Physics-informed learning-based synthesis of functional materials: This project transforms chemical synthesis and composite material discovery by moving beyond traditional one-variable-at-a-time experimentation. Complex chemical processes involving numerous interacting factors require more sophisticated approaches than conventional design of experiments. To address this, an adaptive design of experiment framework integrates computational multiscale modeling with experimental data to optimize dynamic decision variables. Combining first-principles modeling, machine learning, process optimization, and targeted experiments, this approach deepens understanding of material microstructure and composition. Applied to perovskite oxide synthesis, it optimizes parameters such as temperature and annealing time to improve product quality and efficiency. This approach automates exploration, reduces resource use, and accelerates the discovery of advanced materials, enabling more efficient manufacturing. 

​

Digital intelligence architectures for efficient and secure operation of integrated process systems: This research advances sustainability and resilience in integrated chemical manufacturing and energy by developing digital computational technologies for distributed, multiscale polygeneration networks. It addresses complexity and uncertainty through platform-independent modeling, physics-informed surrogates, process data integration, and decision frameworks. Optimal decomposition within a bi-level multi-agent architecture partitions large-scale problems into coordinated subproblems, enabling efficient automation and robust operation. The framework ensures precise product-quality control, real-time energy and material flow management, and resilient operation under variability, embedding transformer-based cyberattack detection and federated state reconstruction for early mitigation. A scalable digital twin supports operator training and rapid evaluation of automation strategies, validated on an ammonia–hydrogen–methanol benchmark to demonstrate scalability, interoperability, and decision-making under infrastructure and data-sharing constraints.

​

Physics-informed machine learning with organ-on-a-chip data for understanding disease progression and drug delivery dynamics: This project develops a framework integrating physiochemical multiscale modeling, organ-on-a-chip data, and advanced computational tools to enhance drug discovery and reduce costs. The integrated approach improves understanding of disease progression and drug delivery while optimizing organ-on-a-chip experiments to minimize trials and increase preclinical efficiency. Collaboration with the Terasaki Institute for Biomedical Innovation provides access to advanced organ-on-a-chip platforms with real-time monitoring. The hybrid modeling framework, combining machine learning and first-principles models, is applied to liver-on-a-chip systems for modeling nonalcoholic fatty liver disease and opioid hepatotoxicity. This methodology seeks to engineer therapeutics with superior efficacy, optimized dosing, and minimized toxicity.

​

Physics-informed machine learning-based predictive modeling for enhancing efficiency, sustainability, and resilience in global food systems: This project develops a PIML framework that integrates plant-based meat extrusion data with mechanistic models based on rheology, heat and mass transfer, and computational fluid dynamics to predict key processing parameters and final product attributes. Computer vision tools analyze the microstructure of plant-based meat products, providing objective quality metrics. Integrated with machine learning, these tools enable advanced feedback control to optimize raw material selection and processing inputs, improving product consistency and quality.

​

Optimization-based control of complex process networks

Complex chemical plants can be considered as integrated networks of combined lumped parameter systems (LPSs) (e.g., well-mixed reactors, staged separators), which can be described by ordinary differential equations (ODEs) and distributed parameter systems (DPSs) which are described by PDEs (e.g., heat exchangers, plug-flow reactors, packed beds). For such systems of systems, the solvability of the MHE/MPC, which involves the repeated online solution of a constrained dynamic optimization problem, becomes more crucial because the underlying optimization problem must be solved in the presence of algebraic-ODE-PDE constraints. Therefore, the MHE/MPC design may not be implementable in practice without reducing the associated computational costs. Such a challenge can be addressed through on-demand model order reduction and system decomposition.

​

Autonomous biomanufacturing for a sustainable, safe, and secure bioeconomy: This project pioneers the integration of next-generation process analytical technology with self-evolving digital twin ecosystems to transform monitoring, control, and optimization in biomanufacturing. Real-time in-line sensors track critical components and process variables. Machine learning and advanced statistics analyze these data to enable predictive rather than reactive operations, improving robustness and efficiency. Digital twins simulate and optimize process dynamics in real time, enabling AI-driven adaptive control and rapid identification of operating conditions that will enhance yield, lower costs, and accelerate scale-up.