We are expanding the role of AI in various stages of process systems engineering, including design, modeling, and optimization. Recent work has explored the use of generative models to generate process flowsheets, supporting ideation during early-stage design. These models act as intelligent assistants, complementing domain knowledge and accelerating creativity in engineering workflows. We are also exploring the possibility of employing large language models (LLMs) for process modeling and optimization. By integrating AI into various frameworks, we aim to address the challenges of process systems engineering efficiently and robustly, adapting to data-rich environments and increasingly complex requirements.

We investigate both established and emerging control strategies to address the growing complexity of modern industrial processes, especially focusing on prevalent scenarios where models are inaccurate or unavailable. Our work explores advanced Model Predictive Control (MPC) schemes to manage complex models and inherent mismatches, alongside data-driven predictive control methods that reduce modeling effort and bias while remaining suitable for chemical systems. We also employ Reinforcement Learning (RL) as a fully data-driven approach to autonomously learn control policies directly from historical or simulated data, effectively handling nonlinearities, delays, and uncertainties without requiring explicit models. Furthermore, we aim to integrate MPC and RL to enhance the performance of conventional MPC and improve the reliability and interpretability of RL, building a unified framework for real-world industrial chemical plant control.

To model complex chemical processes with both accuracy and interpretability, we develop hybrid modeling frameworks that integrate physics-based equations with machine learning architectures. These models leverage mechanistic knowledge to ensure physical consistency while using data-driven components to capture unmodeled dynamics, address uncertainties arising from imperfect physics, and estimate unmeasurable process states or parameters. Recent work also explores the integration of such models into closed-loop control and optimization frameworks, as well as the use of techniques such as structure-aware neural networks and physics-informed learning algorithms to embed differential constraints directly into learning architectures. This approach supports applications in safety-critical industries—enabling robust inference even under sparse sensing or uncertain boundary conditions.

We establish machine learning-based tools to extract actionable insights from process data. Our work involves the development of explainable AI for in-depth fault diagnosis, enabling clear identification of root causes behind abnormal behaviors. To support continuous adaptation in evolving industrial settings, we investigate incremental learning approaches that enable real-time model updates in response to newly emerging fault scenarios. In addressing domain shifts across heterogeneous process environments, we incorporate adaptive learning techniques to enhance model robustness and generalizability under varying operational conditions. We also explore multimodal learning to extract complementary features that are often inaccessible through single-modality data alone. Collectively, we aim to advance intelligent monitoring systems that promote operational safety, support predictive maintenance, and minimize unplanned downtime in modern manufacturing processes.

We apply life cycle assessment (LCA) and techno-economic analysis (TEA) to evaluate and optimize process sustainability and profitability. These methodologies serve as foundational tools for data-informed decision-making in process design and operation. We develop an integrated workflow combining Aspen Plus, Excel, and Python to perform detailed economic evaluations. This framework enables Bayesian optimization for solving complex Mixed-Integer Nonlinear Programming (MINLP) problems, allowing us to explore trade-offs between capital investment, operating cost, and process performance. In parallel, our LCA framework quantifies environmental impacts such as greenhouse gas emissions, energy consumption, and resource depletion across the entire process life cycle. By linking process simulation outputs with LCA databases, we perform scenario analyses to identify environmental hotspots and guide process improvements toward net-zero and circular economy goals. Together, these approaches support the design of economically viable and environmentally responsible chemical processes.