ABSTRACT. The reverse supply chain network (RSCN) involves collecting the End-of-Life (EOL) products from users and transporting them to recovery facilities for recovery or disposal facilities. RSCN contributes to saving material resources. However, RSCN has a trade-off relationship between environmental and economic factors. Therefore, it is desirable to recover the EOL products that as valuable as possible. To identify the valuable EOL products, this study applies the Random Forest (RF). RF realizes a mechanism equivalent to human learning using com-puters based on algorithms. This study designs the RSCN model using integer programming and RF.

ABSTRACT. This paper introduces a robust optimization model for designing a multi-objective, multi-plant supply chain network under demand uncertainty. The model employs Mixed-Integer Programming (MIP) to optimize decisions across multiple components, including manufacturing plants, suppliers, customer zones, and time periods. It aims to minimize total costs encompassing production, transportation, energy, technology, inventory holding, and sourcing costs. Additionally, the model incorporates critical factors such as safety stock, service level tar-gets, and reliability metrics to ensure robust performance under fluctuating demand. The optimization is implemented in GAMS (General Algebraic Modeling System) using the CPLEX solver to derive optimal solutions. The optimization result demonstrates that the model is robust for different scenarios of uncertain demand.

ABSTRACT. Sustainable production and consumption have been sought in the manufacturing industry in recent years, and the exploration of circular systems for collecting and reusing components supplied to the market has advanced. Traditional circular models assume the reuse of parts within the same product category, often leading to the disposal of insufficiently utilized components. This study proposes a collection and reuse plan for lithium-ion batteries, with the aim of maximizing the lifetime power supply from market introduction to disposal through reuse across multiple product categories. In addition, the effectiveness of the plan in small-scale problems is verified.

ABSTRACT. In-home rehabilitation service is provided to elderly and patients who are recognized as the care required. This service is provided to them who have difficulty in self-care or going to the hospital. The scheduling for in-home rehabilitation staff requires consideration of various constraints such as the national qualifications of staff and the time required for visiting patients' homes which places a burden on the staff making the schedules. For ex-ample, the national qualifications of staff contain physical, occupational and speech therapists. Additionally, elderly have time window for in-home rehabilitation service because they receive other health care services or their family have to attend the service. This study developed the route selection model based on work tasks identified through interviews with target hospitals. Additionally, in order to gain insight for route planning by comparing the improved routes with current routes. Furthermore, this study obtains feedback from in-home re-habilitation staff and manager to discuss the proposal model.

ABSTRACT. For the nurse scheduling problem, we proposed a hybrid method using both mathematical programming and machine learning. In the method, by solving a mixed integer programming problem, we obtain a shift schedule of nurses first, and however, the shift schedule is not necessarily acceptable by a chief nurse because there is a hidden condition, which is one of the practical difficulties of the nurse scheduling. Then a deep neural network (DNN) model trained using practically acceptable shift schedules fixes the shift schedule automatically to make it acceptable by a chief nurse. In this paper, we propose a framework to investigate the capability of the method, which is not a simple task because a hidden condition is hidden even if we know acceptable shift schedules. The framework needs another mixed integer programming problem, called the virtual chief nurse problem. The virtual chief nurse problem implements a (virtually) hidden condition, such as the incompatibility of nurses, and its solutions are used as training data of the DNN model. As a result, the DNN model acquires the capability of learning the hidden condition, i.e., modifying the shift schedule appropriately, which implies that our method is promising.

ABSTRACT. In the quality control departments of manufacturing industries, many inspection tasks have been mechanized and automated. However, for certain products that cannot be easily automated, manual inspection remains indispensable. For example, rotary tools (diamond burs) used in dental treatment have microscopic diamond particles electroplated onto their tips, resulting in unique shapes for each tool. Consequently, visual inspection under a microscope is still performed, and inspection accuracy tends to be unstable. To address this issue, we have previously proposed a comprehensive inspection system consisting of three tools: a labeling tool, a worker training tool, and an automatic inspection tool, aimed at stabilizing inspection quality. The labeling tool collects data that experienced workers use to distinguish between acceptable and defective products; however, this in-formation has not yet been incorporated into the discriminant models of the automatic inspection tool. In this study, we construct a multi-input CNN model that utilizes both data obtained from the labeling tool and image data, and we verify the accuracy of the machine learning model through experiments.