Machine learning for optimization and decision support in complex system-of-systems: applications in microelectronics, healthcare, smart cities, and the circular economy

Abstract

Machine Learning (ML) offers powerful tools for optimizing decision-making across complex System of Systems (SoS) environments characterized by decentralization, dynamic interactions, and resource constraints. This work develops AI-driven frameworks tailored to five critical domains: semiconductor manufacturing, electronics quality control, healthcare coordination, smart city infrastructure, and circular economy marketplaces. Each framework integrates domain-specific modeling with ML techniques to enable scalable, adaptive, and cooperative decision-making. In Three-Dimensional Integrated Circuit (3D-IC) manufacturing, a multi-stage ML system combines Random Forests, Convolutional Neural Networks (CNNs), and Long Short- Term Memory (LSTM) models to align defect detection and process optimization with global yield objectives. Dual annealing and adaptive weighting enhance Through-Silicon Via (TSV) formation reliability, supporting system-wide performance gains. For electronics quality control, Natural Language Processing (NLP) and Naïve Bayes classification are applied to technician reports from Printed Circuit Board (PCB) testing, transforming unstructured narratives into structured insights. This approach improves component-level fault detection and demonstrates the role of human-in-the-loop analytics in manufacturing SoS. In healthcare, a Multi-Agent Reinforcement Learning (MARL)framework models coordination among hospitals, telemedicine platforms, and wearable devices. Simulated SoS configurations— Directed, Acknowledged, Collaborative, and Virtual—reveal that decentralized architectures offer superior adaptability, cooperation, and efficiency in Artificial Intelligence of Medical Things (AIoMT)-enabled environments. Smart city infrastructure is addressed through a MARL-based SoS framework that optimizes transportation, energy, and public safety systems via policy learning and augmented reward design. Simulations demonstrate improvements in coordination, efficiency, and system responsiveness. In circular economy digital marketplaces, time-series forecasting (e.g., AutoRegressive Integrated Moving Average (ARIMA), LSTM), Gradient Boosting Regressor (GBR), and reinforcement learning are used for resource prediction and price stabilization. The models enable dynamic supply-demand matching, supporting sustainable resource flows and demonstrating the potential of Artificial Intelligence (AI) for market-based Circular Economy (CE) implementation. Together, these case studies demonstrate how ML techniques can improve system-wide performance, autonomy, and cooperation across interdependent domains. The contributions span algorithm design, SoS coordination strategies, and domain integration, offering a foundation for deploying ML in critical infrastructure and complex engineered systems.