(147am) Simulation, Optimization, and Machine Learning Methods for Sustainable Process and Supply Chain Design

Research Interest: stochastic programming, data science, surrogate-embedded optimization, machine learning, supply chain optimization, plastic advanced recycling, sustainability

With tightening environmental regulation and fierce global competition, developing sustainable processes and making optimal decisions under uncertain market conditions are essential in the current chemical industry. To promote sustainability and digitalization, my Ph.D. thesis work has been focused on leveraging process design, modeling, data science, and mathematical optimization to evaluate and improve emerging renewable chemical manufacturing.

Process Design and Analysis:

Flowsheet design and simulation provide reliable process operation data for detailed economic and environmental analysis, especially for the early stage of technology development. Simulation models could be used to test new strategies and guide debottlenecking. Case studies include circular ethylene from plastic pyrolysis oil, modular microwave-assisted PET glycolysis, high-performance polyester precursor synthesis from furfural, and pressure-sensitive adhesives from lignin. Using simulation results, Bayesian optimization with cost and emission objectives helps to guide the search for optimal reaction conditions.

Process and Supply Chain Optimization:

Due to the complex nature of feedstock, biomass or plastic waste could be converted to various products by different combinations of technologies. To design a biorefinery that handles prices, supply, demand, and yield uncertainties, a superstructure model of stochastic programming with flexibility index constraints is formulated. ReLU neural network surrogate is trained and substitutes the multi-level flexibility constraint to reduce the solution time from hours to 6 seconds while maintaining good approximation (99.7% R²). Moreover, spatial and temporal variations of biomass feedstock affect the biorefinery performance. Thus, I built a two-stage stochastic programming model with rolling horizon planning to design and evaluate the modular biomass supply chain performance.

Global Solvent Pricing:

Chemical prices fluctuate significantly in different regions due to supply and demand imbalances. Using historical price data and exogenous information, I have built different models for price forecasting and market analysis. Next, a stochastic programming model is developed and incorporated into Python Flask web app for non-technical stakeholders to guide business decisions.

Keywords: Sustainable Supply Chains, Process Design & Development, Computing and Systems Engineering