At the CAPE-OPEN 2020 Annual Meeting, Laurence MIDGLEY gave a presentation (access PDF here, access video here) on a software development he made: Distillation Gym.
Laurence MIDGLEY has been researching the use of artificial intelligence to design chemical engineering processes, during the gap between completing his Bachelor’s degree in Chemical Engineering at the University of Cape Town (where he first became interested in the topic), and studying an MPhil in Machine Learning and Machine Intelligence at the University of Cambridge.
He has been working with Jasper van BATEN (AmsterCHEM) and Harry KOOIJMAN (ChemSep) on creating an AI agent that designs processes within COCO simulator. He has used Python as the wrapper to do all the computing.
Reinforcement learning (RL), is a type of machine learning which involves an agent making decisions within an environment to maximise an expected reward. An illustrative example of this is the common application of RL to games (e.g. chess, Atari computer games) where the “environment” is the game simulation, the RL agent is the player and the reward is given for winning the game.
Reinforcement learning presents a novel approach to chemical engineering process synthesis with the potential to be applied to more open ended design problems than conventional computer aided process synthesis techniques. In RL for process synthesis, the environment is the simulator (e.g. COFE, Aspen Plus), the RL agent is the process designer and the reward is the objective function (e.g. profit).
In this talk, I describe a simple proof of concept, in which a reinforcement learning agent designs a hydrocarbon distillation column train simulated with COFE and ChemSep. Furthermore, I discuss how reinforcement learning for process synthesis relates to process simulation in general and how CAPE-OPEN may facilitate this application.