Status of Industrial CFD in India – A Case Study
India is witnessing unprecedented growth in the usage of CFD for industrial applications over the years. A large number of government programs in defense, nuclear and aerospace as well as main-stream civilian applications industries (such as auto, process, energy, heating & ventilation, and energy) sustain CFD services as a viable business. As a result, there are several private companies offering CFD services in India. However, the real impetus for code development has always been military and nuclear applications, as these sectors demand state of art techniques, scale and precision in simulations. They also fund partial development cost. Therefore a couple of companies not only offer services, but also offer a large variety of high quality simulation software applications. These companies are either incubated or/and mentored by premier educational institutes.
The case being elaborated in this document is that of Zeus Numerix Pvt. Ltd. incubated and mentored by Department of Aerospace Engineering Department, Indian Institute Technology Bombay.
Code development at Zeus Numerix
Zeus Numerix has developed and continuously updated its generic three dimensional convective- diffusion equation extensively used in continuum based fluid dynamics for compressible and incompressible fluid flows. A large number of numerical schemes with commonly used turbulence models have been developed, experimented and validated with data in literature and from the national laboratories. Creeping flows with extremely small Reynolds number, hypersonic flows for various gas models and flows with a large range of Nusselt number have been analyzed. However code validation in areas such as multi-phase flows and reactive flows has not taken off.
On the other flip side, the company has used its expertise in CFD code development in other areas such as acoustics, computational electromagnetic, dynamics, etc. Interestingly, it also focuses on fluid dynamic simulations using discrete methods.
The solvers are parallelised using MPI, OpenMP and CUDA programming. The company ensures data portability between solvers belonging to third party applications and its own multi-physics simulation codes. Though most of its solvers have their own pre and post processors, the solvers use ISO file formats for data portability.
Use of CFD in Industry
In India, the main stream industry seems to have accepted CFD tools as the de-facto method for flow analysis now. Several applications can be cited: ventilation / smoke studies in residential and commercial premises, steady and transient simulations in hydraulics, investigation of pumps, turbines and valves in mechanical engineering, flow uniformity studies in boiler, ESP in thermal power sector, etc. In such simulations pressure drop, or flow uniformity is the focus.
However, aerospace and nuclear are the forerunner in driving R&D in CFD. These sectors need a development such that there is a cross over between aerodynamics, dynamics, structures, propulsion, etc. Also, the focus shifts to skin friction, heat transfer coefficient, transition to turbulent boundary layer, aerodynamic derivatives related to moments, damping derivatives, hinge moments, etc. Real need in aerospace is that of unsteady simulations, accurate prediction of stall, conjugate heat transfer studies, combustion instability, etc. Fig. 1 shows unsteady flow past an aircraft under Gaussian gust, which requires three zones of unstructured meshes and coupling of several tools including 6DOF code, meshing tools, interpolation tools, post-processing tools, etc.
Fig. 1: Flow past an aircraft under a gust
CFD for turbo-machinery applications is one more challenging field. These internal flows pose their inherent problems right from the choice of equations, method of solution, choice of boundary conditions, initial conditions, and scalability, etc. Trajectory calculations based of aero-data obtained using CFD and coupled (CFD+6DOF) has become routine in aerospace industry in India. Several other design and analysis methodologies where CFD data is used to complex system can be cited.
The most important impact of CFD usage will be seen when coupled analysis of fluid-structure-inertia becomes a routine analysis tool. Already, a tool for direct transfer of CFD forces (pressure and skin friction) from CFD meshes to FEM meshes for structural analysis is commonly by the Company. It appears that control reversal with un-separated flow on control surface can be predicted with accuracy comparable to wind tunnel results. However, aero-elastic analysis such as flutter with even reduced modeling will take a couple of years. What aerospace community may demand will be dynamics with aero-elastic simulations when flow is separated, which is an order of magnitude complex and may take years, before such tools gain confidence with users and designers. Similar is the case with combustion simulation which is far fairly complex.
Lastly, CFD is only of the several tools that multi-disciplinary optimization studies require. Attempts for optimizing airfoil shape using gradient based and non-gradient based methods are gaining ground. But a fully automated MDA/MDO loop is far from penetrating Indian aerospace industry.
One the most important drawback of continuum based CFD is the requirements of meshes. In fact several papers indicate that man-hours spent in creating the meshes is responsible for higher cost of CFD compared to cost of wind tunnel testing
One of the several types of meshes must be used in continuum based CFD. For high fidelity aerospace applications, structured multi-block meshes are preferred in Zeus Numerix. This is almost de-facto mesh for hypersonic flows. The multi-block grid generation software developed by the company gets extensively used in many defense applications. The software has been customized for configurations, such as missiles, rockets, ram jets, etc. which enables good quality RANS meshes in minutes compared manual mesh generation which otherwise can take several weeks. See Fig. 2. Unstructured polyhedraon as cells are easier to produce, but prism layer generation and creating large meshes (> 100 millions) is a problem. The unstructured meshes can be clustered zonally like structured multi-block meshes. The generic literature on multi-scale multi-physics simulations recommends usage of octtree meshes. The current level of geometry / solution adaptability in octtree, in Zeus Numerix codes is limited 10 for realistic problems.
Fig. 2: Customised multi-block structured mesh generator makes helps creating faster meshes
There are several papers citing inadequacy of continuum based CFD. Flow with large Knudsen number, motion of colloidal solution, flows in micro-channels are some examples. Similarly, multi-scale, multi-physics is one more challenge. Amongst discrete methods frequently found in literature are: Direct Simulation Monte Carlo (DSMC) method, Lattice Boltzman methods(LBM), Molecular Dynamics (MD), Smoothed Particle Hydrodynamics (SPH) methods, etc.
Discrete CFD is becoming popular with CFD community world over, Commercial R&D groups in India is no exception. It appears that there two reasons: (a) these models inherently provide unsteady solutions, which is fluid flows are (b) the numerics in these algorithms is better suited for cheap and energy efficient multi-core graphics processors in platforms likely to be available as time passes. It is heartening to see that commercial groups are developing tools using LBM, DSMC and SPH based CFD applications for civilian and military real life applications.
The Fig. 3 shows simulation of wind over a city using GPUs, as simulated by Zeus Numerix. The emphasis of the simulation was to demonstrate scalability of LBM using CUDA libraries. As per the current practice, a very large number of Cartesian meshes are used. The discrete method currently finding applications is SPH. The method is being used in the analysis of collision of solids.
It is essential to note that collision of solids with velocities higher than 1000 m/s can be modeled as a hydrodynamics problem. The other common application is sloshing of fluids.
Fig. 3: Wind flow over a city using LBM solver
Post-processing, though not considered as a serious activity by CFD scientific community, is an essential and very important activity for CFD users and designers. MDO without post-processing tools is unthinkable. Serious CFD users are known to use one time generated CFD field data for years. The CFD group being discussed develops post-processing tools. Generating fields such as gradient of a scalar or divergence / curl of a vector can be automated / is available in several commercial post-processors. However, it is worth noting that aerod data as required by structural engineers takes several man-hours of a CFD senior engineer. Accumulation of lift and drag along length and span of aircraft, or extracting angle of attack of along radial direction of the blade in a a turbine, converting mixing plane data as initial condition for sliding plane simulation, calculation sources of noise from unsteady simulation, etc. requires an access to CFD code, generic post-processing tools and domain experience. Zeus Numerix therefore owns and continuously updates its post-processing software. It permits even tensor visualization, which some post-processors may not offer. The current version offers stereoscopic-tiled-display as one of the options.
CFD is getting used for solving main-stream industry problems as well as highly complex problems in demanding industry such as defense, nuclear and aerospace industries. There are organizations which boast of CFD product development and CFD services as their revenue generation activity. These organizations have state of art solvers not only in CFD, but also allied areas. The organizations also develop pre and post-processors, including customized versions for reducing turn-around-time for getting numerical solutions.