MULTISCALE MODELING APPROACH FOR DESIGN OF INDUSTRIAL NETWORKS TOWARDS ZERO WASTE AND EMISSIONS: FROM PROCESS TO NETWORKS

This dissertation advances a methodological framework for designing, deriving, and optimizing circular manufacturing networks through the integration of bottom-up physical input-output analysis, data-driven modeling, and dynamic optimization. The research addresses critical limitations in conventional sustainability assessment tools by developing transparent and technology-specific approaches that bridge process-scale engineering detail with economy-wide material flow accounting to enable genuinely circular industrial systems. The work comprises five interconnected studies that progressively construct a comprehensive design framework. First, the By-product Aware Sales Structure (ByPASS) methodology is developed to construct Physical Input-Output Tables (PIOTs) directly from mechanistic process models, overcoming the fixed product structure assumption inherent in conventional supply-and-use table transformations. This bottom-up approach enables the transparent traceability of commodity flows that occur as inter-industry material exchanges, providing the analytical foundation for circular economy design. Second, the ByPASS methodology is applied to systematic circular economy design through a comprehensive case study of the U.S. acetaminophen manufacturing network. The analysis demonstrates substantial waste circularization potential through strategic industrial symbiosis, achieving 78% ammonia waste valorization via urea manufacturing integration, complete CO2 valorization through carbon capture & utilization (CCU) coupling, 70% organic wastewater recovery via synergizing the acetaminophen industry with the green hydrogen production, 97% circularization via integrating linear alkyl benzene production, and near 100% reduction of waste by adding methyl methacrylate (MMA) manufacturing industry. These interventions exhibit sector-level and economy-wide ripple effects that extend far beyond the immediate pharmaceutical network, demonstrating that localized circularity interventions can trigger system-wide sustainability improvements. Third, a multi-tiered reinforcement learning framework is integrated with PIOT-based network representations to autonomously discover economically viable circular economy strategies. The framework operates through hierarchical decision stages encompassing exchange price optimization, buy/sell quantity decisions under market volatility, and process-level material inflow optimization. Simulations demonstrate 65% reduction in virgin material consumption, diversion of over 70% of waste streams to productive reuse, and 12-18% improvement in net revenue relative to linear baseline operations. Fourth, hybrid mechanistic-machine learning models based on Sparse Identification of Nonlinear Dynamics (SINDy) are developed to capture coupled industrial network dynamics while achieving computational speedups of 2-3 orders of magnitude. Applied to an algal biodiesel production network coupled with carbon capture systems, the SINDy framework successfully recovers network-level dynamics from facility-level simulations while preserving physical interpretability and modularity, enabling rapid scenario analysis and robust design optimization under uncertainty. Fifth, these hybrid surrogate models are synthesized through dynamic optimization frameworks that coordinate buffer inventories, shipping schedules, and operational decisions across multi-plant networks. Applied to a four-plant acetaminophen manufacturing network under realistic operational constraints, the framework achieves 70% waste circularization, demonstrating that explicit accounting for process dynamics and inter-plant temporal coordination is essential for unlocking circularity potential. The research demonstrates that substantial progress toward industrial sustainability is achievable through intelligent coordination and data-driven optimization of existing infrastructure. By integrating mechanistic process understanding with economy-wide systems analysis and advanced computational optimization, the developed methodologies provide a scientifically rigorous pathway for transitioning manufacturing networks from linear to circular operation modes. This work contributes to the theoretical foundations of sustainable process systems engineering while providing practical design tools for industrial practitioners and policymakers seeking to accelerate circularity transitions under economic and operational constraints.