Numerical Simulation for Industrial Applications on GARUDA
Transcription
Numerical Simulation for Industrial Applications on GARUDA
Numerical Simulation for Industrial Applications on GARUDA G. R. Shevare Emeritus Fellow, Dept. of Aerospace Engg., IIT Bombay Director, Zeus Numerix Pvt. Ltd. Contents Introduction Industrial vs. Scientific Simulations Ready for Solving Industrial Problems? Solvers - Requirements Meshing and post-processing Case Studies Closure Simulation presented here are the efforts of my colleagues in Zeus Numerix Introduction Context : Experience in developing and deploying simulation tools for industrial products and processes on GARUDA Why Industrial Simulations? For immediate societal needs as compared to quest for knowledge (scientific simulation) “What if scenarios” are easier in numerical simulations Concept to market can be an order of magnitude faster Numerical Simulations must be cheaper (and faster) than experiments Some design must be first time right – space exploration, nuclear reactor, stores separation, etc. Industrial vs. Scientific Simulations Industrial Simulations Scientific Simulations Purpose : to have useful, safe and cheaper products / processes Purpose: understand the nature, gain knowledge Overnight turn-around time Simulation time is less is preferred critical No of simulations can be in very large No of simulations can be handful Multi-disciplinary in nature These are rarely multidisciplinary Computational domain is complex and has to be extracted from CAD Scientist choose simple computational domain Simulations must be affordable Confirmation of “an idea” is the issue Ready for Solving Industrial Problems ? Develop enabling technologies Demonstrate that technologies work Put all technologies together Learning the use of integrated technologies Maturity of knowing when not to use Scientific papers, knowhow Limited success in pioneering applications Generic code useful for industry Extensive validation and convincing industry Industry willing to pay for numerical simulations We are here Two totally separate kinds of activities: Developing and demonstrated enabling technologies Convincing DRDO Labs, PSUs, DAE, CSIR labs and Pvt. Industry to use indigenous integrated HPC technologies Simulations is Interactive with designer in loop (1) CAD and analysis / design Requirements Simulation ready CAD, surface meshes Volume / field meshes (3) Use the data in design (fluid dynamics / heat transfer /structural integrity/ energy consumption / environmental effects) Solution of Equations describing the physics Extracting data for engineering analysis 1) Designer creates and owns CAD (virtual prototype of industrial product) 2a, 2b,2c ) Conduct virtual experiments in 3D space bounded by surface of the CAD model to simulate behavior of the product on HPC 3) Designer modifies his product Solvers Macroscopic simulation tools based on Newton's laws of motion Ampere’s laws, Faraday’s laws and Gauss Law – (Maxwell equations) Use Physics cast in Partial Differential Equations Numerical methods are typically FEM or FVM Often used virtual simulations are: Rigid body dynamics Stress analysis and structural integrity Fluid flow and heat transfer Chemical reactions and Combustion Electromagnetic fields Minimum Features required in a Solver Capable of using multiple spatial discretising schemes Capable of using various mathematical models; PDEs Should have multiple numerical schemes / elements Must permit realistic boundary conditions and initial conditions It should be possible to control data flow in simulation; each simulation is different Simulations should be scalable on ever changing computing hardware / software Should use standard file formats Developing tools is a continuous process… 1st Generation 2nd Generation 3rd Generation 4th Generation 5th Generation 2008 - 2010 2006 - 2008 2011 2010 - 2011 Pre-2006 0.2 million mesh points were the maximum mesh points we could handle 1 million mesh point on serial machine took 4 days 35 million mesh point on 42 CPU cluster took 20 days 2 million mesh point on serial machine took 1 day 35 million mesh point on 84 cores takes less than 14 hours 5th generation solver handles order of magnitude superior Life Cycle of a typical solver Accuracy Problem Size Duration & cost Experimental results vs. Simulation Results NASA Common Research Model Spatial resolution Tiny Coarse Medium Fine ExtraFIne No of cells in millions Comments Approx. 0.5 million Capture trends Approx. 2 million Approx. 5 million Approx. 15 million Approx. 40 million Reasonable results Usable data in design Results match with Expt. Data Scatter exptl. scatter Mesh Generation – divide the 3D space in simple shapes such as hexahedrons/ tetrahedrons Unstructured meshes for high fidelity simulations Futuristic octree meshes for fidelity simulations Post-processing – Extract useful information from large data sets Home grown post-processor a must This is the “other” interface with simulation engineer / designer Home grown post-processors are a must for extracting the engineering data from field data It is impossible to extract heat transfer data in hundreds of subchannels using commercial postprocessors Case # 1 : Simulation of Fire in Buildings Statement of the Problem: Predict temperature and movement of smoke in building under fire Fire Characteristics Fire Intensity Fire Location Fire Coordinates (x,y,z)** Fire Area Duration of Fire 10 MW Cafeteria Corridor (0,0,78.8) (-16.5,0,101.7) 5m x 5m 3m x 3m 5 MINUTES Ambient Conditions Ambient Wind Velocity Ambient Temperature Westerly (1.5 m/s at 10 m) 35 deg C Building Status Refuge Openings Sky Light Openings Atrium - Apartment Interface HVAC Operation 13 All full open Fully open Segregated; smoke proof wall/door NIL Simulation of Fire in Buildings 3.68 m/s 3.58 m/s 3.44 m/s 3.28 m/s 3.09 m/s 2.84 m/s 2.44 m/s 1.50 m/s The tallest building in Mumbai 14 Results with 10 MW fire High temperature dilutes the smoke in terms of CO, soot. Draft is produced due to buoyancy Lower refuge floors are safe. Upper refuge floors will attract smoke. Accumulation can be reduced by removing obstacles from skylight Moderate wind does not affect fire and smoke scenario much The removal of residual smoke from corridors needs jet fans On 24 Cores simulation requires one week Many buildings are simulated in Mumbai and Ratnagiri Case # 2: Airframe-engine intake integration Statement of Problem : Simulate pressure distribution at face of intake of an engine The air entering intake must satisfy the following conditions otherwise the engine stops in flight Distortions for the intakes must be less than 0.1 Pressure recovery should be more than 95 % Turbulence level should be less than 0.01 Maximum Planar wave peak-to-peak should be less than 0.01 Intake should not produce more drag The above conditions must be ensured for various altitude, speed and attitude of aircraft Results Possibly the first-ever “engine fouling” study for military aircraft Approx. 43 million cells Flow at the intake sampled every 1 millisecond. 2 secs. of flight requires 8 days of simulation Flight is cleared Total press pressure variation in the leeward side intake Total press pressure variation in the windward side intake Case # 3 : Heat Transfer in Nuclear Reactor Fast Breeder Reactors use liquid sodium to extract heat generated by nuclear fuel Lower heat extraction leads to meltdown Higher heat extraction leads to inefficiency Due to limitations on the tools and computing power full scale simulations are rare Extracting the useful design data from the simulations requires developing special processing tools Simulation (carried out for IGCAR ) was possible due to GARUDA- possibly a world record Flow of Liquid sodium in Fuel sub-assembly Pressure variation Results from literature Simulations on GARUDA Max of 17 pins only Complete assembly of 217 pins Estimated 60 GB for 100 millions cells Used 8 GB for 60 million cells Used Tet meshes (inferior results) Developed and deployed automated Hex meshes (better results) Case # 4 : Simulation of Noise Noise is unsteady pressure field of extremely small amplitude Flow noise is due to turbulence and interaction of shock with turbulence It has serious implications especially in high speed jets Direct simulation of noise suffer from the scalability and requires peta-scale computing resources Hybrid methods first obtain aerodynamic noise sources. The noise transmission is simulated separately Simulation pertains to calculation of noise around ARIANE launch vehicle Results for ARAINE vehicle 12 M cells One week of simulation for each flight Mach number Case # 5 : Filling of moulds Casting of metals and plastic parts should not produce blow holes or pores Location of risers and rate of filling dictates Quality of mould Productivity Loss of material Numerical Simulations involving PDEs satisfying conservation of mass, momentum and energy for solid, liquid and air can be carried out Heat transfer rate at interface between mould and the solid requires BC based on empirical information Each simulation can take days Case # 5 : Solidification of moulds Blue colour – air Red colour - liquid Initial Simulations - Filling of a jar with water (No solidification attempted yet) Case # 6 : Why model physics through PDEs? Some problems defy usage of PDEs as the mathematical modes Discrete particles (sph) methods is an alternative t = 4 microseconds t =100 microseconds Case # 7 : Stores Separation Studies Aircrafts and stores are designed & manufactured (or acquired) separately When used in a sortie, stores such as drop tanks and missiles are required to be dropped, ejected or launched from aircraft In dense atmosphere there is a possibility of stores colliding with the mother aircraft Wind tunnel tests can not be carried out. Controlled flight tests of separation is very risky and costly Numerical simulation is a cheaper, safer and better option Hundreds of simulations are required for safe separation of stores in the flight envelope Tools required for Stores Separation Studies Interpolate & transfer solution Start Initial volume data to new volume mesh mesh Generate new volume mesh Run Steady State CFD Simulation to Obtain Forces and Moments No Store collides with aircraft objects? Yes Add external forces like, ejector modelling and gravity. Change orientation of store surface mesh. Compute new position and orientation of store from 6-DOF module Yes More dynamics iterations ? No Stop Results Each t takes close to 2.5 hours. This includes: meshing, interpolation of old solution on newly generated mesh, finding out new solution, calculation of aerodynamic loads, finding out new position & orientation and calculation of missed distance 1 m of Stores Separation study requires – approx. 5 days on a HPC Two test flights may get cleared. Two in pipeline Case # 7 : Visibility of aircraft to RADAR Survivability of aircraft is as important as its lethality Survival of aircraft depends on its visibility to enemy RADAR Visibility is measured in terms of radar cross section (RCS) RCS depends on shape of aircraft, its orientation with respect to transmitting / receiving antennas RCS also depends on material of aircraft To reduce RCS, radar absorbing materials are required Estimation of RCS and characterization of materials requires solution Maxwell equations Results Metallic Object Incident Field : 7GHz, Excitation: Along the axis, Vertical polarisation CFDExpert-Lite for Education & SMEs Off-the-shelf CFD software are very costly for small and medium firms of pumps, heat exchangers, etc. CFDExpert-Lite to be launched as a GARUDA application next month 30 Acknowledgements GARUDA Computing Resources are being used engineering simulations IIT Bombay for encouraging “Numerical Simulation for Industrial Applications” DRDO, CSIR& DAE Labs, PSUs, Armed forces for Patronizing Indigenous code development and using the simulation results in their design and operations Numerical Simulations have arrived in Industrial Applications Thank You