Methodologies and Robust Algorithms for Subsurface Simulators
Transcription
Methodologies and Robust Algorithms for Subsurface Simulators
Methodologies and Robust Algorithms for Subsurface Simulators April 27, 2016 Mary F. Wheeler Acknowledge • • • • • • • • Andro Mikelic (U. of Lyon) Thomas Wick (Ricam and Ecole Polytechnique) Ben Ganis Sanghyun Lee Baehyun Min Jing Ping Gurpreet Singh Ashwin Venkatraman CSM Research Areas § CSM website: http://csm.ices.utexas.edu 3/37 BIG DATA: Cross-cutting Initiative Exascale Computing 4/37 UT Strategic Partners Center for Subsurface Modeling Other Departments in UT Director: Mary F. Wheeler v Faculty (6) v Graduate Students (11) • • • • • • • • • • • • • • • • • Mary F. Wheeler Todd Arbogast Sanjay Srinivasan (Penn State U.) Ivan Yotov (U. of Pittsburgh) Andro Mikelic (U. of Lyon) Thomas Wick (Ricam) v Research Associates (2) • • Ben Ganis Gergina Pencheva Tameem Almani Yerlan Amanbek Mohammed Reza Beygi Saumik Dana Recheng Dong Mohammed Jammoul Tea Mikelic Morteza Naraghi Azor Nwachukwu Sogo Shiozawa Deandra White v Postdoctoral Fellows (5) • • • • • Sanghyun Lee Baehyun Min Jing Ping Gurpreet Singh Ashwin Venkatraman v Aerospace Engineering • • Laxminarayan Raja Mark Mear v Bureau of Economic Geology • • • Seyyed Hosseini Susan Hovorka Alexander Sun v Civil Engineering • Chadi El Mohtar v Geosciences • Mrinal Sen v Institute for Geophysics • v Staffs (2) • • Connie Baxter Amy Manley 5/34 Nick Hayman v Petroleum Engineering • • • Matthew Balhoff David DiCarlo David Nicholas Espinoza Collaborators S. Barbeiro H. Florez M. Parashar Z. Benjelloun-To uimi, I. Faille E. Gildin J. Killough T.F. Russell T. Dewers V. Girault S. Sun B. Dindoruk R. Helmig R. Tavakoli H. van Duijn K. Jordan M. Vohralík J. Eaton K. Kumar T. Wick P. Eiseman Y. Lee G. Wittum A. Elsheikh R. Liu I. Yotov A. Firoozabadi A. Mikelić D. Zhang 6/37 Research Topics Covered 1. Engineering and Geophysical Applications • Accurate physics-based simulation for realistic porous media models ü CO2 storage and EOR ü Polymer and surfactant flooding and foam ü Geomechanics: domain decomposition (DD), elasticity, plasticity, and hydraulic fracturing • History matching and reservoir characterization ü Parameter estimation ü Verification & Validation ü Uncertainty quantification 2. Mathematical and Computational Modeling • Formulation & implementation of efficient & accurate parallel multiscale & multi-physics algorithms based on ü Efficient solvers: multi-grid, multi-level multiscale preconditioners, non-linear acceleration ü Unstructured gridding and multipoint flux and mimetic methods ü Transport: DG, EG, CG post-processing ü Phase-field 7/37 Research Initiative: Closed-loop Workflow I. FLUID-DRIVEN FRACTURE PROPAGATION IV. UNCERTAINTY QUANTIFICATION II. PHASE & CHEMICAL EQUILIBRIUM III. SIMULATOR DEVELOPMENT 8/37 I. FLUID-DRIVEN FRACTURE PROPAGATION Objectives Develop a fracture propagation method (phase field) driven by multiphysics, multiphase fluid flow 3-dimensional computations using mesh adaptivity with coupling displacements, phase field, and pressure system In the phase field fracture propagation model, primal-dual active set & fixed-stress iteration is coupled to solve the whole system Demonstrate the potential of the phase field for treating practical engineering applications by providing numerical examples 9/37 Joining and branching of nonplanar hydraulic fractures in 3D heterogeneous media. Advantage of Phase Field Model § Classical theory of crack propagation [Griffith 1921] § Diffusive crack zones for free discontinuity problems § Γ-Convergent approximation [Ambrosio-Tortorelli 1992] § Variational methods based on energy minimization -Marigo 1998], [Miehe et al. 2010] Variation methods based energy minimization [Real fractures] [Interface approach] [Diffusive approach using Phase field] 10/37 [Francfort Advantage of Phase Field Model § Fixed-topology approach avoiding re-meshing § Determine crack nucleation, propagation, and the path automatically § Simple to handle joining and branching of (multiple) cracks § Promising findings for future ideas as an indicator function base d on theory and numerical simulations Before joining After joining 11/37 Branching Governing System: Biot’s System Biot’s system Fracture with maximum pressure • Pressure Diffraction System • The pressure starts to decrease when the fracture starts to propagate. • Mechanics and Phase Field • Linear Elasticity • Newton Iteration (a) Phase Field • Primal-dual Active Set Method 12/37 (b) Pressure Numerical Examples of Phase Field Multiple fractures propagating near the well bore • Fracture propagation • Pressure distribution 13/37 Numerical Examples of Phase Field 3 parallel fractures in 3D domain • Not all fractures are growing because of the stress-shadowed effect. Horizontal well Fractures 14/37 Numerical Examples of Phase Field Multiple fractures propagating near the well bore • Dynamic mesh adaptivity: predictor-corrector method [Heister-Wheeler-Wick, 2014] Heterogeneous Young’s Modulus Fracture Adaptive Mesh 15/37 Numerical Examples of Phase Field A fracture in layered media with different fracture toughness values Fracture = 1 Pa-m Injection well = 100 Pa-m 16/37 II. GEOCHEMICAL REACTIONS Objectives Model geochemical reactions of injected CO2 in carbonate reservoirs during EOR or CO2 sequestration In situ brines with reactive ionic species Gas Phase CO2, nC14, & H 2O Oleic Phase CO2, nC14, & H 2O Study the effect of reactive species on CO2 concentration Aqueous Phase Quantify the effect using changes in miscibilit y conditions during CO2 EOR CO2, CO32-, HCO3-, Ca2+, Cl-, CaCO3, H+, OH-, & H2O Solid Phase CaCO3 [ Phase & Chemical Equilibrium of CO2 ] 17/37 Coupling with a Compositional Simulator Coupled phase and chemical equilibrium • Gibbs Free Energy Minimization Gibbs free energy minimization Phase behavior Reactions Stochiometric approach • Equation of State (EOS) where rij : rate of change of component i in phase j due to chemical equilibrium (rij = 0 if no reaction) 18/37 Example: Effect of Reactions on CO2 Concentration Concentration Profiles at Pavg = 2,200 psi after 0.12 hours of CO2 injection • Red curve ( • Black curve ( ): concentration profiles with equilibrium reactions ): concentration profiles without equilibrium reactions • Reactions in aqueous phase consume CO2 altering phase equilibrium and hence miscibil ity conditions. • Higher in situ water saturation results in lower CO2 concentrations. CO2 w/ reaction CO2 w/o reaction C14 w/ reaction C14 w/o reaction CO2 w/ reaction CO2 w/o reaction C14 w/ reaction C14 w/o reaction Sw = 0.3 Sw = 0.5 19/37 III. SIMULATOR DEVELOPMENT Objectives Complete simulator development with numerical schemes f or coupled processes Develop computational methods for coupled p rocesses based on multiscale discretization fo r flow, geomechanics & geochemistry Development of efficient solvers & pre-conditioners Model CO2 storage field sites & perform compositional simulations Cranfield site, Mississippi 20/37 Framework of IPARS § IPARS (Integrated Parallel Accurate Reservoir Simulator) Development of integrated flow, geochemistry, and geomechanics framework 21/37 Upscaling: Bridging Lab to Field Upscale measured rock properties (fluid flow & geomechanics) Objectives o scale relevant to field processes t Development of homogenization schemes c ombining numerical and analytical approache s, e.g. multiscale mortar method Particular emphasis will be put on including n atural fractures in effective properties and lo calization effects Obtain field scale constitutive parameters to p erform coupled fluid flow and geomechanic al numerical simulation 22/37 ⎧σ = C ε − αpc I ⎪ ⎨ ⎪⎩ϕ = pc / N + αε V Model Field Sites Objectives Modeling, simulation & uncertainty analysis with application to CO2 storage sites (Cranfield, MS & Frio, TX) Measure mechanical properties in laboratory Site 1: Cranfield, MS Collect other existing data (seismic, well logs, etc.) Measure impact of geochemical alteration on mechanical properties Study rock dissolution and its effect on weake ning the rocks and creating leakage pathways Enhanced simulation for studying and quantif ying parameters, e.g. reservoir over pressure, chemical and thermal loading 23/37 Site 2: Frio, TX Poro-plasticitiy § Geomechanical Effects of CO2 Injection with a Poro-plasticity Model @(⇢( Fluid Flow 1 0 + ↵"v + M (p @t Stress Equilibrium @(⇢( 0 + ↵"v + M1 (p @t 00 Hooke’s Law ✓ p0))) K (rp ⇢grh) q=0 µ r · ( 00 + o ↵(p p0)I) + f = 0 +r· ⇢ ◆ K + r · ⇢ (rp ⇢grh) q=0 De : (" µ"p) 00 1 r · ( + o ↵(p p0)I) + f = 0 = 00 "˙ = Plastic Strain Evolution 2 375581 [psi] 0.25 Druker-Prager Yield Surface (ru + rT u) = De : (" "p) @F (" =00) 1 (ru + rT u)00 ,2 at Y ( ) @ 00 00 @F ( )00 "˙p = "0, ˙p = at Y (00 , ) < at Y0 ( Yield and Flow Functions E ⌫ ✓ p0))) " = Strain-Displacement Relation p ◆ @ p "˙ = 0, at Y ( 00 )<0 Y = q + ✓ m ⌧0 Y = q+✓ m 0 F =F q=+q + m m ⌧00 24/37 00 =0 )=0 Domain Decomposition for Geomechanics Pay-zone / Nonpay-zone Advantages of fixed stress splitting Schematic of 3D domain • Mortar finite elements and interface iterati ons are no longer required, but still possible to use. Reservoir • We have a new parallel poro-plasticity model that may be coupled with elastic a nd poroelastic subdomains. 2D cross-section !e,e • Savings with multi-rate methods have b Ωe s. We foresee it would be even greater in np,e !N Ωp poroelastic-elastic problems. !D 25/37 h H een demonstrated on poroelastic problem !p,e !p,p nΩ IV. UNCERTAINTY QUANTIFICATION § Calibration process of rock and fluid properties in subsurface models A Priori Model History Matching (permeability, md) (permeability, md) Multi-modal A Posteriori Model Gaussian 26/37 Multi-modal Carbon Capture & Storage Projects Worldwide Snohvit Sleipner K-12B RECOPOL CO2 SINK Hokkaido Nagaoka Sibilla In Salah Qinshui Basin Abu Dhabi 1st CO2 EOR in Middle East Teapot Dome Rangely Burlington Zama Penn West Alberta ECBM Weyburn Mountaineer West Pearl Queen Cranfield Frio Gorgon Otway Basin Cerro Fortunoso Coal Bed Depleted Oil Field ECBM Projects EOR Projects Gas Production Fields Saline Aquifer 27/37 Parameterization & Data Assimilation § History matching coupled w/ level-set parameterization, MFDFrac, and EnKF 1 Initialization 2 Level-Set Parameterization • • Generate initial fractured realizations Experiments Mesh generation Realization Convert non-Gaussian to Gaussian parameters • Φ: level set at the node • r: fracture length • θ: fracture orientation 3 Simulation using MFDFrac 4 Inverse Modeling using EnKF • • Mimetic Difference Approach Internal Fracture Boundaries Intersecting Fractures Flow EnKF for updating Gaussian parameters Ensemble mean of initial fracture realizations 28/37 Ensemble mean of final fracture realizations Parameterization & Data Assimilation Experimental setup Numerical model • Collaboration w/ Dr. N. Espinoza • Injector: 5 • Producers: 2, 3, 8, 9 29/37 Parameterization & Data Assimilation Numerical results • Blue straight lines indicate major fractures that mostly affect production rate. 30/37 Parallel Multi-objective Optimization for CCS at Cranfield Reservoir Characterization & Optimization Simulator IPARS • Global-objective evolution ES .exe strategy Stop • Serial • Multi-objective algorithm NSGA-II .exe Start • Multi-objective Builder .exe evolution strategy • PC For i=1:N Evaluate model genetic … Evaluate model algorithm • Supercomputer Start Evaluate model GA .exe Storage • Parallel • Global-objective genetic Run Evaluate model Algorithm Evaluate model OS ε-MOES .exe Stop 31/37 Simulation of Pulse Test for CO2 Leakage Detection Permeability distribution in Cranfield, MS • 661,760 = 20x188x176 grid cells Permeability (md) • Grid size: 4 ft x 50 ft x 50 ft DAS (drilled area of study) 80 ft x z y 32/37 Simulation of Pulse Test for CO2 Leakage Detection Bottomhole pressure match at the monitoring well Amplitude • Reproduction of both amplitude and wavelength of pulse patterns Wavelength 33/37 V. ONGOING WORKS § Modeling of proppant-filled fractures Proppant-filled fractures in a poroelastic medium [L.-Mikelic-Wheeler-Wick 2016] Fracture Proppant Proppant in a fracture • • • Transport system is coupled Enriched Galerkin (locally conservative) Non-Newtonian flow (power-law) Proppant 34/37 V. ONGOING WORKS § Modeling of viscous fingering Experimental results [S. Malhotra – E.R. Lehman – M.M. Sharma 2014] Simulation results 35/37 V. ONGOING WORKS § Stochastic fracture propagation Coupling with probability map [L.-Wheeler-Wick-Srinivasan] • InSAR (surface deformation map) • Obtain probability map • Initialize hydraulic fractures • Interactions with natural fractures 36/37 Concluding Remarks Developing a coupled flow-geomechanics-geochemistry simulator Capturing the coupled effect of non-ideal aqueous phase equilibrium and reactions with traditional EOS calculations Homogenization / Poro-plasticity / Fracture Propagation / Model calibration / Multi-rate coupling / Domain Decomposition Synergy effects from combination of modules 37/37 Acknowledgements Thank you for your attention Contact: mfw@ices.utexas.edu 38/37