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 1×10⁵. The model predicted the Reynolds-averaged velocity and pressure field around the airfoil for arbitrary angles of attack between 10.0° and 17.5°. The model was trained with data from CFD 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 were trained. 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 of labeled training data in the domain. The trained network was capable of predicting the developing flow separation on the suction surface and exhibited excellent agreement with CFD results even in the proximity to the wall for interpolations as well as extrapolations from the labeled data set.

A method was developed to perform shape optimization of a tidal stream turbine hydrofoil using a multi-objective genetic algorithm. A bezier curve parameterized the refrence hydrofoil profoil NACA 63815. Shape optimization of this hydrofoil maximized its lift-to-darg ratio and minimized its pressure coefficient, thereby increasing the turbines power output power and improving its cavitation characteristics. The Elitist Non-dominated Sorting Genetic Algorithm (NSGA-II) was employed to perform the shape optimization. A comparative study of two-and three-dimensional optimizations was carried out. The effect of varing the angle of attack on the quality of optimized results was also studied. predictions based on two-dimensional panel method results was also studied. Preditions based on a two-dimensional panel method and on a computational fluid dynamics code were compared to experimental measurments.

Unsupervised physics-informed deep learning can be used to solve computational physics problems by training neural networks to satisfy the underlying equations and boundary conditions without labeled data. Parameters such as network architecture and training method determine the training success. However, the best choice is unknown a priori as it is case specific. Here, we investigated network shapes, sizes, and types for unsupervised physics-informed deep learning of the two-dimensional Reynolds-averaged flow around cylinders. We trained mixed-variable networks and compared them to traditional models. Several network architectures with different shape factors and sizes were evaluated. The models were trained to solve the Reynolds-averaged Navier-Stokes equations incorporating Prandtl's mixing length turbulence model. No training data were deployed to train the models. The superiority of the mixed-variable approach was confirmed for the investigated high Reynolds number flow. The mixed-variable models were sensitive to the network shape. For the two cylinders, differently deep networks showed superior performance. The best fitting models were able to capture important flow phenomena such as stagnation regions, boundary layers, flow separation, and recirculation. We also encountered difficulties when predicting high Reynolds number flows without training data

Experimental investigations were conducted to uncover the impact of cavitation control - through the use of riblets (mesoscale surface structuring) - on cavitating flows around a circular cylinder. First, the dynamics of cavitation in the flow behind a finite cylinder (without riblets) was unveiled by visualizing the cavitation clouds and measuring the lift force fluctuations acting on the cylinder. Second, as a significant step forward, a comprehensive explanation was provided for the mechanism of cavitation control using two bio-inspired riblet morphologies positioned in different orientations and locations on the cylinder. For the first time, the effects of these tiny structures on the flow dynamics and the associated cavitation process were scrutinized. This showed that scalloped riblets, with their curved design, induced secondary vortices near their tips and distorted primary streamwise vortices, and that high velocity gradients near the jagged pattern peaks of sawtooth riblets delayed flow separation, which affected cavitation.

In this work, we conducted a numerical study on the cavitation flow around a circular cylinder through the implementation of a porous coating. The primary objective addressed the effectiveness of utilizing a porous surface to control cavitation. We analyzed the cavitation dynamics around the cylinder and the hydrodynamic performance at different permeability levels of the porous surfaces. The flow was governed by the density based homogeneous mixture model and the volume penalization method was used to deal with the porous layer. A high-order accurate numerical method was adopted for the simulation of the cavitating flow through solving the preconditioned multiphase equations. The results revealed that the application of porous layer led to effective suppression of the cavitation vortex shedding. In addition, a reduction of the shedding frequency was obtained which accompanied by thinner and elongated vortices in the wake region of the cylinder. With the proper porous layer, the inception of the cavitation on the cylinder was suppressed and the amplitude of pressure pulsations due to the cavitation shedding mechanism was mitigated.