Physics-informed neural networks (PINN) can be used to predict flow fields with a minimum of simulated or measured training data. As most technical flows are turbulent, PINNs based on the Reynolds-averaged Navier–Stokes (RANS) equations incorporating a turbulence model are needed. Several studies demonstrated the capability of PINNs to solve the Naver–Stokes equations for laminar flows. However, little work has been published concerning the application of PINNs to solve the RANS equations for turbulent flows. This study applied a RANS-based PINN approach to a backward-facing step flow at a Reynolds number of 5100. The standard k- model, the mixing length model, an equation-free and an equation-free pseudo-Reynolds stress model were applied. The results compared favorably to DNS data when provided with three vertical lines of labeled training data. For five lines of training data, all models predicted the separated shear layer and the associated vortex more accurately.
Physics-informed neural networks are a promising method to yield surrogate models of flow fields. We present a metamodeling technique for variable geometries based on physics-informed neural networks. The method was applied to the DU99W350 airfoil at a Reynolds number of $$1\times 10^{5}$$ . Using our technique, the angle of attack was introduced as an additional input parameter of the network and the model was trained to predict the Reynolds-averaged velocity and pressure fields around the airfoil for arbitrary angles of attack between 10.0 and 17.5 $$^{\circ }$$ . Furthermore, we present an effective method to generate the training points for the parameterized geometry. The model was trained with data from simulations for a limited set of angles of attack. Additionally, satisfaction of the a priori known boundary conditions as well as the Reynolds-averaged Navier–Stokes equations was attained. A sensitivity analysis concerning the Reynolds number, the amount and distribution of training data, and the turbulence model was conducted showing the superiority of the pseudo-Reynolds stress method and the demand for labeled training data in the domain. The trained network was capable of predicting the flow separation progressing with angle of attack on the suction surface and exhibited excellent agreement with numerically simulated results, even in the proximity of the wall for interpolations as well as extrapolations from the labeled data set. Our study demonstrates that physics-informed neural networks can be used to obtain accurate flow field surrogate models of variable geometries.
This systematic literature review investigates the use of physics-informed neural networks (PINNs) in electromagnetics by examining peer-reviewed articles and conference papers. By integrating governing physical laws into the loss function of a neural network, PINNs offer a mesh-free method in scientific computing. Records published between 2020 and 2025 were retrieved from the databases Scopus, Web of Science, and IEEE Xplore. The initial dataset comprised 500 records, from which 292 unique publications were identified. These were screened, yielding a final set of 139 publications that met predefined eligibility criteria. The analysis reveals growth in research activity, with a pronounced increase from 2022 onward. The literature predominantly addresses electrodynamic problems, employs feedforward neural network architectures, and adopts physics-only training. Two-dimensional problem formulations dominate, with three-dimensional formulations concentrated almost exclusively in electrodynamics, and no publications addressing electroquasistatics were identified. Contingency tables show that methodological choices are not independent of problem characteristics: medium selection correlates with physics regime, and architectural diversity increases with spatial dimensionality. Based on these findings, priorities for future work include: addressing the gap in electroquasistatics, extending three-dimensional formulations to static and quasistatic regimes, broader architectural experimentation in lower-dimensional settings, and increased integration of labeled data in static electromagnetics. To support methodological consistency and reproducibility, a reporting checklist for future PINN-based electromagnetics publications is proposed.
This systematic literature review (SLR) investigates the use of physics-informed neural networks (PINNs) in electromagnetics by examining peer-reviewed articles and conference papers. By integrating governing physical laws into the loss function of a neural network (NN), PINNs offer a mesh-free method in scientific computing. The reviewed records were retrieved from the databases Scopus, Web of Science, and IEEE Xplore, published between 2020 and 2025. The initial dataset comprised 500 records of which 292 unique publications were identified. These were screened, yielding a final set of 139 publications that met predefined inclusion criteria. The analysis reveals a growth in research activity, with a pronounced increase from 2022 onward. The literature predominantly addresses electrodynamic problems, employs feedforward neural network architectures and adopts unsupervised, physics-driven training paradigms. Two-dimensional problem formulations are considerably more prevalent than three-dimensional formulations, and advanced architectures remain limited. Further analysis reveals the associations between the extracted characteristics. The choice of medium dependends on the physics regime, and architectural diversity increases with spatial dimensionality. The review's conclusions identify potential priorities for future work: extending three-dimensional formulations to the static and quasistatic electromagnetic regimes, broader architectural experimentation particularly in lower-dimensional settings, and increased use of semi-supervised learning in static electromagnetic applications.