Abstract

This talk explores how process industries can increase autonomy by combining situational awareness with reliable decision-making algorithms. First, a safe learning and optimization framework will be presented that uses Gaussian process surrogate models to enable guided exploration of operating conditions while improving key performance indicators such as quality, throughput, and energy efficiency; an industrial case study will be used to illustrate the overall architecture, show how domain knowledge improves stability during early learning, and outline adversarially robust optimization strategies for maintaining performance under disturbances. Next, the focus will shift to how AI can support control system development: both model-based approaches—where controllers are built around identified models, including grey-box (physics + data) and fully data-driven models—and model-free approaches will be discussed, with particular emphasis on offline reinforcement learning to learn control policies directly from historical plant data. Finally, an agentic AI workflow for handling abnormal situations will be described, where a large language model proposes candidate actions, a validation agent tests them using digital twin simulations to assess feasibility and safety, and a reprompting agent helps refine the plan through an iterative loop. Overall, the talk outlines a practical route to greater autonomy—layered over existing control systems—combining interpretable state estimation, safe learning, and human oversight to improve performance and resilience without compromising safety. 

Biography

Mehmet Mercangöz is an Associate Professor in the Department of Chemical Engineering at Imperial College London, where he leads the Autonomous Industrial Systems Laboratory (AISL). His research focuses on the development of intelligent, adaptive, and autonomous systems for industrial operations, combining methods from optimization, control theory, machine learning, and artificial intelligence to address challenges in process systems engineering, with applications spanning energy systems, gas compression, manufacturing, and industrial decarbonization. Prior to joining Imperial, he worked at ABB in various R&D roles, developing model predictive control solutions for gas compressors — a technology that today underpins the reliable transport of natural gas from Norway to many European countries — and contributing to the development of advanced energy storage technologies and numerous optimization and control applications. He holds a PhD in Chemical Engineering from the University of California, Santa Barbara.