Data-driven fluid modeling using physics-informed neural networks

compressible_flow/NICFD_nozzle/PhysicsInformed/0:generate_config.py

@@ -0,0 +1,44 @@
+###############################################################################################
+#       #      _____ __  _____      ____        __        __  ____                   #        #
+#       #     / ___// / / /__ \    / __ \____ _/ /_____ _/  |/  (_)___  ___  _____   #        #
+#       #     \__ \/ / / /__/ /   / / / / __ `/ __/ __ `/ /|_/ / / __ \/ _ \/ ___/   #        #
+#       #    ___/ / /_/ // __/   / /_/ / /_/ / /_/ /_/ / /  / / / / / /  __/ /       #        #
+#       #   /____/\____//____/  /_____/\__,_/\__/\__,_/_/  /_/_/_/ /_/\___/_/        #        #
+#       #                                                                            #        #
+###############################################################################################
+
+############################# FILE NAME: 0:generate_config.py #################################
+#=============================================================================================#
+# author: Evert Bunschoten                                                                    |
+#    :PhD Candidate ,                                                                         |
+#    :Flight Power and Propulsion                                                             |
+#    :TU Delft,                                                                               |
+#    :The Netherlands                                                                         |
+#                                                                                             |
+#                                                                                             |
+# Description:                                                                                |
+#  Generate configuration for defining a physics-informed neural network for modeling the     | 
+#  fluid properties of siloxane MM in NICFD with the data-driven fluid model in SU2.          |
+#                                                                                             |
+# Version: 2.0.0                                                                              |
+#                                                                                             |
+#=============================================================================================#
+
+from su2dataminer.config import Config_NICFD
+
+# The fluid data for siloxane MM are generated with the CoolProp module using the Helmoltz
+# equation of state.
+fluid_name = "MM"
+EoS_type = "HEOS"
+
+Config = Config_NICFD()
+Config.SetFluid(fluid_name)
+Config.SetEquationOfState(EoS_type)
+
+# Fluid data are generated on a density-energy grid rather than pressure-temperature.
+Config.UsePTGrid(False)
+
+# Display configuration settings and save config object.
+Config.SetConfigName("SU2DataMiner_MM")
+Config.PrintBanner()
+Config.SaveConfig()

compressible_flow/NICFD_nozzle/PhysicsInformed/1:generate_fluid_data.py

@@ -0,0 +1,40 @@
+###############################################################################################
+#       #      _____ __  _____      ____        __        __  ____                   #        #
+#       #     / ___// / / /__ \    / __ \____ _/ /_____ _/  |/  (_)___  ___  _____   #        #
+#       #     \__ \/ / / /__/ /   / / / / __ `/ __/ __ `/ /|_/ / / __ \/ _ \/ ___/   #        #
+#       #    ___/ / /_/ // __/   / /_/ / /_/ / /_/ /_/ / /  / / / / / /  __/ /       #        #
+#       #   /____/\____//____/  /_____/\__,_/\__/\__,_/_/  /_/_/_/ /_/\___/_/        #        #
+#       #                                                                            #        #
+###############################################################################################
+
+########################### FILE NAME: 1:generate_fluid_data.py ###############################
+#=============================================================================================#
+# author: Evert Bunschoten                                                                    |
+#    :PhD Candidate ,                                                                         |
+#    :Flight Power and Propulsion                                                             |
+#    :TU Delft,                                                                               |
+#    :The Netherlands                                                                         |
+#                                                                                             |
+#                                                                                             |
+# Description:                                                                                |
+#  Generate the single-phase fluid data for siloxane MM which is used to train the            |
+# physics-informed neural network.                                                            |
+#                                                                                             |
+# Version: 2.0.0                                                                              |
+#                                                                                             |
+#=============================================================================================#
+
+from su2dataminer.config import Config_NICFD
+from su2dataminer.generate_data import DataGenerator_CoolProp
+
+# Load SU2 DataMiner configuration.
+Config = Config_NICFD("SU2DataMiner_MM.cfg")
+
+# Initiate data generator.
+DG = DataGenerator_CoolProp(Config)
+DG.PreprocessData()
+DG.ComputeData()
+
+# Visualize and save fluid data.
+DG.VisualizeFluidData()
+DG.SaveData()

compressible_flow/NICFD_nozzle/PhysicsInformed/2:train_PINN.py

@@ -0,0 +1,53 @@
+###############################################################################################
+#       #      _____ __  _____      ____        __        __  ____                   #        #
+#       #     / ___// / / /__ \    / __ \____ _/ /_____ _/  |/  (_)___  ___  _____   #        #
+#       #     \__ \/ / / /__/ /   / / / / __ `/ __/ __ `/ /|_/ / / __ \/ _ \/ ___/   #        #
+#       #    ___/ / /_/ // __/   / /_/ / /_/ / /_/ /_/ / /  / / / / / /  __/ /       #        #
+#       #   /____/\____//____/  /_____/\__,_/\__/\__,_/_/  /_/_/_/ /_/\___/_/        #        #
+#       #                                                                            #        #
+###############################################################################################
+
+################################ FILE NAME: 2:train_PINN.py ###################################
+#=============================================================================================#
+# author: Evert Bunschoten                                                                    |
+#    :PhD Candidate ,                                                                         |
+#    :Flight Power and Propulsion                                                             |
+#    :TU Delft,                                                                               |
+#    :The Netherlands                                                                         |
+#                                                                                             |
+#                                                                                             |
+# Description:                                                                                |
+#  Initate physics-informed machine learning process for training the neural network used to  |
+#  model the fluid properties of siloxane MM in NICFD with the SU2 data-driven fluid model.   |
+#                                                                                             |
+# Version: 2.0.0                                                                              |
+#                                                                                             |
+#=============================================================================================#
+
+from su2dataminer.config import Config_NICFD
+from su2dataminer.manifold import TrainMLP_NICFD
+
+# Load SU2 DataMiner configuration.
+Config = Config_NICFD("SU2DataMiner_MM.cfg")
+
+# Set learning rate parameters and define network architecture.
+Eval = TrainMLP_NICFD(Config)
+
+# Initial learning rate: 10^-3, learning rate decay parameter: 9.8787, mini-batch size: 2^6.
+Eval.SetAlphaExpo(-3.0)
+Eval.SetLRDecay(+9.8787e-01)
+Eval.SetBatchExpo(6)
+
+# Network architecture: two hidden layers with 12 nodes.
+Eval.SetHiddenLayers([12,12])
+# Hidden layer activation function: exp(x)
+Eval.SetActivationFunction("exponential")
+# Display training progress in the terminal.
+Eval.SetVerbose(1)
+
+# Initiate training process.
+Eval.CommenceTraining()
+Eval.TrainPostprocessing()
+
+Config.UpdateMLPHyperParams(Eval)
+Config.SaveConfig()

compressible_flow/NICFD_nozzle/PhysicsInformed/3:run_SU2.py

@@ -0,0 +1,256 @@
+#!/usr/bin/env python
+
+## \file 3:run_SU2.py
+#  \brief NICFD simulation of supersonic expansion of siloxane MM.
+#  \version 8.1.0 "Harrier"
+#
+# SU2 Project Website: https://su2code.github.io
+#
+# The SU2 Project is maintained by the SU2 Foundation
+# (http://su2foundation.org)
+#
+# Copyright 2012-2024, SU2 Contributors (cf. AUTHORS.md)
+#
+# SU2 is free software; you can redistribute it and/or
+# modify it under the terms of the GNU Lesser General Public
+# License as published by the Free Software Foundation; either
+# version 2.1 of the License, or (at your option) any later version.
+#
+# SU2 is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+# Lesser General Public License for more details.
+#
+# You should have received a copy of the GNU Lesser General Public
+# License along with SU2. If not, see <http://www.gnu.org/licenses/>.
+
+import numpy as np
+import gmsh 
+import pysu2
+import CoolProp.CoolProp as CP
+from mpi4py import MPI
+from su2dataminer.config import Config_NICFD
+
+
+def WriteSU2Config(Config:Config_NICFD):
+    """
+    Write SU2 configuration options.
+    """
+
+    # Retrieve critical pressure and critical temperature from the fluid.
+    fluid = CP.AbstractState(Config.GetEquationOfState(),Config.GetFluid())
+    p_critical, T_critical = fluid.p_critical(), fluid.T_critical()
+
+    p_t_in = p_critical / 1.0716  # Inlet total pressure.
+    T_t_in = T_critical / 0.9919  # Inlet total temperature.
+    p_ratio = 10                # Total-to-static pressure ratio.
+    p_s_out = p_t_in / p_ratio  # Outlet static pressure.
+
+    val_viscosity=1e-5  # Constant viscosity value.
+    val_Prandtl=8.25e-1 # Laminar and turbulent Prandtl number.
+
+    # Write ASCII file with MLP weights and biases.
+    Config.WriteSU2MLP("MLP_siloxane_MM")
+
+    # SU2 config options for NICFD nozzle simulation.
+    su2_options = """               
+    SOLVER= RANS
+    KIND_TURB_MODEL= SA
+    SA_OPTIONS= NONE
+    MATH_PROBLEM= DIRECT
+    RESTART_SOL=NO
+    MACH_NUMBER= 0.05
+    AoA= 0.0
+
+    FREESTREAM_PRESSURE= __Pt_INLET__
+
+    FREESTREAM_TEMPERATURE= __Tt_INLET__
+
+    FREESTREAM_TURBULENCEINTENSITY = 0.1
+
+    FREESTREAM_TURB2LAMVISCRATIO = 100.0
+
+    FREESTREAM_OPTION= TEMPERATURE_FS
+
+    INIT_OPTION= TD_CONDITIONS
+
+    REF_DIMENSIONALIZATION= DIMENSIONAL
+
+    FLUID_MODEL= DATADRIVEN_FLUID
+    USE_PINN=YES
+    INTERPOLATION_METHOD = MLP
+    FILENAMES_INTERPOLATOR = (__PINN_FILENAME__)
+
+    VISCOSITY_MODEL= CONSTANT_VISCOSITY
+
+    MU_CONSTANT= __VAL_MU__
+
+    CONDUCTIVITY_MODEL= CONSTANT_PRANDTL 
+    PRANDTL_LAM=__VAL_PRANDTL__
+    TURBULENT_CONDUCTIVITY_MODEL=CONSTANT_PRANDTL_TURB
+    PRANDTL_TURB=__VAL_PRANDTL__
+
+    MARKER_HEATFLUX= (wall, 0.0)
+
+
+    MARKER_EULER=(symmetry)
+
+    MARKER_GILES= (inflow, TOTAL_CONDITIONS_PT, __Pt_INLET__, __Tt_INLET__, 1.0, 0.0, 0.0, 0.8, 0.8,\
+                outflow, STATIC_PRESSURE,__Ps_OUTLET__, 0.0, 0.0, 0.0, 0.0, 0.8, 0.8)
+    SPATIAL_FOURIER=NO
+    TURBOMACHINERY_KIND= AXIAL
+    TURBO_PERF_KIND=TURBINE
+    RAMP_OUTLET_PRESSURE=YES
+    RAMP_OUTLET_PRESSURE_COEFF= (__Pt_INLET__, 10, 200)
+    AVERAGE_PROCESS_KIND= AREA
+    MIXEDOUT_COEFF=(0.1, 1e-5, 15.0)
+    MARKER_TURBOMACHINERY= (inflow, outflow)
+
+    NUM_METHOD_GRAD= WEIGHTED_LEAST_SQUARES
+    CFL_NUMBER= 4
+    CFL_ADAPT= YES
+    CFL_ADAPT_PARAM= ( 0.99, 1.01, 1.0, 1000.0)
+
+    LINEAR_SOLVER= FGMRES
+    LINEAR_SOLVER_PREC= LU_SGS
+    LINEAR_SOLVER_ERROR= 1E-5
+    LINEAR_SOLVER_ITER= 10
+
+
+    CONV_NUM_METHOD_FLOW= ROE
+    MUSCL_FLOW= NO
+    SLOPE_LIMITER_FLOW= VENKATAKRISHNAN
+    VENKAT_LIMITER_COEFF=0.5
+
+    TIME_DISCRE_FLOW= EULER_IMPLICIT
+    CONV_NUM_METHOD_TURB= SCALAR_UPWIND
+    MUSCL_TURB = NO
+    SLOPE_LIMITER_TURB= VENKATAKRISHNAN
+    TIME_DISCRE_TURB= EULER_IMPLICIT
+    CFL_REDUCTION_TURB= 0.1
+
+    ITER=10000
+
+    CONV_RESIDUAL_MINVAL= -16
+    CONV_STARTITER= 10
+    CONV_CAUCHY_ELEMS= 100
+    CONV_CAUCHY_EPS= 1E-6
+
+    MESH_FILENAME= __MESH_FILENAME__
+
+    OUTPUT_WRT_FREQ= 20
+    OUTPUT_FILES=RESTART,PARAVIEW_MULTIBLOCK
+    SCREEN_OUTPUT= (INNER_ITER, RMS_RES)
+    HISTORY_OUTPUT= (INNER_ITER, WALL_TIME, RMS_RES, CFL_NUMBER)
+    VOLUME_OUTPUT= (SOLUTION, PRIMITIVE, RESIDUAL, MPI)
+    """
+
+    # Set fluid specific quantities in config options.
+    su2_options = su2_options.replace("__Pt_INLET__", "%.6e" % p_t_in)
+    su2_options = su2_options.replace("__Tt_INLET__", "%.6e" % T_t_in)
+    su2_options = su2_options.replace("__Ps_OUTLET__", "%.6e" % p_s_out)
+    su2_options = su2_options.replace("__VAL_MU__", "%.1e" % val_viscosity)
+    su2_options = su2_options.replace("__VAL_PRANDTL__", "%.2e" % val_Prandtl)
+    su2_options = su2_options.replace("__PINN_FILENAME__", "MLP_siloxane_MM.mlp")
+    return su2_options
+
+def GenerateMesh():
+    """
+    Generate ORCHID nozzle mesh using gmsh
+    """
+
+    mesh_filename="nozzle_mesh.su2"
+    t_bl = 5e-3             # Boundary layer refinement thickness.
+    mesh_size_max = 2e-4    # Maximum cell size.
+
+    # Load ORCHID nozzle curve geometry.
+    curve_filename = "nozzle_curve.csv"
+    X = np.loadtxt(curve_filename,delimiter=',',skiprows=1)
+    x_wall, y_wall = X[:, 0], X[:, 1]
+    order = x_wall.argsort()
+    x_wall = x_wall[order]
+    y_wall = y_wall[order]
+
+    x_max, x_min = max(x_wall), min(x_wall)
+    L = x_max - x_min 
+
+    bl_offset = L * t_bl 
+    Np_axial = int(L / mesh_size_max)
+
+    # Initialize meshing procedure.
+    gmsh.initialize()
+    gmsh.option.setNumber("Mesh.MeshSizeMax", mesh_size_max)
+    gmsh.option.setNumber("Mesh.SaveAll", 0)
+    gmsh.model.add("FLUID")
+    factory = gmsh.model.geo
+
+    mesher = gmsh.model.mesh
+    wall_pts = []
+    for x, y in zip(x_wall, y_wall):
+        wall_pts.append(factory.addPoint(x,y,0))
+    boundary_curve = factory.addSpline(wall_pts)
+
+    bl_pts = []
+    for x, y in zip(x_wall, y_wall):
+        bl_pts.append(factory.addPoint(x,y-bl_offset,0))
+    bl_curve = factory.addSpline(bl_pts)
+
+    sym_pts = [factory.addPoint(x_min, 0, 0), factory.addPoint(x_max, 0, 0)]
+    sym_line = factory.addLine(sym_pts[0], sym_pts[1])
+
+    bl_inlet_crv = factory.addLine(bl_pts[0],wall_pts[0])
+    bl_outlet_crv = factory.addLine(bl_pts[-1],wall_pts[-1])
+
+    freeflow_inlet_crv = factory.addLine(sym_pts[0], bl_pts[0])
+    freeflow_outlet_crv = factory.addLine(sym_pts[-1], bl_pts[-1])
+
+    curvloop_freeflow = factory.addCurveLoop([sym_line, freeflow_outlet_crv, -bl_curve, -freeflow_inlet_crv])
+    curvloop_bl = factory.addCurveLoop([bl_curve, bl_outlet_crv, -boundary_curve, -bl_inlet_crv])
+
+    freeflow_plane = factory.addPlaneSurface([curvloop_freeflow])
+    bl_plane = factory.addPlaneSurface([curvloop_bl])
+
+    inlet_tag = factory.addPhysicalGroup(1, [freeflow_inlet_crv, bl_inlet_crv], name="inflow")
+    outlet_tag = factory.addPhysicalGroup(1, [freeflow_outlet_crv, bl_outlet_crv], name="outflow")
+    sym_tag = factory.addPhysicalGroup(1, [sym_line],name="symmetry")
+    wall_tag = factory.addPhysicalGroup(1, [boundary_curve],name="wall")
+    fluid_tag = factory.addPhysicalGroup(2, [freeflow_plane,bl_plane],name="fluid")
+    factory.synchronize()
+
+    mesher.setTransfiniteCurve(tag=bl_inlet_crv, numNodes=15, coef=0.8)
+    mesher.setTransfiniteCurve(tag=bl_outlet_crv, numNodes=15, coef=0.8)
+    mesher.setTransfiniteCurve(tag=boundary_curve, numNodes=Np_axial, coef=1.0)
+    mesher.setTransfiniteCurve(tag=bl_curve, numNodes=Np_axial, coef=1.0)
+    mesher.setTransfiniteCurve(tag=sym_line, numNodes=Np_axial, coef=1.0)
+
+    mesher.setTransfiniteSurface(tag=bl_plane, cornerTags=[bl_pts[0],bl_pts[-1],wall_pts[-1],wall_pts[0]])
+    mesher.setRecombine(dim=2,tag=bl_plane)
+    gmsh.model.mesh.generate(2)
+    gmsh.write(mesh_filename)
+    gmsh.finalize()
+
+    return mesh_filename
+
+comm = MPI.COMM_WORLD
+rank = comm.Get_rank()
+
+if rank == 0:
+    # Load SU2 DataMiner configuration.
+    Config = Config_NICFD("SU2DataMiner_MM.cfg")
+
+    # Write SU2 configuration options.
+    su2_options = WriteSU2Config(Config)
+
+    # Generate computational mesh.
+    mesh_filename = GenerateMesh()
+    su2_options = su2_options.replace("__MESH_FILENAME__", mesh_filename)
+
+    # Write SU2 configuration file.
+    with open("config_NICFD_PINN.cfg", "w") as fid:
+        fid.write(su2_options)
+comm.Barrier()
+
+# Initialize NICFD simulation.
+driver = pysu2.CSinglezoneDriver("config_NICFD_PINN.cfg",1, comm)
+driver.StartSolver()
+driver.Finalize()