Imperial College London
Alternate

ProfessorBenoitChachuat

Faculty of Engineering, Department of Chemical Engineering

Professor of Process Systems Engineering
 
 
 
//

Contact

 

b.chachuat Website

 
 
//

Location

 

609Roderic Hill BuildingSouth Kensington Campus

//

Summary

 

Overview

The overall goal of our research is to develop safe, economically and environmentally sustainable chemical and biological processes through the synergistic use of advanced computational modelling and optimisation methods and process data. Our vision of process systems engineering has a strong focus on rigorous computation to predict the performance at scale of both existing and novel technology and empower decision-making, and we work in close collaboration with experimentalists. We furthermore develop software tools that implement these methods to ensure their dissemination (https://github.com/omega-icl).

The principal areas of our research are currently threefold, with about identical emphasis on each area.

Analysis, optimization and Control of UNCERTAIN PROCESSES

show research

This entails making optimal decisions, either off-line or in real-time, despite limited or uncertain process knowledge. A particular focus has been on set-theoretic approaches for rigorous uncertainty propagation in nonlinear dynamic systems, including their implementation into software packages (MC and CRONOS). This unique capability can be exploited in algorithms to provide global optimality or robustness certificates. Another main focus has been on real-time optimization methods that rely on process data and feedback to drive an uncertain system to optimality. We have recently set out to extend this paradigm by embedding machine learning to compensate for the lack of physical process understanding.

Recent papers:

Design of Multiscale Systems

show research

This involves the transformation of trial-and-error practice for designing complex processes into systematic methods and tools. Our approach has relied on the combination of detailed process modelling with economic / environmental life-cycle assessment and risk management. The use of superstructure-based optimisation offers a natural and powerful framework in this context, but it remains a formidable computational challenge due to the combinatorial and nonlinear nature of these problems (MINLP) and the presence of uncertainty. On-going applications in our group have now widened to polygeneration from natural resources (wastewater, algal biomass, sugarcane, natural gas), catalytic CO2 conversion systems (methanol, higher hydrocarbons), and more recently vaccine manufacturing.

Recent papers:

Process ModelLING and Model Development Tools

show research

This entails constructing predictive mathematical models for novel process technology via the combination of mechanistic knowledge and process data. Our approach aims to strike the right balance between development effort, run-time complexity, and extrapolability. We have pioneered novel methods to support such model development: global optimisation and set-membership estimation (C package CRONOS), and more recently Bayesian estimation (python package DEUS) and optimal experiment design (python package pyDex). Another area of research is hybrid modelling, which combines parametric and nonparametric submodels based on different knowledge sources.

Recent papers: