Abstract
This project aims to develop a scalable physics-informed modelling methodology that can be utilized in
manufacturing industry with an initial focus on pharmaceutical and chemical processes. By combining hard
constraints, like mass and energy balances with soft constraints, like heat transfer coefficient and liquid
enthalpy, Hard-Soft Physics Informed Neural Networks (PINNs) will enhance data efficiency and ensure
physical validity. The methodology will be evaluated against established industry baselines and other hybrid
models using a solvent-switch use case at Johnson & Johnson, focusing on convergence, accuracy, data
efficiency, and integration with reinforcement learning (RL). This methodology will be the cornerstone for
resolving the scalability bottleneck for providing an end-to-end process optimization solution.
Researcher(s)
Research team(s)
Project type(s)