Multiple applications of Numerical Simulations and High Performance Computing in an Energy Company Philippe Ricoux TOTAL/DS - PDF

Please download to get full document.

View again

of 55
All materials on our website are shared by users. If you have any questions about copyright issues, please report us to resolve them. We are always happy to assist you.
Information Report
Category:

Screenplays & Play

Published:

Views: 4 | Pages: 55

Extension: PDF | Download: 0

Share
Related documents
Description
Multiple applications of Numerical Simulations and High Performance Computing in an Energy Company Philippe Ricoux TOTAL/DS ENERGY TRANSITION Energy demand in Mboe/d +45% +31% Non OECD
Transcript
Multiple applications of Numerical Simulations and High Performance Computing in an Energy Company Philippe Ricoux TOTAL/DS ENERGY TRANSITION Energy demand in Mboe/d +45% +31% Non OECD OECD Under Total s vision, energy demand will increase by 31% between 2010 and 2035 driven by non-oecd countries needs To secure enough energy supplies to respond to the needs of 8.6 billion people in 2035 (vs. 7 B. in 2012), the energy industry will need to focus on: Diversifying the energy supply Dedicate each energy to uses where it is most efficient : oil in transport and petrochemicals gas in power generation, heating Progressively reduce the carbon content of the energy mix Contribute to improve energy efficiency. Reliable energy at an acceptable price and compatible with environment is key for the development A new energy model will be a long process that will require technology, innovation and massive investments 2 SIGNIFICANT RESOURCES YET TO BE PRODUCED SOURCE: TOTAL (STR, 02/2012) Oil resources Gas resources ~3,500 Bboe ~2,500 Bboe 100 Tight oil Unconventional resources Years of production at current pace* Extra heavy oil Unrisked shale oil / tight oil Yet to find and increased recovery rate Unrisked res. shale & tight gas, CBM Years of production at current pace* Identified resources Already produced 690 * Based on 2011 production: oil 31 Gboe -- gas 21 Gboe 3 VERY LARGE NON CONVENTIONAL GAS RESOURCES STILL TO BE DEVELOPED Billions boe Non conventional gas resources Europe Amérique Latine CEI Latin America CEI Africa Afrique Moyen Middle Orient East Amérique North America du Nord Asie Asia 50 0 Coal bed methane* Tight gas* Shale gas* Potential constraints : geological structures, production costs, environmental, logistics Favorable areas : United States, Latin America, China Source s: Total, IEA, CERA 4 PROGRESSIVE CHANGE OF THE WORLD ENERGY MIX Mboe/d % 2% 10% 6% 2% 3% 10% 5% 25% 6% 3% 11% 6% 21% Solar, wind, others Hydro Biomass Nuclear Coal Fossil energies to represent 74% of energy supply in 2035 Gas to become the second-largest energy source before % Share of nuclear to remain constant % 24% 25% Gas Strong growth of new energies, notably solar and wind 32% 31% 28% Oil Renewable energies are promised to a bright future but substitution of fossil fuels will be slow 5 Gt CO 2 eq Gt CO 2 eq ENERGY RELATED CO 2 EMISSIONS Ref.: IEA WEO 2012 For an efficient climate scenario, energy efficiency/savings are key, but New Energies and Carbon Capture and Storage (CCS) are necessary 6 TOTAL: POOL OF ENERGIES, STRATEGIC CHOICES Heavy Oil production E.O.R. simulations thermal chemical polymers Storage polysaccharides Simple degradation Saccharose Dilute acid Fructose Glucose Starch Amylase Glucose Structural polysaccharides Celluloses Complex degradation CellulaseS C 6 Glucose Hemicelluloses HémicellulaseS C 5 Xylose Arabinose + Preparing deployment of CCS 7 A WORLD OF PROMISING ADVANCED TECHNOLOGIES Facing energy future: implications for R&D strategy IT Technologies Material sciences Biotechnologies Nanotechnologies New analytical techniques Advanced computing Major implications at present Solutions for the future 8 RADAR FOR IT TECHNOLOGIES MATURITY EMERGING ADOLESCENT FIRST ROLLOUT MAINSTREAM Advanced Robotics Geolocation Augmented Reality Data Analytics Drones Industrial Mobile Devices Natural Language Processing Wearable Computing Self service BI High Performance Computing Wireless Networks Search Engines Hadoop GIS HTML5 NOSQL RFID NFC Immersive video conferencing 2D Barcodes Media Analytics Data Visualization Virtual Reality 3D printing Semantic Web M2M Text and Speech Translation Biometrics Li-Fi Holography Voice Recognition Gesture Recognition Short term Mid term Long term Availability 9 SIMULATION/HPC: A TOOL FOR UNDERSTANDING, CONCEPTION AND INNOVATION Intensive Computing for Numerical Simulation : Necessary, Unavoidable Simulation and HPC for a better Understanding of major complex scientific problems: Earth System: Goelogy,Geomecanic, global changes (climate, ocean, ), natural risks, Physics: Particles, chemical activity, Astrophysics, Thermodynamics, Life Sciences: Pharmacy, Genome, Biomechanics Industrial challenges: Geosciences, Aeronautics, turbulent combustion, multi-fluid flows, new materials,, Simulation for Conception, Optimization, Innovation A tool for R&D and Engineering is in the service of processes Material Structure: Rheology, Fluid/Structure coupling, compounds, New Material Design: with more and more Molecular Simulation, nanomaterials, nanosystems Process Engineering: oil&gas, Automotive, Crash Test, Aeronautics, Benefits of Numerical Simulation : Better Understanding with a huge reduction of errors and risks Increase range of parameters variation (closer limits) with reduction of dangerous or expansive experiments Large «time saving» of development phases, before pilot Necessary way to go further: Work together Collaboration, Multi disciplinary teams: Share tools and algorithms, merge skill, Multi domains Team Building, workgroup : Maths, Computer Science, Applicative experts, Engineers, 10 TOTAL NUMERICAL SIMULATION AND HPC 11 PROCESSOR S EVOLUTION PARRALEL COMPUTING 12 MASSIVELY PARALLEL COMPUTING EX: IBM Blue Gene ( L or Q) 16 Nodes (left) 2* 440 cores /Architecture of core (right) 13 HPC (HIGH PERFORMANCE COMPUTING), DEFINITIONS AND TENDANCIES Floating Operation per sec (Flops) Capacity 1 «Normal» PC (Intel 4 Dual Core) ~ 3 GF (10 9 Flops) 1 TeraFlops = Flops (~300 Processors) 1 PetaFlops = Flops (~ Processors) 1 ExaFlops = Flops (several Millions Cores) 1 ZetaFlops = Flops (several Billions Cores???) CPU (Processors), GPU (Graphic cards, accelerators) 45nm Design des processeurs (Intel) happertown 32nm convergence CPUs et nehalem accélérateurs? westmere Knights Corner (accélérateur) 22nm sandybridge 16nm IvyBridge Technologie 3D 12nm 8nm 4nm Tendancies TOTAL: 3PF in World: PF en 2014 ; 100 PF en 2015 Rupture(s) : 1 EF en 2020, CPU, 100 cores/cpu, cores Electric Power: 1 MW /1PF in 2012 (5 MW in 2009), Expected : 1EF 20/40 MW At Exascale Vs Petascale: Capacity *1000 ; Memory *100 ; Memory/core *0.1 Impose to change all programming software Synchronization and communication avoiding Optimal Coupling Architecture / Algorithm / Application THE TOP500 LIST: TWENTY YEARS OF INSIGHT INTO HPC PERFORMANCE Nov YEARS 6 YEARS Oil & Gas Companies TOTAL 2.3 PF 2013 CPU 25 PF 2015/6 New Proc. 15 SOME SUPER COMPUTERS Titan, Oak Rigde, PF, CRAY, CPU/GPU Tianhe-2, China, PF, CPU (Ivy Intel) Pangea,TOTAL, 2.3 PF, SGI, CPU Sequoia, LLNL, PF, IBM, CPU 16 DEPTH IMAGING: AN PARALLEL ALGORITHM Wave Equation (hyperbolic) Approximation : Helmholtz Equation (elliptic) Billions unknown variables, Large solvers Link with computer science 17 FWI ALGORITHM General workflow for FWI, initial model: legacy velocities, well logs, and non-seismic measurements for velocity analysis. FWI is the best Approach today to determine reservoir properties. A data misfit results after several iterations, producing local and global minima depending on the starting models. One of the challenges with FWI using gradient or gradient-descent methods is the convergence to the local minima. Technique very sensitive to the starting velocity model, especially when 3-D is considered 18 SEISMIC DATA 1200m 1200m cquisition Geometries Conventional NAZ WAZ Explo 2 Larger Xline offset RAZ/Full WAZ Development Multi 7-9 Azimuth M$ 30-70M$ Wide Azimuth Rich M$ Azimuth Cost for ~200 km² Seismic and geological Data unconventional 19 HPC for Depth Imaging : 3 fundamental steps Numerical analyst Numerical Methods HPC Computing HPC implementation Cores + Options GPU Geo-physics Maths for Physic Modeling Studies Embarrassingly Parallel approximation 20 WEST AFRICA 21 HPC OPPORTUNITIES IN TOTAL: NEXT STEPS IN DEPTH IMAGING Combinaison of Physics, Numerics, Uncertainties (UQ) Involving maths modling for a more accurate approximation of the physics of propagation: More realistic: elastic, visco-elastic, poro-visco elastic Hybrid representations of waves equation Others physics: EM, micro gravimetric, More and more adapted numerics: Sub domains, automatic mesh generation Finite Elements,... explicit or implicit Massively parallel solvers, embedded solvers,. Performing approximations Uncertainties, Optimization Stochastic Methods thank to HPC. Robust optimization basis of inverse problem Computer Science Load Balancing Programming, Resilience, Challenge: Integrated Approach of Oil System : interaction geology geophysic : foot hills, non conventional reservoirs, Same Roadmap in BP, Chevron Absolute Need of multi skills Multidisciplinary teams 22 HPC & Reservoir Modeling From Pore to Darcy Needs of new and efficient reservoir simulations Heavy oil : combining Maxwell law + Darcy law, simulating SAGD (evolution of the steam chamber) Multi Fluids including polymers, MultiPhase Flows, Multi Physics, including geomechanic, Chemistry, E.O.R. mechanisms requiring thermo-hydraulic modeling, chemical reaction simulation CO2 project: simulation of storage (CO2 migration) and well integrity, predicting long term behavior Pore Network Modeling: Modeling mechanisms at pore scale Processing requirements could result in resources comparable to seismic imaging. Heavy Oil production Multi Scale, Different Physics at different scales Upscaling laws from nanometer to meter Essential for new fractured reservoirs, Shale Oil, Shale Gas Chess & Hytec Select main geochemical reactions at a local scale Lab experiment for calibration and matching GEM E.O.R. simulations thermal chemical polymers Activate main reactions for global reservoir/geochemistry simulation Matching performance needs by applying parallelism techniques at all levels 23 RESERVOIR SIMU. MULTI PHYSICS, MULTI SCALE Maximization of Oil reservoir production, Oil recovery (EOR) Multi-phase flow, Darcy's law modified with stochastic relative permeability: (stochastic PDE, macro law) for each phase of each component Where β indicates the phase, is the relative permeability (between 0 and 1) for the phase, and is the fluid pressure in the phase, which is the sum of the pressure in a reference phase (usually the gas phase) and the capillary pressure (capillary pressure is negative), and μ β the viscosity in the phase. Estimation of macro parameters such as relative permeability fundamental for reservoir simulation Many sources Many scales (10-5 to 10 8 cm) Sparse Not always reliable Observed data not at the same scale than models 24 RESERVOIR MODELLING: BASIS Transport Black Oil Model (o,g,w) Compositional Model Composition / chemistry / Thermodynamics Boundary conditions : Well, aquifers, bloc limits (kr=0) 25 RESERVOIR MODELLING: COMPLEXITY Reservoir: geostatistic fine representation of K (permeability tensor), Φ (flows vector) Stochastic PDE: Uncertainties Homogenization on a meshing : K, Φ constant by mesh kr(s), Pc(S) by lithology (estimation from transport at pore level) Anisotropy: Kxy/Kz from 1 to 10, K based on the principal slopes Coupled equations : Elliptic/Parabolic in Pressure and Hyperbolic/Parabolic non linear saturations/compositions Non linear Closure laws (kr, Pc, densities, viscosities, thermodynamic equilibrium) Coupled Resolution in P-S (c a ) : Local conservation equations for compressible Transport-thermodynamics coupling Implicit discretization in time Large Δt Locally conservative schema in space Stability of multi phase transport (approx) Local Explicit schema for flows (Φ) depending on variables of neighbor mesh of the ridge Cost on implicit way Best Extract of physics : Discrete conservation law Physical acceptable solution on «rough» meshing Homogenization degenerate in 26 RESERVOIR MODELING: PERSPECTIVES Darcy stochastic EDP Difficult ways to parallelization (few hundreds cores, 3000 in history matching) Future: Many cores (up to cores) application Refined meshing (close to wells) Adaptive Intelligent Mesh Improvement of Domain decomposition methods Time domain decomposition / Parallelization of time Hybrid numerical schema for fractured reservoir Discontinuous Galerkin Fine Discretization, CFL limit (IMPRES), Billons meshes New linear algebra: Factorization, Directed Acyclic Graphs (DAG) Communication Avoiding Algorithms Performing Embedded iterative solvers (Modified Newton) Disruptive new non linear new algebra, Qualitative Computing Physics coupling: Thermal, Thermodynamic, chemistry, transport.. Multi Scale methods Different time scale: split chemistry (saturation, low scale) and transport (large macro scale) Upscaling laws from pore network modeling 27 DENSE LINEAR ALGEBRA, PLASMA (CF. JACK DONGARRA) Large improvement on tasks schedule, reducing synchronization 28 29 IN SITU TECHNIQUES AND METHODS Data Reduction The transform by itself is reversible, and does not compress the data Subsample, Single precision or double precision, Direct scalar quantization, Adaptive scalar quantization, Vector quantization (VQ) or block quantization (Linde-Buzo-Gray (LBG) algorithm similar to the K-means method in data clustering) Transform-based compression: FT, Discrete Cosine Transform DCT, Wavelet Transform Feature Extraction Large-scale scientific simulations generate massive amounts of data that must be validated and analyzed for understanding and possibly new discovery. Quality Assessment Issues In Situ Visualization 30 SEA ICE MODELLING Sea Ice Floes Towards an accurate continuum dynamical model for sea ice Dynamics of an Assembly of Rigid Ice Floes Numerical simulation of ice performance of ships Forecast of gas dispatch schedule from Artic area 31 HPC FOR INDUSTRIAL MULTI SCALE PROBLEMS Petro Chemical Process 32 MULTI SCALE CFD : FCC RISER / MULTI SCALE - HPC Fluid Catalytic Cracking Micro (cata) Meso (cm 3?) Multi Scale in FCC Turbulence Macro (m) Experimental Snapshot of a the particle volume fraction field In the three-dimensional fluidized bed. Simulation Neptune Many cores Runs Scalability proven up to 4096 cores 3D Validation Pilot scale validation Validation in dilute area (TDH, transport disengagement height). Mesh up to 3 M cells on bubble / laminar / turbulent regime Mesh sensitivity Neptune optimum = cells/core But : imental Need much more cores to simulate 3D industrial scale Riser experimental Multi Scale need HPC 33 HPC FOR COMBUSTION Engine combustion - Aerospace, Gas Turbine, Helicopter Efficiency / Safety / Competitiveness - Automotive Efficiency / Environment / Life Cycle 34 Validation by experiments thanks to High Speed Imaging, Particle Image Velocimetry (HS PIV) HPC AND SAFETY: MULTI SCALE IN EXPLOSIONS Experiments Performed by Buncefield 2007 Understanding for safety Integration in on using codes of danger studies (large economical issue) Simulation LES Ref: Large Eddy Simulation of Vented Deflagration Quillatre P; Vermorel O; Poinsot T; Ricoux Ph Industrial & Engineering Chemistry Research, Feb MULTI SCALE IN EXPLOSIONS: LES FILTERING NS EQUATIONS Resolved Subgrid DNS LES RANS Resolved field Resolved Modeled Modeled information LES filter cutoff Navier-Stokes Equations for a compressible reactive flow: Only the terms below grid size are modelled Resolved Modelled 36 HPC & Numerical Simulation in Hutchinson Material Structure & Acoustics compounds «One of key technologies contributing to be a world Leader» Permitted Source : Hutchinson New Products Development Assistance - Performances forecast (static & dynamic stress, acoustic, ) - Length of life warranty (constraints, distortions,...) - Optimization Process Implementation Assistance : Injection, Extrusion.. - Equipment Conception (molds, tools,..) - Global Process Monitoring, optimization and Control : extrusion, injection, vulcanisation, pressing, 25 Mars NEW FEM approach : Iso geometry (IGA) : FEM vs IGA mesh Objective: New Efficient Mechanical Structure Simulation Method CADs (like IRIT) use NURBS (non-uniform rational B-splines) IGA use NURBS for the PDE solver CHALLENGE: treatment of nonmatching patch interfaces, regular gluing when possible. 38 PIPES NETWORK OPTIMIZATION (MINLP) 39 GAS GATHERING NETWORK OPTIMIZATION Data Results OPTIMIZATION Optimizes the costs (CAPEX + DRILLEX) Thousands of feasible solutions are explored The quality of the tool s solutions depends on the input data: (GIS, TEC/GEO, FP, ICS, EST) The pressure drop calculations are done online by Pipesim It quickly generates several alternatives 40 AHNET Solution is very close to the real one. The final result is introduced automatically in Pipesim to verify that it satisfies pressure drop constraints in all periods 41 Molecular Simulation: Multiscale Modeling Computational Tools: Density Functional Theory Monte Carlo & Molecular Dynamics Dynamic Density Functional Method. Finite Element Analysis 10-3 s 10-6 s Calculated properties: 10-9 s s MD, CPMD QM/MM Interaction of Segments, c QSAR MC, MD Coarse Grain kmc Thermo EP m 10-8 m 10-6 m 10-4 m Density of Segments 42 Mechanical Morphology Structure, Density, Diffusion Energetics, Spectroscopy,Chemre actions FIRST PRINCIPLE CALCULATIONS Kinetic H = E Potentiel H = T n + T e + V nn + V ne + V ee nuclear and electron and electron-electron, electron-nuclear, and nuclear-nuclear interaction H el = E el H el = T e + V ne + V ee Adiabatic Approximation ˆ V ee N i j 1 r ij V ee is the problem term Wavefunction methods: non-interacting reference systems - 0 Hartree (no exchange or correlation) Hartree-Fock (exact exchange) Post Hatree-Fock Density based approaches (DFT) 0 H E 0 KS non-interacting reference system - 0 In principle exact, but don t know form of E XC... 43 MOLECULAR MODELLING Simulate, understand, predict Two basic informations 1) interatomic forces at Temperature T vibrations (ps) at the adsorption site over a long time (ms) needed to «activate» the chemical bonds then a hop (diffusion step) (ps) 2) Electronic properties dipole moment, magnetism, NMR, excited states, spectrosocpy only first principle methods. Sometimes DFT not enough. MD not possible for the full process (including successive hops) 44 KINETIC MONTE CARLO (KMC) Kinetic Monte Carlo: coarse-grained hops Molecular Dynamics: Hole trajectory is calculated Rare-event dynamics on a grid List of all important processes Rate constants for each process Cu 2+, Cu +, Cu(0) Bulk diffusion Surface diffusion Adsorption Cu 2+ Cu + Cu(0) Suppressor (PEG) Bulk diffusion Surface diffusion Adsorption/desorption Blocking of adsorbed site Decreasing Cu 2+ Cu + rate in its proximity Accellerator (SPS) Bulk diffusion Surface diffusion Adsorption/desorption Increases the rate of Cu 2+ Cu + in its proximity Can remove an adsorbed PEG Can switch positions with Cu 2+ 45 KMC SIMULATIONS FOR FLAT COPPER SURFACE Additive free case Effect of Suppressor 46 COMPATIBILIZER SELECTION PP density No Compatibiliser Polypropylene (PP) + EPDM (Vistalon 8600) Strong segregation in both AFM: large domains (dark: EPDM) Simulation: Lamellar phase (finite size effects) 47 \ COMPATIBILIZER SELECTION: PP DENSITIES AFM images PP + EPDM + Vistamax 1% 5% 10% PP + EPDM + PPC 48 \ Compatibilizer concentration COMPATIBILIZER SELECTION: DENSITIES AT 10% COMPAT. AFM images PP + EPDM + Vistamax PP Compat EPDM PP + EPDM + PPC 49 \ 3 alloys components TOTAL NUMERICAL SIMULATION & HPC APPLICATIONS Oil & Gas (E&P) Seismic, Reservoir, Wells, Safety / Explosion Turbulence, Flame speed, detonation, Pipes, Risers, complex fluids transport Separation, Hydro cyclone (Oil Sands), FPSO, Molecular Simulation for Thermodynamics Refining Fluidized Bed Reactors : FCC, DHC, Combustion, Engine combustion, Hydro conversion of heavy hydrocarbons, Fischer-Tropsch Reactors Molecular Simulation for new lubricant & tribology Chemical Plants Slurry Loop, Polymerization, Swelling (PE) Multiphase Catalytic Reactors Molecular Simulation for Catalyst Specialties Compound Materials Deformation, Structure Calculations (Hutchinson) Meso Scale: RepresentativeVolume Element (RVE) Acoustics in com
Recommended
View more...
We Need Your Support
Thank you for visiting our website and your interest in our free products and services. We are nonprofit website to share and download documents. To the running of this website, we need your help to support us.

Thanks to everyone for your continued support.

No, Thanks