Vadlamani Nagabhushana Rao

Temperature prediction is critical for many industrial and everyday applications. Numerical Weather Prediction (NWP) models using high-performance computing is the most sought technique to forecast weather, including temperature. However, NWP is complex in nature and computationally expensive. In this paper, the temperature is forecast using data-driven Machine Learning techniques, which are not computationally intensive and are further accelerated using GPUs. Two deep learning models: A stacked Long Short-Term Memory (LSTM) and Random Forest Regressor (RFR), are developed and validated using the standard ERA5 data (at 850hPa, above the atmospheric boundary layer). In addition, the models are tested against the ground-level observations (inside the atmospheric boundary layer) for twenty different locations in India. The performance of univariate and multivariate models is also analyzed for the real-time dataset. Root Mean Square Error (RMSE) obtained by the LSTM and RFR are 0.47 and 0.23, respectively, for ERA5 data. When compared to the numerical weather prediction model - operational IFS, the RMSE using LSTM and RFR is smaller by 65% and 83%, respectively. The LSTM and RFR models forecast temperature with an average RMSE of 0.7 for the real-time data at twenty locations. The GPU-enabled LSTM model performed 64 times faster than the CPU-enabled model. The developed RNN models are made publicly available at https://github.com/arasuadrian/RNN-Models.

Linear stability analysis of an inviscid liquid sheet with different velocity profiles across its thickness is reported. The velocity profiles for which there is a progressive increase or decrease in velocities between the two interfaces are demonstrated to be inherently unstable even in the absence of the destabilizing aerodynamic shear at the liquid-gas interfaces. Compared to a flat velocity profile, a linear or a parabolic profile, symmetric at the center line of the sheet reduced both the maximum growth rate and the wavelength range over which the waves grow. The convective acceleration from the velocity gradient is found to stabilize longer waves while the growth of shorter waves is hampered by the combined effect of the surface tension and a decrease in the interface velocity between gas and liquid media. The wave forms are dominantly sinuous for symmetric velocity profiles; however, with larger velocity gradients the dilatational modes are observed. The inherent instability of liquid sheets with a progressive change in velocities between the interfaces is seen to arise from the differential convective acceleration at the two interfaces in the plane of reference of the liquid sheets. © 2009 American Institute of Physics.

A computationally efficient high-order solver is developed to compute the wall distances by solving the relevant partial differential equations, namely: Eikonal, Hamilton–Jacobi (HJ) and Poisson equations. In contrast to the upwind schemes widely used in the literature, we explore the suitability of high-order central difference schemes (explicit/compact) for the wall-distance computation. While solving the Hamilton–Jacobi equation, the high-order central difference schemes performed approximately 1.4–2.8 times faster than the upwind schemes with a marginal improvement in the solution accuracy. A new pseudo HJ formulation based on the localized artificial diffusivity (LAD) approach has been proposed. It is demonstrated to predict results with an accuracy comparable to that of the Eikonal equation and the simulations are ≈ 1.5 times faster than the baseline HJ solver using upwind schemes. A curvature correction is also incorporated in the HJ equation to correct for the near-wall errors due to concave/convex wall curvatures. We demonstrate the efficacy of the proposed methods on both the steady and unsteady test cases and exploit the unsteady wall-distance solver to estimate the instantaneous shape and burning surface area of a dendrite propellant grain in a solid propellant rocket motor.

In this paper, we study the interaction of roughness induced streaks with pulses of free-stream turbulence towards promoting bypass transition of a laminar boundary layer. A series of eddy resolving simulations are carried out to simulate a transitional boundary layer developing over a flat plate under zero pressure gradient. An isolated roughness element of fixed height is modelled using immersed boundary method and Jarrin’s synthetic eddy method is used to impose free-stream turbulence (FST) at the inlet. We impose pulses of FST to trigger bypass transition in contrast to the continuous forcing typically studied in the literature. For a fixed turbulence intensity and length scale, the pulsing frequency of FST is varied. Interaction between the Klebanoff streaks due to FST with the roughness-induced high and low speed streaks is studied for different pulsing frequencies. Variation in the slope of streamwise evolution of skin friction coefficient is discussed in terms of the relaxation time of the roughness induced streaks which are disturbed when FST pulses pass over it. The dynamics of this interaction and the role it plays in subsequently triggering the bypass transition has been investigated.

In this paper, we investigate the coupled interaction between a new short intake design with a modern fan in a high-bypass ratio civil engine, specifically under the off-design condition of high incidence. The interaction is expected to be much more significant than that on a conventional intake. The performance of both the intake-alone and rotor-alone configurations are examined under isolation. Subsequently, a comprehensive understanding on the two-way interaction between intake and fan is presented. This includes the effect of fan on intake angles of attack (AoA) tolerance (FoI) and the effect of circumferential and radial flow distortion induced by the intake on the fan performance (IoF). In the FoI scenario, the rotor effectively redistributes the mass flow at the fan-face. The AoA tolerance of the short-intake design has increased by ≈4 deg when compared with the intake-alone configuration. Dynamic nature of distortion due to shock unsteadiness has been quantified. ST plots and power spectral density (PSD) of pressure fluctuations show the existence of a spectral gap between the shock unsteadiness and blade passing, with almost an order of magnitude difference in the corresponding frequencies. In the IoF scenario, both the “large” (O(360 deg)) and “small” scale distortion (O(10-60 deg)) induced by the intake results in a non-uniform inflow to the rotor. Sector analysis reveals a substantial variation in the local operating condition of the fan as opposed to its steady characteristic. Streamline curvature, upwash, and wake thickening are identified to be the three key factors affecting the fan performance. These underlying mechanisms are discussed in detail to provide further insights into the physical understanding of the fan-intake interaction. In addition to the shock-induced separation on the intake lip, the current study shows that shorter intakes are much more prone to the upwash effect at higher AoA. Insufficient flow straightening along the engine axis is reconfirmed to be one of the limiting factors for the short-intake design.

This work investigates the influence of V-shaped riblets on skin friction drag reduction of transitional and turbulent boundary layers under zero pressure gradient. High fidelity eddy resolving simulations are used to demonstrate the efficacy of riblets. The riblets are represented using the Boundary Data Immersion Method (BDIM). In this paper, we validate the BDIM framework on two test cases: flow past a circular cylinder and minimal span channel with sinusoidal roughness. Subsequently we report the results on a transitioning boundary layer under the influence of free stream turbulence. The role of riblets is analyzed through the boundary layer integral parameters and skin friction coefficient. Time and span averaged turbulent statistics are presented in discerning the influence of riblets close to the wall. Analysis of flow statistics over a single riblet element is also performed through time and phase averaging. The efficacy of riblets in delaying the onset of turbulent breakdown and reducing skin friction drag is demonstrated through averaged and instantaneous flow quantities.

This paper studies the computational modelling of the flow separation over the engine nacelle lips under the off-design condition of significant crosswind. A numerical framework is set up to reproduce the general flow characteristics under crosswinds with increasing engine mass flow rate, which include: low speed separation, attached flow and high speed shock-induced separation. A quasi-3D (Q3D) duct extraction method from the full 3D (F3D) simulations has been developed. Results obtained from the Q3D simulations are shown to largely reproduce the trends observed (Isentropic Mach number variations and high-speed separation behaviour) in the 3D intake, substantially reducing the computational cost by a factor of 50. The agreement between the F3D and Q3D simulations is encouraging when the flow either fully attached or with modest levels of separation, but degrades when the flow fully detaches. Results are shown to deviate beyond this limit, since the captured streamtube shape (and hence the corresponding Q3D duct shape) changes with the mass flow rate. Interestingly, the drooped intake investigated in the current study is prone to earlier separation under crosswinds when compared to an axisymmetric intake. Implications of these results on the industrial nacelle lip design is also discussed.

Information loss in numerical physics simulations can arise from various sources when solving discretised partial differential equations. In particular, errors related to numerical precision (“sub-precision errors”) can accumulate in the quantities of interest when the simulations are performed using low-precision 16-bit floating-point arithmetic compared to an equivalent 64-bit simulation. On the other hand, low-precision computation is less resource intensive than high-precision computation. Several machine learning techniques proposed recently have been successful in correcting errors due to coarse spatial discretisation. In this work, we extend these techniques to improve CFD simulations performed with low numerical precision. We quantify the precision-related errors accumulated in a Kolmogorov forced turbulence test case. Subsequently, we employ a Convolutional Neural Network together with a fully differentiable numerical solver performing 16-bit arithmetic to learn a tightly-coupled ML-CFD hybrid solver.1 Compared to the 16-bit solver, we demonstrate the efficacy of the hybrid solver towards improving various metrics pertaining to the statistical and pointwise accuracy of the simulation.

Low pressure turbines (LPT) typically operate at low Reynolds numbers, of O(105), resulting in transitional boundary layers on the suction surface. These boundary layers are prone to separation under the strong local adverse pressure gradients on the aft portion of the blade. Intermittent free-stream turbulence, periodic wakes shed by the upstream blades and surface roughness due to gradual degradation of the blades have been shown to suppress the separation bubble on the suction surface of the LPT blade. Although this generally leads to a profile loss reduction, some of the benefit is offset by a loss increase associated with a larger turbulent wetted area. In this work, we explore a strategy where the losses in both the transitional and turbulent boundary layers can be reduced. In particular, we employ surface roughness in the transitional regime to trip the boundary layer and reduce the separation bubble related losses. Subsequently, we explore the possibility of utilizing riblets in the turbulent regime to further reduce the losses due to the turbulent wetted area. To investigate the efficacy of this 'rough-riblet blade surface', high fidelity eddy-resolving simulations are carried out on the configuration of a flat surface subjected to streamwise varying pressure gradients. A contoured upper wall is used to mimic the pressure distribution typically encountered on the suction surface of a high lift turbine blade. The Boundary Data Immersion Method is used to represent different riblet shapes and roughness elements, while a digital filtering approach is used to impose free-stream turbulence at the inlet. The influence of different riblet geometries on skin friction drag of transitional and turbulent boundary layers under streamwise varying pressure gradients are discussed in detail. The effect of different riblet shapes in conjunction with the roughness elements on the passage losses is shown through boundary layer integral parameters and Reynolds stresses.

Transitional boundary layers on low-pressure turbines (LPTs) are prone to separation on the suction surface of the blade under strong local adverse pressure gradients. Intermittent freestream turbulence, periodic wakes shed by the upstream blades, and surface roughness due to in-service degradation of the blades are shown to suppress the separation. Although this generally leads to a profile loss reduction, some of the benefits are offset by a loss increase associated with an increased turbulent wetted area. In this work, we explore a strategy where the losses in both the transitional and turbulent boundary layers can be reduced. In particular, we employ surface roughness in the transitional regime to reduce the separation bubble-related losses and riblets in the turbulent regime to further reduce the losses due to the turbulent wetted area. The efficacy of this 'rough-riblet blade surface' is studied using high-fidelity eddy resolving simulations on the configuration of a flat surface subjected to streamwise varying pressure gradients. Two riblet shapes, sawtooth and scalloped, are considered. When compared to the roughness alone configuration, scalloped riblets reduced the skin friction drag by ≈10% and are much more effective than the sawtooth riblets. Through the streamwise evolution of the boundary layer parameters such as trailing edge momentum thickness, maximum turbulent kinetic energy, and Reynolds stresses, the additional losses incurred at the junction between the smooth wall and riblet leading edge are highlighted.