Stefano Riva

Stefano Riva

Researcher in computational methods for nuclear reactors, with a focus on Reduced Order Modelling and Machine Learning for state estimation. Passionate about bridging physics-based models and data-driven approaches.

pyforce

pyforce is a Python framework for data-driven reduced-order modelling (DDROM), developed within the ERMETE Lab, Politecnico di Milano, for applications to complex multi-physics problems, with a strong focus on nuclear engineering.

pyforce logo

The package is part of the broader ROSE project (Reduced Order modelling with data-driven techniques for multi-physics problems), which aims to reduce the computational complexity of large-scale models, identify optimal sensor locations, and integrate experimental measurements to improve physical predictions.

Description

pyforce provides a unified environment for applying modern reduced-order modelling techniques to simulation data. Version 1.0.0 introduces a major redesign of the codebase: instead of relying on finite-element environments, it now uses pyvista for mesh handling, integrals and data import, and stores fields as NumPy arrays. This makes the framework solver-agnostic and compatible with any tool capable of exporting VTK files.

The general workflow is divided into:

  • Offline phase: dimensionality reduction, extraction of a reduced basis, and optimal sensor placement.
  • Online phase: data assimilation, reconstruction of system states, and estimation of non-observable fields.

The framework gathers several algorithms used in reduced-order modelling and data assimilation, including:

  • Singular Value Decomposition and Proper Orthogonal Decomposition
  • (Generalised) Empirical Interpolation Method, with optional regularisation
  • Parameterised-Background Data-Weak formulation
  • SGreedy algorithm for optimal sensor placement
  • Indirect reconstruction techniques for unobserved fields

These tools have been successfully applied to fluid dynamics, thermal–hydraulics, magnetohydrodynamics, and reactor physics problems.

Tutorials and Documentation

Detailed documentation, introductory guides, and a growing set of tutorials are available online. They cover the basic concepts of SVD/POD, GEIM, sensor placement, data assimilation, and applications to both fluid-dynamics and neutronics case studies, with examples for training ROMs and reconstructing physical fields.

Authors

pyforce is developed and maintained at the Nuclear Reactors Group – ERMETE Lab by Stefano Riva, under the supervision of Dr. Carolina Introini and Prof. Antonio Cammi.