Podcast: Data-Driven Fluid Mechanics & Academic Life

In this podcast episode, I joined Jousef Mourad for an open conversation on data-driven fluid mechanics, the evolving role of machine learning in engineering, and the interplay between physics-based modeling and modern data-centric approaches.

Beyond research, we also discussed the realities of academic life, the PhD journey, career choices, and the challenges and rewards of working at the interface between theory, computation, and experiments. The episode is intended as an informal and accessible exchange, aimed at students and early-career researchers interested in fluid dynamics and scientific research more broadly.

ERC Starting Grant: Experience & Advice (FNRS Workshop)

Recorded panel organized by NCP-FNRS to share feedback and strategic guidance for researchers interested in ERC and Horizon Europe funding schemes.

The discussion covers proposal structure, evaluation criteria, positioning of scientific vision, and practical aspects of preparing competitive European grant applications.

Data-Driven Modal Decompositions in Fluid Dynamics (2-day short course)

An intensive two-day course on data-driven modal analysis, delivered at the American University of Beirut, invited by Prof. Sara Najem. The course revisits the mathematical foundations of modal decompositions and connects them to practical goals: coherent-structure identification, filtering, data compression, and reduced-order modeling.

The programme is organized as four lecture blocks plus hands-on tutorials: Linear autoencoders and links to modal analysis, POD, DFT and DMD, and multiscale POD (mPOD), with afternoon sessions dedicated to guided exercises and implementation.

Generalized and Multiscale Modal Analysis (lecture)

This lecture introduces a unified view of modal decompositions for fluid-mechanics data, starting from the classical idea of extracting coherent structures from snapshots and then expanding it to generalized inner products and multi-resolution / multi-scale formulations. The emphasis is on how the choice of metric (e.g. energy-weighted, physics-informed, or data-driven kernels) changes what “optimal” means, and how scale separation can be enforced to isolate phenomena evolving on different time (or frequency) ranges.

The discussion links these ideas to practical workflows in experiments and simulations: denoising, identifying dominant structures, building reduced-order representations, and interpreting modes.

Continuous and Discrete LTI Systems (lecture)

This lecture reviews the core ideas behind Linear Time-Invariant (LTI) systems in both continuous time and discrete time (SISO setting). It starts from basic notions of signals and system representations, then builds up the key properties of LTI systems in the time and frequency domains.

A central part is the bridge between the two worlds: the mapping between continuous and discrete formulations (sampling, conformal mapping), and how this connects differential vs difference equation models. The lecture closes with two practical outcomes: using linear models for time-series analysis / forecasting and designing digital filters for multi-resolution analysis.

Constrained Regression for Data Assimilation of Velocity Fields

Mini-course introducing physics-constrained regression techniques for the super-resolution and data assimilation of velocity fields from image velocimetry. The lectures present the mathematical foundations of constrained least-squares, regularization, and gradient-aware formulations.

The course also introduces SPICY (Super-resolution and Pressure from Image veloCimetrY), an open-source software developed at VKI for meshless data assimilation in fluid mechanics.

Data-Driven Modal Analysis: Theory, Algorithms, and Practice

Mini-course dedicated to data-driven modal analysis for the decomposition of complex flow fields in time and space. The lectures cover Proper Orthogonal Decomposition (POD), Dynamic Mode Decomposition (DMD), and generalized multiscale extensions for transient and non-stationary flows.

The course introduces MODULO, an open-source Python framework for modal decompositions designed for flexibility, reproducibility, and integration with experimental and numerical fluid dynamics workflows.