PCGym #17
Description
Your name, department, and University
Ammar Ali, Department of Chemical Engineering, Imperial College London
Name(s) and department(s) of anyone else relevant to this project
Dr Antonio del Rio Chanona
Please write a brief description of the application area of project
The project focuses on process control, a critical component of process industries, which produce essential materials such as fuels, pharmaceuticals, polymers, and specialty chemicals. These industries rely on large-scale, energy-intensive processes where safe, reliable, and efficient control is crucial for product quality, cost reduction, and sustainability.
However, many traditional control systems, like PID or model predictive control (MPC), struggle with high nonlinearity, uncertainty, and safety constraints that characterize modern chemical plants. Reinforcement learning (RL) and model-free control offer a promising path forward, but their safe deployment and benchmarking remain open challenges.
Please describe the project.
PCGym is an open-source simulation and benchmarking platform, a "gymnasium" for process control, analogous to OpenAI Gym but specifically designed for the chemical and process systems community.
The platform provides a unified environment where researchers and students can:
- Implement and test reinforcement learning, model-based, and model-free controllers.
- Benchmark algorithms across a range of realistic process control environments, such as continuous stirred tank reactors, distillation columns, and chemical production units.
- Study issues of safety, uncertainty propagation, and controllability that are specific to industrial chemical processes.
By providing standardized process environments with transparent APIs, PCGym enables reproducible research, encourages collaboration, and helps bridge the gap between academic control theory and industrial implementation.
What will be the outputs of this project?
- An open-source Python library implementing the PCGym framework.
- A suite of standardized process control environments for benchmarking RL and advanced control algorithms.
- Example tutorials and notebooks demonstrating how to deploy and evaluate agents on process control problems.
- Community contributions extending the benchmark suite to new chemical processes and control tasks.
- Potential publications or conference presentations demonstrating the use of RL for process control under uncertainty and safety constraints.
Which programming language(s) will this project use?
Python
Links to any relevant existing software repositories, etc.
https://github.com/MaximilianB2/pc-gym
Links to any relevant papers, blog posts, etc.
arXiv:2410.22093 – PCGym: A Gymnasium for Process Control Benchmarking
Project website
Make project public
- I understand that this project proposal will be public