Computational Fluid Dynamics group
The research work of the Computational Fluid Dynamics (CFD) group ranges from the design, implementation and demonstration of novel computational technologies to the analysis of complex flow problems encountered in mechanical, aeronautical and civil engineering, and in biomedical applications. Many of our current projects are interdisciplinary concentrating on a wide rage of topics within the field of energy, bioengineering, thermofluids and combustion. The algorithm development work includes the implementation of time- and frequency-domain solvers for the analysis of unsteady periodic flows, the development of accurate and computationally efficient low-speed preconditioners for the compressible Navier-Stokes solver, and novel approaches to high-performance parallel computing for CFD applications. The group has close connections with the other research divisions and groupings across the University, other academic institutions, NHS and industries in the UK and abroad. Some of our ongoing research projects are reported below.
The main research programmes of the group are presently:
- Combustion modelling
- Biomedical flows
- Low-speed preconditioning for compressible URANS solvers
- Frequency-domain Navier-Stokes solvers for wind turbine unsteady aerodynamics
- Other renewable energy applications
- Hybrid parallelization technology
- Farfield boundary conditions for turbomachinery flows
- Industrial turbomachinery projects
- Micro- and nano-thermofluids
- Natural transition and boundary layer flows
Combustion modelling
Dr. Manosh Paul
Current research in this area concentrates on the prediction of turbulent reacting flow, combustion species and emissions (NOx, Soot) by using an in-house Large Eddy Simulation (LES) code. A recently developed Discrete Ordinate Method (DOM) code, which is coupled with the LES in a general body-fitted coordinate system has enabled the group to model accurately the flame-radiation interaction. The code has successfully been applied to the Rolls-Royce Tay Gas Turbine Combustor. Current research in this area is being carried out in a close collaboration with academics in Imperial College London.
Biomedical flows
Dr. Manosh Paul
Project 1. We are working closely with medical doctors and consultants at Glasgow Royal Infirmary, Southern General Hospital and Queen Margaret’s Hospital in a number of biomedical research projects. In the framework of a research programme funded by Bayer Public Limited Company, we are investigating the dependence of contrast media viscosity on the injection rate. The objective is to study the resulting variations of the vessel wall shear stress in relation to the risk of contrast extravasations. We are also performing patient specific multi-phase simulations to develop a novel CFD-based treatment modality for delivering the chemotherapy dose for cancer treatment in Head and Neck vessel network.
Project 2. The analysis of blood flow dynamics plays an important role in the diagnosis of arterial diseases such as atherosclerosis and stenosis; for example the quantification of arterial stenosis by duplex ultrasound and quantitative flow MRI techniques rely on measurements of the flow velocity/acceleration to infer the degree of underlying stenosis. We are working on the development of a Fluid Structure Interaction (FSI) methodology based on LES to investigate the pathophysiology of transient blood flow through stenosis. A recent collaboration with Dr Craig J Daly, Medical, Veterinary & Life Sciences has been established to study the blood flow at the Resistance Vascular Branch Points.
Low-speed preconditioning for compressible URANS solvers
Dr. Sergio Campobasso
Two computational approaches are available for the solution of low Mach number flows. One is the use of the incompressible Navier-Stokes (NS) equations, and the other is the use of the compressible NS equations . One of the advantages of the compressible formulation is the possibility of using the solver to perform aeroacoustic analyses. Another advantage is the capability of analyzing problems in which large differences of flow velocities exist. An example is provided by the flow field past a vertical take-off aircraft, where the velocities in the lifting jet are substantially higher than the velocity of the aircraft. When using the compressible equations for low-speed flows, however, the accuracy of the solution decreases due to the large disparity between the magnitudes of convective and acoustic eigenvalues. Such a disparity results in unbalanced amounts of numerical dissipation, and this has a detrimental effect on the solution accuracy. In addition, the convergence rate of explicit multigrid solvers is also impaired by such a disparity. Both problems can be solved by supplementing the compressible solver with Low-speed preconditioning (LSP), a mathematical/numerical method that enables one to optimize the amount of numerical dissipation and maintain a high-convergence rate independently of the flow speed. This enables one to use a single CFD solver to analyze low-, high- and mixed-speed flows of compressible fluids. Due to the abovesaid flexibility of the compressible formulation, we have chosen to develop a compressible unsteady Reynolds-averaged Navier-Stokes (URANS) solver featuring an accurate and computationally efficient LSP algorithm. The figure on the right-hand-side reports the Mach number contours past a wind turbine airfoil for a freestream Mach number of 0.001 computed with our compressible solver featuring LSP.
Frequency-domain Navier-Stokes solvers for wind turbine unsteady aerodynamics
Dr. Sergio Campobasso
Several Horizontal Axis Wind Turbine (HAWT) unsteady flows can be viewed as periodic. This is the case of the flow regimes associated with flow-induced vibrations (dynamic stall) and the yawed wind conditions. The latter phenomenon occurs when the wind direction is not orthogonal to the rotor plane. The reliable design of modern large HAWT's increasingly relies on the accuracy by which such complex aerodynamic phenomena can be predicted. High-fidelity computational tools such as those based on the 3D Navier-Stokes (NS) equations potentially possess the required accuracy, but their solution in the time-domain requires very long run-times even when intensive parallel computing is used. The formulation and solution of the NS equations in the frequency-domain, however, allows one to reduce run-times by at least one order of magnitude with respect to those required by the corresponding time-domain solver. This project focuses on the development of a frequency-domain solver of the compressible NS equations based on a harmonic balance formulation of the equations. The solver features an optimized low-speed preconditioner, and uses an explicit multigrid approach for the integration of the frequency-domain equations. Due to the use of a compressible formulation, the code is equally applicable to low-speed as well as transonic and supersonic flows. The research code being developed is also particularly well suited for turbomachinery applications. In these problems, the use of suitable periodicity boundary conditions allows one to simulate the flow field of a single blade passage rather than the flow of the whole blade-row being considered. The figure on the right-hand-side reports the lift coefficient hysteresis cycle of a HAWT blade section in yawed wind computed with our TD solver (curve labeled TD 128) and HB solver using 3, 4 or 5 harmonics (curves labeled HB n). The HB calculation with 3 harmonics is about 20 times faster than the TD simulation.
Other renewable energy applications
Dr. Manosh Paul
Project 1. In a recent collaborative project with Dr Ian Watson on biomass combustion, we are investigating the performance of 3-phase (gas, solid and tar) combustion of biomass, and how the important parameters such as temperature, pressure, particle size and shape, moisture content and so forth affect the combustion process. Ultimately, the goal is to achieve a suitable combination of parameters that would potentially reduce emissions by maximising the combustion efficiency.
Project 2. We are investigating the effect of mounting type and geometry on the temperature of a Photovoltaic (PV) panel, and the effects of heat convection in the air gap between panel and building. The temperature of a PV panel has a strong effect on its efficiency. From a series of experiments and CFD modelling our aim is to achieve the optimum mounting conditions for a PV panel. This is a joint project with the Energy and Management Department of the University of L'Aquila, Italy.
Hybrid parallelization technology
Dr. Sergio Campobasso
The computer analysis of large problems, such as the calculation of the unsteady flow past horizontal axis wind turbines (HAWT's) based on the Navier-Stokes equations requires large computational resources and the use of high-performance parallel computing. Until now, the majority of available CFD solvers have been using distributed parallel computing based on the message-passing interface (MPI) paradigm. The MPI parallelization is usually quite efficient, but certain features of the application code at hand may limit the effectiveness of a pure MPI parallelization. This is the case of multi-block Harmonic Balance (HB) Navier-Stokes solvers. The geometry and size of the multi-block computational domain required to compute an unsteady periodic flow field using the HB solver or its time-domain (TD) counterpart is identical. The memory requirement of the HB solver, however, is about 2N times that of the TD solver, where N is the number of complex harmonics retained in the representation of the periodic solution. The consequence of this is that a given multi-block computational domain suitable for a TD analysis, may become unsuitable for a HB analysis due to the fact that the increased memory requirement of each block may exceed the memory of the computational nodes of the cluster. One way of solving this problem would be to further split the given domain into smaller blocks, but for large problems this may reduce the parallel performance of the MPI parallelization due to excessive communications among processes. Given that the computational nodes of modern clusters often consist of multi-core shared-memory processors, a more effective solution appears to be that of using a hybrid parallelization strategy, namely a distributed-memory MPI structure across computational nodes whereby each node takes a geometric block, and a distributed shared-memory OpenMP structure across the cores of a given node, whereby each core takes a particular harmonic of the sought periodic flow field. This hybrid parallelization structure is presently being implemented in our HB NS solver in collaboration with EPCC. The figure on the right-hand-side reports the computational speed-up of our HB CFD featuring a preliminary pure MPI parallelization.
Farfield boundary conditions for turbomachinery flows
Dr. Sergio Campobasso
Fan, compressor and turbine blades operate in annuli which, in the majority of cases, are delimited by non-parallel end walls. The 3D CFD analysis of turbomachinery blade-rows based on the Euler or Navier-Stokes (NS) equations focuses on a set made up of one row, one stage or multiple stages. In all cases, the computational domain is the portion of the annulus containing the blade set under investigation. The inflow and outflow boundaries are often delimited by non-parallel end walls, and this may result in strong radial and streamwise gradients of the flow field. Such gradients may also occur when the end walls are parallel but the mean radius of the annulus varies along the flow path. In the case of subsonic farfield boundaries, 1D characteristics are typically used to implement the boundary conditions (BC’s) at both the inflow and outflow boundaries of the domain, in addition to user-given data such as total pressure, total temperature and flow angles at the inlet boundary and static pressure at the outlet one. In the presence of flow variations in the streamwise direction at either farfield boundary, the straightforward ’conventional’ use of 1D characteristics reduces the global accuracy of the solver. For example, a CFD solver using a second-order space discretization delivers results which are only first-order accurate. This circumstance may significantly reduce the overall accuracy and reliability of engineering analyses, particularly when the farfield boundaries are relatively close to the blade edges. Functionals of engineering interest most likely affected by the use of insufficiently accurate farfield BC’s are the flow incidence and the mass flow rate. The former parameter crucially affects the blade loading, while the latter one is also important to determine the overall characteristics of turbomachinery components.
As for the practical implementation of the farfield BC’s, these can be based on a) the use of auxiliary flow states, introduced to compute the numerical fluxes on the boundary faces with the same scheme used for the interior faces, or b) the direct enforcement of suitable boundary fluxes. The accuracy loss arising in the presence of strong flow gradients at the farfield boundaries is common to both architectures. This project focuses on the analysis of this issue, and the design of simple and effective enhancements of both BC implementations to resolve this accuracy problem. More specifically, the main objective of this research is three-fold. It aims to a) thoroughly investigate the accuracy loss arising in the presence of streamwise flow gradients at the farfield boundaries when using flux enforcement or auxiliary state BC’s, b) provide simple and effective solutions to eliminate or greatly reduce such a loss, and c) provide effective benchmarks for the validation of the accuracy of turbomachinery farfield BC’s to be used by turbomachinery engineers and CFD developers. The higher accuracy of the new farfield BC’s we have developed so far with respect to that of more conventional implementations is based on 1) an improved use of 1D characteristic data, and 2) an extension of certain features of the scheme used for the interior face flux construction to the boundary faces.The figure on the right-hand-side reports the Mach number contours of a diverging turbomachinery annulus computed using the new farfield boundary conditions. The black contours refer to the computed result, whereas the analytical solution (available for this problem) is reported with red lines.
Industrial turbomachinery projects
Dr. Manosh Paul
The group has a strong connection with the turbomachinery industry. In a recently completed EPSRC-CASE research project with the Howden Group Technology, the aerodynamics efficiency of a two-stage centrifugal fan has been investigated.
The flow cavitation that causes a great deal of noise in weir pump and damages the impeller is investigated with the aid of CFD.
The group has recently investigated the efficiency of a hydraulic hammer/cannon designed and manufactured by Clansman Dynamics Ltd.
We are working with SMDH Ltd in Glasgow to investigate the performance of a new Compact Turbine.
Micro and Nano Thermofluids
Dr. Manosh Paul
In a recent joint project with Dr Phil Dobson, we are designing, manufacturing and investigating the performance of micro scale heat exchangers with the aid of novel computational fluid dynamics (CFD) and experimental techniques. Various design parameters of heat exchanger will be varied and their effects on the performance will be investigated under various important Thermofluids boundary and input conditions.
Moreover, in order to understand the principle of combustion in micro-devices and to determine the relevant factors affecting micro-combustion, several micro-combustors have recently been designed for powering micro-electromechanical system (MEMS) devices and their performances are being investigated. This is a collaborative project with the Dipartimento di Ingegneria e Fisica dell’ambiente, Universita’ della Basilicata, Italy.
Natural transition and boundary layer flows
Dr. Manosh Paul
The group has expertise in solving fundamental fluid mechanics problems relating to the boundary layer flow transition and separation. Current research includes the applications of in-house developed Direct Numerical Simulation (DNS) and Parallel Flow Solver (PFS) codes to the study of both linear and nonlinear mechanisms of stability and transition of free convection boundary layer flow over a plane surface, and this is carried out in a close collaboration with academics at the University of Bath.
Recently, the concept of the flow transition is applied to a new project that involves the design of an efficient natural convection heating system which would potentially run by solar energy.