Andrei Sleptchenko

Within the Industry 5.0 vision, designing resilient, sustainable, and human-centric maintenance facilities plays a critical role in industrial operations. This paper focuses on optimizing the design of a maintenance system for expensive equipment. The proposed maintenance system replaces faulty parts with ready-to-use spares from inventory and repairs the failed parts to be returned to the stock in as-good-as-new condition. The repair tasks are performed by multi-skilled servers (e.g., technicians) who can handle specific types of failed parts. Our objective is to enhance system performance by minimizing the overall cost, including factors such as inventory holding costs for spare parts, backorder costs, server costs, and training costs for multi-skilled servers. We investigate the impact of incorporating preemptive priorities at the repair facility, both under baseline demand and under demand surges. A simulation-based Binary Particle Swarm Optimization (BPSO) is utilized to optimize skill-server assignments and base-stock levels. Multiple priority rules were evaluated to identify the optimal policy, based on their respective impacts on system performance and resilience. The results show that prioritizing repair tasks based on service times reduces total system costs by approximately 4.8% and yields higher resilient performance under demand surges. The paper provides valuable insights into designing efficient, human-centric, and resilient maintenance facilities for expensive equipment used in industrial applications.

Missed appointments in outpatient healthcare settings continue to be a widespread operational and clinical challenge, resulting in inefficiencies, disrupting care continuity, and contributing to financial losses. Addressing this problem requires analytical frameworks that extend beyond descriptive reporting toward interpretable, decision-oriented models. This study develops a knowledge-based probabilistic framework that integrates advanced machine learning (ML), probabilistic graphical models, and causal inference to predict and mitigate nonattendance. Leveraging 110,526 outpatient records from Brazilian healthcare facilities, the framework applies a structured pipeline that combines systematic feature selection, hybrid class balancing, and model training within a stratified nested cross-validation design to ensure unbiased performance estimation and reduce overfitting. The best-performing model achieved an accuracy of 0.93, an AUC of 0.83, and a macro-F1 of 0.83, surpassing benchmark approaches and providing a foundation for subsequent analyses. Building on these predictive results, Bayesian belief networks (BBNs) were employed to capture probabilistic dependencies among patient, temporal, and appointment-related variables, thereby enhancing interpretability, supporting scenario-based reasoning, and providing the foundation for causal analyses. In the final stage, average treatment effect (ATE) estimation and uplift modeling were applied to evaluate the impact of SMS reminders. Results demonstrate an average 15% reduction in nonattendance probability, with the strongest uplift observed among patients with repeated missed visits. This reduction corresponds to an estimated annual saving of approximately $670,000 in unused clinical capacity. Collectively, the integration of predictive modeling, probabilistic knowledge representation, and causal inference provides a coherent framework that is both analytically sound and operationally relevant, offering actionable tools to optimize scheduling, allocate resources efficiently, and enhance patient access.

Emergency departments (EDs) are frequently challenged by fluctuating patient volumes and resource constraints, leading to overcrowding and operational inefficiencies. This study presents a framework that combines design thinking and discrete event simulation (DES) to optimize ED processes and improve patient outcomes. By using both methods in conjunction with the design of experiments, the digital twin evaluates the effects of various resource configurations on key performance indicators, such as length of stay (LOS). The model highlights critical resource bottlenecks and explores alternative scenarios to improve ED performance. A case study demonstrates that targeted adjustments in nurse staffing and bed management significantly improve operational productivity and reduce patient LOS. The proposed approach provides hospital administrators with practical information to optimize resource allocation and streamline ED operations, contributing to better patient experience. This methodology is particularly suited for healthcare systems that face different patient demands and limited resources.