Skip to main content
Upcoming Events:
Artificial intelligence approaches towards hybridizing analytical & data-driven decision-making

Artificial intelligence approaches towards hybridizing analytical & data-driven decision-making

Date & Time:
   Add All to Calendar
Location:

BSB 106

The AI community typically focuses on models learned solely from data. But chemical engineering applications may also require explicit, parametric models, e.g. modeling process constraints, operations constraints and cost objectives

Overview

The AI community typically focuses on models learned solely from data. But chemical engineering applications may also require explicit, parametric models, e.g. modeling process constraints, operations constraints and cost objectives. So we consider integrating AI-based approaches together with more traditional process engineering strategies:

  • Design of experiments for model discrimination. We bridge the gap between classical, analytical methods and Monte Carlo-based approaches with surrogate models. 
  • Optimizing regression trees. We quantify the risk of evaluating a new data point and integrate tree models into larger decision-making problems.
  • Scheduling under uncertainty. For processes with equipment degradation, historical data and Bayesian optimisation approximate the uncertainty set. We use explainable AI tools to develop a theoretical and practical framework for explainable scheduling.

This presentation highlights ongoing collaborations with BASF, Royal Mail and Schlumberger.

Ruth Misener, Senior Lecturer