to the situation in each restaurant; for example, the number of customers and their relative positions are always changing. Hence, an operator can effectively reduce the infection risk by determining a seat layout dynamically, as opposed to traditional methods. In addition, the magnitude of risk intended by an operator may change according to the situation in the restaurant, social conditions, and so on. Therefore, this study proposes an operational model for restaurants to reduce the magnitude of infection risk using a simplified parameter θ. The parameter θ is the threshold of infection risk for the entire restaurant space for an arbitrary time. By simulating the proposed model in a virtual restaurant, it is confirmed that the model can easily control the infection risk using a single parameter and contribute remarkably to reducing the infection risk with a slight loss of proceeds.

This paper focuses on the automation of work measurement using a camera to capture workers' hands in an assembly factory and analyzing the image to estimate the work content. In recent years, the technology to collect, analyze, and visualize manufacturing data from various devices installed in processing machines has been improving. However, it has not yet reached the point where workers' manual operations can be analyzed and visualized in real time. In this study, it is proposed that high estimate accuracy can be achieved using two analysis procedures to analyze a worker's image with deep learning: setting the region of interest and estimating the work content from the features in the region of interest. As a result of an evaluation experiment conducted at an actual factory using the proposed method, 99.5% of the work was correctly estimated. Based on these results, the method proposed in this study is now being examined in anticipation of introducing it for practical application.

In general, assembly lines are connected by conveyors. Therefore, the movement of products is synchronized and they all move at the same speed. In recent years, factories in which products are moved individually without synchronization have been introduced. In this study, we examine the difference in productivity between the synchronized line and the individual transport line.