Gert Desmet | Searcharxiv

arXiv2025

Steady inertial flow of a compressible fluid in a spatially periodic channel under large pressure drops: a multiscale semi-analytical approach

Spatially-periodic channels are increasingly attracting attention as an efficient alternative to packed columns for a number of analytical and engineering processes. In incompressible flows, the periodic geometry allows to compute the flow structure by solving the Navier-Stokes (NS) equations in the minimal periodic cell of the structure, however large the pressure gradient. Besides, when gas flow under large pressure drops are dealt with, the velocity field is not periodic because of the density dependence on pressure. In this case, the momentum balance equations must be solved numerically on the entire channel, thereby requiring massive computational effort. Based on the marked separation of scales between the length of the periodic cell and the overall channel length characterizing many applications we develop a general method for predicting both the large-scale pressure and velocity profiles, and the small-scale flow structure. The approach proposed is based on the assumption that the local dimensionless pressure drop as a function of the Reynolds number, say $g({\rm Re})$, can be estimated from the solution of the incompressible NS equations within the minimal periodic cell of the channel. From the knowledge of $g({\rm Re})$, the pressure profile $P(Z)$ vs the large scale axial coordinate $Z$ is derived analytically by quadratures. We show how qualitatively different profiles $P(Z)$ can be obtained depending on the equation of state the gas. The approach is validated by comparing the predicted profiles with the full-scale numerical solution of the compressible NS equations in different axially-symmetric periodic channel geometries.

physics.flu-dyn↗

arXiv2023

MolGraph: a Python package for the implementation of molecular graphs and graph neural networks with TensorFlow and Keras

Molecular machine learning (ML) has proven important for tackling various molecular problems, such as predicting molecular properties based on molecular descriptors or fingerprints. Since relatively recently, graph neural network (GNN) algorithms have been implemented for molecular ML, showing comparable or superior performance to descriptor or fingerprint-based approaches. Although various tools and packages exist to apply GNNs in molecular ML, a new GNN package, named MolGraph, was developed in this work with the motivation to create GNN model pipelines highly compatible with the TensorFlow and Keras application programming interface (API). MolGraph also implements a chemistry module to accommodate the generation of small molecular graphs, which can be passed to a GNN algorithm to solve a molecular ML problem. To validate the GNNs, they were benchmarked against the datasets of MoleculeNet, as well as three chromatographic retention time datasets. The results on these benchmarks illustrate that the GNNs performed as expected. Additionally, the GNNs proved useful for molecular identification and improved interpretability of chromatographic retention time data. MolGraph is available at https://github.com/akensert/molgraph. Installation, tutorials and implementation details can be found at https://molgraph.readthedocs.io/en/latest/.

cs.LG↗