Physics-informed machine learning for process modeling and optimization

Physics-informed neural networks (PINNs) include physical laws, such as balance equations, in the training of neural networks. This reduces the amount of data needed and prevents overfitting. We use physics-informed machine learning to model and optimize chemical reactors and processes.

Challenge

Data-driven methods often lack the domain knowledge that is well established in engineering. In chemical engineering, data are often scarce and expensive, and models must respect mass and energy balances to be trusted in design and operation. At the same time, systematic process design based on rigorous (superstructure) optimization is computationally demanding and still relies on engineering experience and heuristics.

Our approach

We combine deep learning with knowledge from first principles to improve the reliability and interpretability of models for chemical processes, from physical phenomena at the micro scale to unit operations and whole plants. Our methods include:

  • Physics-informed neural networks for dynamic models of chemical reactors, combined with time-series transformers.
  • Hard-constrained neural networks that satisfy balance equations exactly or up to a set tolerance, for example nonlinear enthalpy balances (Picard-KKT-hPINN) and general nonlinear constraints (ENFORCE).
  • Physics-informed surrogate models in optimization. Embedding high-fidelity surrogates into optimization routines reduces problem complexity and speeds up solvers (see optimization with surrogate models).

This work is part of our research on hybrid modeling.

Key publications

  • Lastrucci, G., Karia, T., Gromotka, Z., & Schweidtmann, A. M. (2025). Picard-KKT-hPINN: Enforcing nonlinear enthalpy balances for physically consistent neural networks. Systems and Control Transactions (Proceedings of ESCAPE 35). doi:10.69997/sct.108423 · Details
  • Lastrucci, G., & Schweidtmann, A. M. (2025). ENFORCE: Nonlinear constrained learning with adaptive-depth neural projection. arXiv preprint. arXiv:2502.06774 · Details
  • Lastrucci, G., Theisen, M. F., & Schweidtmann, A. M. (2024). Physics-informed neural networks and time-series transformer for modeling of chemical reactors. Computer Aided Chemical Engineering, 53, 571–576. doi:10.1016/B978-0-443-28824-1.50096-X · Details
← All projects