ABSTRACT. In this talk, we deal with a preventive replacement problem for multi-unit systems. First, we assume that the units in systems are identical, and the failures of units are independent and follow an exponential distribution. Secondly, it is assumed that the replacement time is negligible, the fixed and variable costs to replace failed units are included, and the expected cost rate is used as an optimization criterion.

We propose a preventive replacement policy based on the mean residual life of the system. If the system is failed before the mean residual life of the system is greater than the specified threshold (the decision variable), then the system is replaced correctively. Otherwise, the system is replaced preventively when the mean residual life of the system is less than the threshold. A procedure to obtain the expected cost rate is developed and an algorithm to find the optimal threshold of the mean residual life is also proposed under some well-known system structures in reliability.

Numerical examples are studied to know the effects of model parameters on the optimal solutions in various system structures. The optimality of the policy proposed is investigated by some machine learning techniques approximately in cases of discrete failure distributions. For further studies, we analyze preventive replacement policies for extended failure and cost models and promising optimization problems.

ABSTRACT. In this paper, we study an optimal post-warranty age replacement policy for second-hand products. The second-hand product, with an age x (>0), is purchased with a fixed length of non-renewing warranty period. During this period, the second-hand product undergoes periodic preventive maintenance at a prorated cost, while only minimal repairs are performed by the dealer at each failure. After the warranty expires, the second-hand product is minimally repaired at failure and is replaced either at the N-th failure or at the pre-scheduled replacement time, whichever comes first. The costs for both minimal repairs and replacement are borne solely by the user. The aim of this paper is to determine not only the optimal number of post-warranty product failures for a given length of self-maintenance period but also the optimal length of the self-maintenance period for a given number of post-warranty product failures. To this end, we derive a formula to evaluate the expected cost rate during the second-hand product’s life cycle, assuming a certain cost structure for maintaining the product during its life cycle, and discuss the optimality of the post-warranty age replacement policy.

ABSTRACT. Holding spare parts on hand is crucial for addressing maintenance demands, especially when the maintenance needs replacement. Managing inventory parameters efficiently is essential for balancing inventory costs and part availability. This study proposes an optimization model for managing spare parts considering deterioration and lead times. This study utilizes the periodic inventory policy (R, S). The maintenance cost, ordering cost, holding cost, and shortage cost equations are modified by incorporating the lead time function and spare parts deterioration function. Delay lead time directly increases total costs, while the speed of spare parts deterio-ration reduces them. The total cost can be minimized by balancing maintenance intervals and maximum inventory. The algorithm was developed to solve the model; moreover, numerical examples verify the deduction of the proposed model

ABSTRACT. This study discusses a repairable system consisting of two types of primary components, two types of spare components, and one unreliable repairman. The repairman is responsible for repairing failed components, but during the repair process, the repairman can also experience unexpected breakdowns. Additionally, the repair rate varies depending on different types of failed components being serviced. When the repairman experiences a breakdown, he is not entirely unable to repair the failed components; instead, he continues the repairs at a reduced service rate. When there are no failed components in the system, the repairman may perform the preventive maintenance. For this repairable system, we apply Cramer's rule to solve the steady-state equations and then numerically obtain the joint steady-state probabilities of the two types of failed components. Utilizing the steady-state probability distribution, we develop the steady-state availability and the steady-state failure frequency of the system. Finally, numerical examples are given to study the effects of various system parameters on the system’s steady-state availability and failure frequency.

ABSTRACT. The development of advanced sensor technologies has led to an increasing interest in component predictive maintenance. However, designing a predictive maintenance approach for multiple components is still a challenging problem as it requires effective integration of prediction and scheduling under resource constraints. This paper presents a dynamic predictive maintenance framework for multiple components that integrates remaining useful life (RUL) prediction into maintenance plan. First, a hybrid deep learning model is designed for predicting the RULs of multiple components, which combines a one-dimensional convolutional neural network and a bidirectional long short-term memory network with temporal self-attention mechanism. Meanwhile, a Bayesian optimization method is proposed to further improve the RUL prediction performance of hybrid deep learning model. Then, a novel mixed integer linear programming (MILP) models is developed to handle multiple components predictive maintenance under constrained maintenance resources, which targets to minimize the maximum maintenance time for all components. The experimental results show that the proposed dynamic predictive maintenance framework can flexibly provide accurate RUL predictions and maintenance planning to reduce maintenance time.

ABSTRACT. Group recommendation is currently a research area with increasing importance because of the diversity of scenarios in which it is useful. However, existing recommender systems have difficulty in aggregating individual users’ tastes into a group’s preference properly. Furthermore, it is unable to say that one of approaches is better and always applicable. Online product reviews have become an important source of information for consumers purchase decisions. Therefore, a Hybrid Preferences-Recommendations Aggregation algorithm (HPRAA) to group recommender systems will be developed which features: 1. Aggregate all group members’ preferences into common aggregated group model. 2. It is unlikely to find unexpectedly recommended items and it is easier to optimize the group recommendation process. Dining is something that everyone does every day. Dining with colleagues, family, and friends is a social activity that promotes emotional communication. However, it also comes with the risk of disagreements. Therefore, this study applies proposed algorithm to group dinner recommender system. The proposed method was compared with Average, Least misery method and Borda count. Experimental results show that the proposed method is superior to the state-of-the-art methods in NDCG. These achievements prove that our method, taking advantage of preference aggregation and recommendation aggregation, can supplement accurate user preference for group dinner.

ABSTRACT. E. Z. Demirci, and N. K. Erkip and Tanaka et al. constructed an optimal resource allocation problem to explore the budget allocation problem of a firm executing a new product development project to develop and manufacture a new generation of an existing product. The firm is to bring a product to market through a distributor while aiming to maximize its market potential. Our goal can be achieved by investing in R&D and using incentives for distributors to sale more products. On the other hand, given a set budget and high distributor’s incentives, we need to consider what decisions firms must make. In other words, in this study, we consider the optimal budget allocation problem, which considers product development and market penetration within a set budget or high incentives from distributor. As a result, we can also provide valuable managerial insights on how and when to determine the budget and the incentive values for distributors.

ABSTRACT. The electricity company is currently having several challenges, one of which is the energy transition to achieve net zero emission in 2060. Therefore, the electricity company must demonstrate the ability to increase profit through the electricity company's financial indicator return on assets (RoA) at least 4%. In the previous research, several methods were developed to solve the problems, namely by optimizing asset interventions. However, these methods cannot be applied in the electricity company because optimization was only performed at the equipment level and/or at the power generation level. This is because assets in the power generation can also generate revenue, which can be one of the considerations in the optimization. Therefore, the electricity company seeks to increase asset productivity by prioritizing asset intervention at the corporate level and using the economic lifetime method. The prioritization is carried out based on key performance indicators (KPIs) of corporate, the power system priority index based on prediction of the power system risk cost, prediction of the highest equivalent annualized profit with combinations of all intervention scenarios (with the intervention scenario being carried out now or postponed in the following years), all compliance criteria (environmental, regulatory, and safety), and prediction of corporate capital cost limit per year.

ABSTRACT. This study considers the portfolio optimization problem of diversified investment in companies that treat real estate investment trusts (REIT) as financial instruments. Portfolio optimization involves finding an optimal portfolio for investors. By considering the trade-off be-tween the two indicators of risk and return, I find a way to minimize risk under a constant return. In this study which uses a genetic algorithm, the portfolio optimization problem is solved based on stock selection. The Markowitz model is used to determine the optimal investment ratio. In this study, I solved a portfolio based on stock selection used a genetic algorithm and it has challenged CVaR to determine the optimal investment ratio.

ABSTRACT. The paper deals with a dynamic maintenance optimization problem for multi-indenture systems with a modular structure. The components age over time and fail randomly, a discrete Weibull distribution is used to model the failure time with the aging process. The system's state is defined by the binary variable states set of individual components, and the failures of the redundant components can be identified by inspections. Two maintenance strategies for the maintenance object are considered: component-level maintenance and module-level maintenance, different maintenance objects can take different maintenance actions. Since the state transitions and effective ages of all components in the system are independent of their history. The stochastic decision process of the maintenance strategy to minimize the maintenance cost rate is formulated as a finite discrete Markov decision process with a discrete state space. A reinforcement learning (RL) algorithm is developed to determine the optimal maintenance objects and the corresponding maintenance actions based on the system state set over a finite time horizon. Numerical examples illustrate the applicability and effectiveness of the RL algorithm.

ABSTRACT. In this paper, we consider a condition-based preventive replacement policy for two-dimensional lamp systems in which identical lamps cover the system service area. Each lamp located in the service area covers a circular area with the same radius and if the working lamps cannot cover the whole service area, then the system is failed. The failure time of each lamp follows an exponential distribution and can be replaced by a new one. We propose a pre-ventive replacement policy based on the mean residual life and if the mean residual life of the system is less than the given threshold, then we replace all lamps failed. The expected cost rate is used as an optimization criterion and we want to find the optimal threshold to minimize the expected cost rate. The procedures to calculate the expected cost rate and to find the optimal solutions are explained and numerical examples are also studied to analyze the proposed pre-ventive replacement policy.

ABSTRACT. Production capacity and efficiency are critical performance measurements in the manufacturing industry. A capable maintenance center can improve the equipment’s uptime and enhance the machine's maintainability. This research investigates a production system with multiple unreliable machines, warm standbys, heterogeneous failure modes, repairman’s working breakdown, and delayed recovery policy. The state-transition-rate diagram is presented and the differential-difference equations are listed. The classical Laplace transformation method is employed to compute the transition probabilities as well as the system reliability. Moreover, the mean time to failure of system is evaluated. A sensitivity analysis is performed to demonstrate the influences of system parameters on critical reliability indicators. With numerical and graphical results, managerial insights are summarized for decision-makers.

ABSTRACT. Networks, which model various real-life systems, require effective performance measurement and reliability maximization. System administrators focus on configuring the optimal type and quantity of binary-state components in each network's arc or node to achieve this goal. While previous studies have employed metaheuristic algorithms and analytical methods based on minimal paths or cuts, these approaches may encounter memory limitations in large and complex network topologies. To address this issue, our study integrates the Genetic Algorithm (GA) with Monte Carlo simulation (MCS), namely GA-MCS, to tackle the component assignment problem for transmission reliability. GA, known for its multi-point search capability, is utilized to identify the optimal component assignment. The MCS is used to estimate the transmission reliability for each configuration, continuing until the confidence interval width meets a predefined threshold. A case study on a wind power supply network is adopted to demonstrate applicability of the GA-MCS.

ABSTRACT. This study addresses optimizing the arrangement of sliding window systems for enhancing system reliability. Our focus lies in utilizing the properties of optimal arrangements to improve the performance of genetic algorithms and ant colony optimization. Our approach introduces a reordering step derived from previous work. Through numerical experiments, we demonstrate how integrating this reordering step into metaheuristic algorithms enhances computational efficiency and solution quality. These findings contribute to a deeper understanding of the optimization of sliding window systems.

ABSTRACT. Although many network systems have been designed in the real world, it is known that evaluating network systems is complex and takes a lot of computing time. The purpose of this study is to improve the method of obtaining quasi-Pareto solutions for a Bi-Objective Network. As in the previous studies, a genetic algorithm is used to improve both the discovery rate and error rate of the quasi-Pareto solution by improving the method of reducing the search space and the method of intersection. In order to improve the method of obtaining the quasi-Pareto solution, I conducted an additional investigation in addition to the properties of the quasi-Pareto solution shown in previous studies. Based on the properties obtained, the proposed algorithm uses a genetic algorithm to efficiently search for networks, and it generated initial population that predisposition toward to be quasi-Pareto solutions.

ABSTRACT. Major Depression Disorder, as the second most common disease in the world, has brought serious depression to society and families. With the rapid development of electroencephalogram(EEG) technology, the characteristics and potential biological mechanisms of brain networks in major depression based on EEG data have become one of the current research hotspots. In this paper, based on the open source depression database MODMA dataset, a brain network model was constructed using coherence coefficient to analyze and compare the topological properties and community division of the brain network between major depression patients and healthy people. On this basis, a seepage evolution based analysis method was proposed for the study of brain network resilience in major depression patients. One set of network resilience simulation experiment and two sets of network recovery ability simulation experiments were designed to simulate the process of brain network damage and its state transition, exploring the changes in brain network resilience in major depression patients.

ABSTRACT. The redundancy allocation problem (RAP) aims to determine the allocation of redundant components for each subsystem under resource constraints to maximize system reliability. The general RAP with k-out-of-n subsystems and heterogeneous components can satisfy the various requirements of reliability optimization engineering. However, the existing reliability formulation and model are specific for RAP with k-out-of-n subsystems and homogeneous components and RAP with 1-out-of-n subsystems and heterogeneous components. This paper considers a novel general RAP with k-out-of-n subsystems and heterogeneous components under mixed redundancy strategy. Firstly, a continuous time Markov chain theory is used to exactly calculate the reliability of subsystem. Secondly, the considered model for RAP is developed, which allows heterogeneous components and the minimum number of operating components can take any value. Finally, Numerical experiments verify that the model for RAP demonstrates high reliability.

ABSTRACT. Importance measure (IM) has made significant progress, but there are still omissions in addressing optimization problems with multiple resource constraints. Therefore, this paper focuses on a multi-constrained selective maintenance problem (SMP) under imperfect maintenance. System reliability and multiple constraints for cost and time are both noticed in IM construction to match the SMP better. Two novel IMs are proposed based on a Lagrangian function: SMP-based IM from a cost perspective (SIMc) and SMP-based IM from a time perspective (SIMt). For a component, SIMc (SIMt) reveals the comprehensive effect of cost (time) resource consumption on system reliability and other resources. Then two hybrid genetic algorithms guided by SIMc and SIMt respectively are designed, where the proposed IM is used to adjust the solutions locally. The advantages of the hybrid algorithms in quality of solutions, convergence, and robustness are demonstrated via numerical experiments.

ABSTRACT. The use of deep learning for Remaining Useful Life (RUL) prediction has grown, bringing significant progress. However, a common obstacle in real-world applications is the challenge of dealing with missing data in condition monitoring data within test datasets. The absence of complete sensor data can severely impact the accuracy of model predictions. To address this issue, this study introduces a novel data-filling method named DiffFMD, which employs diffusion models. This method first utilizes a self-supervised learning strategy to train the diffusion models, effectively learning the distribution characteristics of the missing data in the test datasets. Subsequently, a denoising network is designed to integrate generative conditions and precisely capture the spatiotemporal correlations in monitoring data, thereby estimating noise levels. Experimental validation demonstrates that the proposed method effectively fills in the missing data and significantly enhances the accuracy of the RUL prediction model in test scenarios where data is incomplete.

ABSTRACT. Due to the trade system is closed related to the national economy, interruption of trade would cause economy stagnated and throw the economy into a tailspin for the development of the trade globalization. Hence how to evaluate the resilience of trade system is critical. The topology property of the trade system has drawn great attention, but the flow of the product between countries is a dynamical process, the interaction between countries should be considered. In this paper, we propose a trade network model to establish the trade network based on the trade data. Then a resilience evaluation model which considers the interaction between countries is introduced. The average activity is calculated to evaluate the resilience of the trade network as the time changes. The results show the trade network becomes more and more homogeneous as the year increases. For the resilience of the trade network, the topology property is positive to the average activity of the trade network. More generally, the proposed model demonstrates how trade system resilience is and opens an innovative resilience evaluation model which considers the dynamics between nodes in the trade network.

ABSTRACT. This paper develops a performance evaluation procedure for a production system comprising both automatic and manual lines by applying stochastic-flow production network (SPN) model. Considering rework operations and the fact that rework from the automatic line must be transferred to the manual line due to closure issues on the automatic line, a graphic-based method is proposed to construct the production system as a network SPN, which is a novel network model for practical needs. A simple algorithm is developed to generate the minimal capacity vectors that each station should provide to meet the given demand. Finally, the probability that the production network can meet the demand in terms of the minimal capacity vectors is evaluated. A simplified production system is used to demonstrate the procedure, and further decision-making issues are discussed on the basis of the derived demand satisfaction probability.

ABSTRACT. This paper deals with network resilience for power grid networks by using graph theory. The demand of each substation is modeled as the weight of a node,and the transmission capacity of a line is incorporated as the capacity of an edge in the network. The total power supplied to all node is adopted as a performance measure of the network. Resilience of the network upon the simultaneous failures of edges is analyzed. Important measure is applied to obtain the repair sequence with high resilience in the analysis. Examples of resilience analysis for an actual power grid are shown.

ABSTRACT. Network Interdiction Problems are related to identifying components that are interdicted. The objective is to minimize the network flows of an evader, by interdiction activities. An evader aims to maximize the expected flows which are transmitted from a source to a sink without being detected. In this study, flows are detected stochastically at each arc, and the penalty cost is imposed on an evader when flows are detected. In this situation, we need to obtain the flows that effectively maximize the expected profits considering the detection probability, and determine which arc is the best to interdict.

ABSTRACT. Many real-world systems are modeled in the form of stochastic flow networks (SFN). This paper assesses the likelihood, called system reliability, that an SFN can send a given demand through a quickest path (QP) linking source and terminal in a fixed time. This specific computational issue, referred to as the QP flow system reliability problem, is classified as NP-hard. As an alternative, the Monte Carlo simulation is developed to randomly generate capacity vectors representing the current states of the network for estimating the QP system reliability problem. The proposed algorithm is tested thoroughly on a large-scale realistic network to show its high performance and speed.