Research

My research lies at the intersection of computational mechanics, constitutive modelling and scientific machine learning. I develop physics-constrained data-driven methods for complex materials, together with analytical, numerical and experimental approaches for structures subjected to extreme dynamic loading.

Thermodynamics-constrained constitutive modelling

I develop machine-learning methods for discovering constitutive equations directly from mechanical observations while enforcing the governing physical constraints by construction. This work includes Thermodynamics-based Artificial Neural Networks, autonomous discovery of internal variables, Neural Integration for Constitutive Equations, and the learning of inelastic constitutive models from stress–strain data under hard thermodynamic constraints. The objective is to obtain constitutive models that remain thermodynamically admissible, interpretable and reliable under loading paths not represented during training.

Framework for learning inelastic constitutive models from stress-strain data under hard thermodynamic constraints — Learning inelastic constitutive behaviour from stress–strain data under hard thermodynamic constraints.
*Collaborators: F. Vincent, V. Acary, I. Einav, I. Stefanou.*

Data-driven multiscale mechanics

I study heterogeneous materials whose macroscopic response emerges from nonlinear and irreversible mechanisms evolving at finer spatial scales. My research combines micromechanical simulations, computational homogenisation and thermodynamics-aware machine learning to construct efficient reduced-order constitutive descriptions. Learned internal variables provide a compact representation of the microscopic state, while constitutive surrogates retain the essential effects of path dependence, dissipation and material heterogeneity.

Masonry under extreme loading, reduced-scale experiments and scaling laws

Part of my research focuses on the investigation of the response and failure of masonry, monumental structures and free-standing artefacts subjected to explosions and other extreme dynamic actions. These systems combine complex geometry, unilateral contact, friction, cracking, rocking, block separation and fluid–structure interaction. My work uses analytical models, discrete-element methods, nonlinear finite-element simulations and coupled solid–fluid calculations to identify damage and collapse mechanisms and support mechanically informed risk assessment. Within the PhD thesis of A. Morsel, in collaboration with I. Stefanou and P. Kotronis, we developed scaling approaches and reduced-scale experiments for reproducing the response of masonry and blocky structures under blast-type loading. The work examines how rocking, sliding, uplift, impact and collapse mechanisms can be represented consistently at laboratory scale. Controlled shock waves, high-speed imaging, pressure measurements and three-dimensional motion tracking provide detailed observations for validating computational models and linking laboratory measurements with full-scale structural behaviour.

Filippo Masi

Thermodynamics-constrained constitutive modelling

Data-driven multiscale mechanics

Masonry under extreme loading, reduced-scale experiments and scaling laws