process.ExecuteFinalizeSolutionStep()
#TODO: decide if it shall be done only when output is processed or not
for process in list_of_processes:
process.ExecuteBeforeOutputStep()
if (output_time <= out):
#TODO: following lines shall not be needed once the gid_io is adapted to using the parameters
nodal_results = []
for i in range(output_settings["nodal_results"].size()):
nodal_results.append(
output_settings["nodal_results"][i].GetString())
gauss_points_results = []
for i in range(output_settings["gauss_points_results"].size()):
gauss_points_results.append(
output_settings["gauss_points_results"][i].GetString())
gid_io.write_results(time, fluid_model_part, nodal_results,
gauss_points_results)
out = 0
for process in list_of_processes:
process.ExecuteAfterOutputStep()
out = out + Dt
gid_io.finalize_results()
for process in list_of_processes:
process.ExecuteFinalize()
class VertexMorphingMethod:
# --------------------------------------------------------------------------
def __init__(self,opt_model_part,config,analyzer):
# For GID output
self.gid_io = GiDOutput(config.design_surface_name,
config.VolumeOutput,
config.GiDPostMode,
config.GiDMultiFileFlag,
config.GiDWriteMeshFlag,
config.GiDWriteConditionsFlag)
# Add variables needed for shape optimization
opt_model_part.AddNodalSolutionStepVariable(NORMAL)
opt_model_part.AddNodalSolutionStepVariable(NORMALIZED_SURFACE_NORMAL)
opt_model_part.AddNodalSolutionStepVariable(OBJECTIVE_SENSITIVITY)
opt_model_part.AddNodalSolutionStepVariable(OBJECTIVE_SURFACE_SENSITIVITY)
opt_model_part.AddNodalSolutionStepVariable(MAPPED_OBJECTIVE_SENSITIVITY)
opt_model_part.AddNodalSolutionStepVariable(CONSTRAINT_SENSITIVITY)
opt_model_part.AddNodalSolutionStepVariable(CONSTRAINT_SURFACE_SENSITIVITY)
opt_model_part.AddNodalSolutionStepVariable(MAPPED_CONSTRAINT_SENSITIVITY)
opt_model_part.AddNodalSolutionStepVariable(DESIGN_UPDATE)
opt_model_part.AddNodalSolutionStepVariable(DESIGN_CHANGE_ABSOLUTE)
opt_model_part.AddNodalSolutionStepVariable(SEARCH_DIRECTION)
opt_model_part.AddNodalSolutionStepVariable(SHAPE_UPDATE)
opt_model_part.AddNodalSolutionStepVariable(SHAPE_CHANGE_ABSOLUTE)
opt_model_part.AddNodalSolutionStepVariable(IS_ON_BOUNDARY)
opt_model_part.AddNodalSolutionStepVariable(BOUNDARY_PLANE)
opt_model_part.AddNodalSolutionStepVariable(SHAPE_UPDATES_DEACTIVATED)
opt_model_part.AddNodalSolutionStepVariable(SENSITIVITIES_DEACTIVATED)
# Read and print model part (design surface)
buffer_size = 1
model_part_io = ModelPartIO(config.design_surface_name)
model_part_io.ReadModelPart(opt_model_part)
opt_model_part.SetBufferSize(buffer_size)
opt_model_part.ProcessInfo.SetValue(DOMAIN_SIZE,config.domain_size)
print("\nThe following design surface was defined:\n\n",opt_model_part)
# Set configurations
self.config = config
# Set analyzer
self.analyzer = analyzer
# Set response functions
self.objectives = config.objectives
self.constraints = config.constraints
print("> The following objectives are defined:\n")
for func_id in config.objectives:
print(func_id,":",config.objectives[func_id],"\n")
print("> The following constraints are defined:\n")
for func_id in config.constraints:
print(func_id,":",config.constraints[func_id],"\n")
# Create controller object
self.controller = Controller( config );
# Model parameters
self.opt_model_part = opt_model_part
# Create mapper to map between geometry and design space
self.mapper = VertexMorphingMapper( self.opt_model_part,
self.config.filter_function,
self.config.use_mesh_preserving_filter_matrix,
self.config.filter_size,
self.config.perform_edge_damping,
self.config.damped_edges )
# Toolbox to perform optimization
self.opt_utils = OptimizationUtilities( self.opt_model_part,
self.objectives,
self.constraints,
self.config.step_size,
self.config.normalize_search_direction )
# Toolbox to pre & post process geometry data
self.geom_utils = GeometryUtilities( self.opt_model_part )
# --------------------------------------------------------------------------
def optimize(self):
print("\n> ==============================================================================================================")
print("> Starting optimization using the following algorithm: ",self.config.optimization_algorithm)
print("> ==============================================================================================================\n")
# Print time stamp
print(time.ctime())
# Start timer and assign to object such that total time of opt may be measured at each step
self.opt_start_time = time.time()
# Initialize design output in GID Format and print baseline design
self.gid_io.initialize_results(self.opt_model_part)
self.gid_io.write_results(0, self.opt_model_part, self.config.nodal_results, [])
# Call for for specified optimization algorithm
if(self.config.optimization_algorithm == "steepest_descent"):
self.start_steepest_descent()
elif(self.config.optimization_algorithm == "augmented_lagrange"):
self.start_augmented_lagrange()
elif(self.config.optimization_algorithm == "penalized_projection"):
self.start_penalized_projection()
else:
sys.exit("Specified optimization_algorithm not implemented!")
# Finalize design output in GID formad
self.gid_io.finalize_results()
# Stop timer
opt_end_time = time.time()
print("\n> ==============================================================================================================")
print("> Finished optimization in ",round(opt_end_time - self.opt_start_time,2)," s!")
print("> ==============================================================================================================\n")
# --------------------------------------------------------------------------
def start_steepest_descent(self):
# Flags to trigger proper function calls
constraints_given = False
# Get Id of objective
only_F_id = None
for F_id in self.objectives:
only_F_id = F_id
break
# Initialize file where design evolution is recorded
with open(self.config.design_history_directory+"/"+self.config.design_history_file, 'w') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append("itr\t")
row.append("\tf\t")
row.append("\tdf_absolute[%]\t")
row.append("\tdf_relative[%]\t")
row.append("\tstep_size[-]\t")
row.append("\tt_iteration[s]\t")
row.append("\tt_total[s]")
historyWriter.writerow(row)
# Define initial design (initial design corresponds to a zero shape update)
# Note that we assume an incremental design variable in vertex morphing
X = {}
for node in self.opt_model_part.Nodes:
X[node.Id] = [0.,0.,0.]
# Miscellaneous working variables for data management
initial_f = 0.0
previous_f = 0.0
# Start optimization loop
for opt_itr in range(1,self.config.max_opt_iterations+1):
# Some output
print("\n>===================================================================")
print("> ",time.ctime(),": Starting optimization iteration ",opt_itr)
print(">===================================================================\n")
# Start measuring time needed for current optimization step
start_time = time.time()
# Set controller to evaluate objective
self.controller.initialize_controls()
self.controller.get_controls()[only_F_id]["calc_value"] = 1
# Set to evaluate objective gradient
self.controller.get_controls()[only_F_id]["calc_gradient"] = 1
# Initialize response container
response = self.controller.create_response_container()
# Start analyzer according to specified controls
iterator = str(opt_itr) + ".0"
self.analyzer( X, self.controller.get_controls(), iterator, response )
# Store gradients on the nodes of the model_part
self.store_grads_on_nodes( response[only_F_id]["gradient"] )
# Compute unit surface normals at each node of current design
self.geom_utils.compute_unit_surface_normals()
# Project received gradients on unit surface normal at each node to obtain normal gradients
self.geom_utils.project_grad_on_unit_surface_normal( constraints_given )
# Compute mapping matrix
self.mapper.compute_mapping_matrix()
# Map sensitivities to design space
self.mapper.map_sensitivities_to_design_space( constraints_given )
# Compute search direction
self.opt_utils.compute_search_direction_steepest_descent()
# # Adjustment of step size
# if( opt_itr > 1 and response[only_F_id]["value"]>previous_f):
# self.config.step_size = self.config.step_size/2
# Compute design variable update (do one step in the optimization algorithm)
self.opt_utils.compute_design_update()
# Map design update to geometry space
self.mapper.map_design_update_to_geometry_space()
# Compute and output some measures to track changes in the objective function
delta_f_absolute = 0.0
delta_f_relative = 0.0
print("\n> Current value of objective function = ",response[only_F_id]["value"])
if(opt_itr>1):
delta_f_absolute = 100*(response[only_F_id]["value"]-initial_f)/initial_f
delta_f_relative = 100*(response[only_F_id]["value"]-previous_f)/initial_f
print("> Absolut change of objective function = ",round(delta_f_absolute,6)," [%]")
print("> Relative change of objective function = ",round(delta_f_relative,6)," [%]")
# Take time needed for current optimization step
end_time = time.time()
time_current_step = round(end_time - start_time,2)
time_optimization = round(end_time - self.opt_start_time,2)
print("\n> Time needed for current optimization step = ",time_current_step,"s")
print("> Time needed for total optimization so far = ",time_optimization,"s")
# Write design in GID format
self.gid_io.write_results(opt_itr, self.opt_model_part, self.config.nodal_results, [])
# Write design history to file
with open(self.config.design_history_directory+"/"+self.config.design_history_file, 'a') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append(str(opt_itr)+"\t")
row.append("\t"+str("%.12f"%(response[only_F_id]["value"]))+"\t")
row.append("\t"+str("%.2f"%(delta_f_absolute))+"\t")
row.append("\t"+str("%.6f"%(delta_f_relative))+"\t")
row.append("\t"+str(self.config.step_size)+"\t")
row.append("\t"+str("%.1f"%(time_current_step))+"\t")
row.append("\t"+str("%.1f"%(time_optimization)))
historyWriter.writerow(row)
# Check convergence
if(opt_itr>1):
# Check if maximum iterations were reached
if(opt_itr==self.config.max_opt_iterations):
print("\n> Maximal iterations of optimization problem reached!")
break
# Check for relative tolerance
if(abs(delta_f_relative)<self.config.relative_tolerance_objective):
print("\n> Optimization problem converged within a relative objective tolerance of ",self.config.relative_tolerance_objective,"%.")
break
# Check if value of objective increases
if(response[only_F_id]["value"]>previous_f):
print("\n> Value of objective function increased!")
break
# Update design
X = self.get_design()
# Store values of for next iteration
previous_f = response[only_F_id]["value"]
# Store initial objective value
if(opt_itr==1):
initial_f = response[only_F_id]["value"]
# --------------------------------------------------------------------------
def start_augmented_lagrange(self):
# README!!!
# Note that the current implementation assumes that only one scalar objective & constraint is present
# Flags to trigger proper function calls
constraints_given = True
# Get Id of objective
only_F_id = None
for F_id in self.objectives:
only_F_id = F_id
break
# Get Id of constraint
only_C_id = None
for C_id in self.constraints:
only_C_id = C_id
break
# Initialize the optimization algorithm
self.vm_utils.initialize_augmented_lagrange( self.constraints,
self.config.penalty_fac_0,
self.config.gamma,
self.config.penalty_fac_max,
self.config.lambda_0 )
# Initialize file where design evolution is recorded
with open(self.config.design_history_directory+"/"+self.config.design_history_file, 'w') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append("itr\t")
row.append("\tsub_itr\t")
row.append("\tl\t")
row.append("\tdl_relative[%]\t")
row.append("\tf\t")
row.append("\tdf_absolute[%]\t")
row.append("\tpenalty_fac\t")
row.append("\tC["+str(only_C_id)+"]:"+str(self.constraints[only_C_id]["type"])+"\t")
row.append("\tlambda["+str(only_C_id)+"]")
historyWriter.writerow(row)
# Define initial design (initial design corresponds to a zero shape update)
# Note that we assume an incremental design variable in vertex morphing
X = {}
for node in self.opt_model_part.Nodes:
X[node.Id] = [0.,0.,0.]
# Miscellaneous working variables for data management
initial_f = 0.0
initial_l = 0.0
previous_l = 0.0
# Start primary optimization loop
for opt_itr in range(1,self.config.max_opt_iterations+1):
# Some output
print("\n>===================================================================")
print("> ",time.ctime(),": Starting optimization iteration ",opt_itr)
print(">===================================================================\n")
# Start measuring time needed for current optimization step
start_time = time.time()
# Solve optimization subproblem
for sub_opt_itr in range(1,self.config.max_sub_opt_iterations+1):
# Some output
print("\n>===============================================")
print("> Starting suboptimization iteration ",sub_opt_itr)
print(">===============================================\n")
# Start measuring time needed for current suboptimization step
subopt_start_time = time.time()
# Set controller to evaluate objective and constraint
self.controller.initialize_controls()
self.controller.get_controls()[only_F_id]["calc_value"] = 1
self.controller.get_controls()[only_C_id]["calc_value"] = 1
# Set controller to evaluate objective and constraint gradient
self.controller.get_controls()[only_F_id]["calc_gradient"] = 1
self.controller.get_controls()[only_C_id]["calc_gradient"] = 1
# Initialize response container
response = self.controller.create_response_container()
# Start analyzer according to specified controls
iterator = str(opt_itr) + "." + str(sub_opt_itr)
self.analyzer( X, self.controller.get_controls(), iterator, response )
# Evaluate Lagrange function
l = self.opt_utils.get_value_of_augmented_lagrangian( only_F_id, self.constraints, response )
# Store gradients on the nodes of the model_part
self.store_grads_on_nodes( response[only_F_id]["gradient"], response[only_C_id]["gradient"] )
# Compute unit surface normals at each node of current design
self.geom_utils.compute_unit_surface_normals()
# Project received gradients on unit surface normal at each node to obtain normal gradients
self.geom_utils.project_grad_on_unit_surface_normal( constraints_given )
# Compute mapping matrix
self.mapper.compute_mapping_matrix()
# Map sensitivities to design space
self.mapper.map_sensitivities_to_design_space( constraints_given )
# Compute search direction
self.opt_utils.compute_search_direction_augmented_lagrange( self.constraints, response )
# Compute design variable update (do one step in the optimization algorithm)
self.opt_utils.compute_design_update()
# Map design update to geometry space
self.mapper.map_design_update_to_geometry_space()
# Compute and output some measures to track changes in the objective function
delta_f_absolute = 0.0
delta_l_relative = 0.0
print("\n> Current value of Lagrange function = ",round(l,12))
if(sub_opt_itr>1):
delta_f_absolute = 100*(response[only_F_id]["value"]-initial_f)/initial_f
delta_l_relative = 100*(l-previous_l)/initial_l
print("\n> Relative change of Lagrange function = ",round(delta_l_relative,6)," [%]")
# We write every major and every suboptimization iteration in design history
with open(self.config.design_history_directory+"/"+self.config.design_history_file, 'a') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append(str(opt_itr)+"\t")
row.append("\t"+str(sub_opt_itr)+"\t")
row.append("\t"+str("%.12f"%(l))+"\t")
row.append("\t"+str("%.6f"%(delta_l_relative))+"\t")
row.append("\t"+str("%.12f"%(response[only_F_id]["value"]))+"\t")
row.append("\t"+str("%.2f"%(delta_f_absolute))+"\t")
row.append("\t"+str("%.2f"%(self.vm_utils.get_penalty_fac()))+"\t")
row.append("\t"+str("%.12f"%(response[only_C_id]["value"]))+"\t")
row.append("\t"+str("%.12f"%(self.vm_utils.get_lambda(only_C_id))))
historyWriter.writerow(row)
# Write design in GID format
write_itr = float(iterator)
self.gid_io.write_results(write_itr, self.opt_model_part, self.config.nodal_results, [])
# Check convergence (We ensure that at least 2 subiterations are done)
if(sub_opt_itr>3):
# Check if maximum iterations were reached
if(sub_opt_itr==self.config.max_sub_opt_iterations):
print("\n> Maximal iterations of supoptimization problem reached!")
break
# Check for relative tolerance
if(abs(delta_l_relative)<self.config.relative_tolerance_sub_opt):
print("\n> Optimization subproblem converged within a relative objective tolerance of ",self.config.relative_tolerance_sub_opt,"%.")
break
# Check if value of lagrangian increases
if(l>previous_l):
print("\n> Value of Lagrange function increased!")
break
# Update design
X = self.get_design()
# Store value of Lagrange function for next iteration
previous_l = l
# Store initial objective value
if(opt_itr==1 and sub_opt_itr==1):
initial_f = response[only_F_id]["value"]
initial_l = l
# Take time needed for current suboptimization step as well as for the overall opt so far
subopt_end_time = time.time()
print("\n> Time needed for current suboptimization step = ",round(subopt_end_time - subopt_start_time,2),"s")
print("\n> Time needed for total optimization so far = ",round(subopt_end_time - self.opt_start_time,2),"s")
# Check Convergence (More convergence criterion for major optimization iteration to be implemented!)
# Check if maximum iterations were reached
if(opt_itr==self.config.max_opt_iterations):
print("\n> Maximal iterations of optimization problem reached!")
break
# Update lagrange multipliers and penalty factor
self.vm_utils.udpate_augmented_lagrange_parameters( self.constraints, response )
# Take time needed for current optimization step
end_time = time.time()
print("\n> Time needed for current optimization step = ",round(end_time - start_time,2),"s")
# --------------------------------------------------------------------------
def start_penalized_projection(self):
# README!!!
# Note that the current implementation assumes that only one scalar objective & constraint is present
# Flags to trigger proper function calls
constraints_given = True
# Get Id of objective
only_F_id = None
for F_id in self.objectives:
only_F_id = F_id
break
# Get Id of constraint
only_C_id = None
for C_id in self.constraints:
only_C_id = C_id
break
# Initialize file where design evolution is recorded
with open(self.config.design_history_directory+"/"+self.config.design_history_file, 'w') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append("itr\t")
row.append("\tf\t")
row.append("\tdf_absolute[%]\t")
row.append("\tdf_relative[%]\t")
row.append("\tc["+str(only_C_id)+"]:"+str(self.constraints[only_C_id]["type"])+"\t")
row.append("\tc["+str(only_C_id)+"] / reference_value[%]"+"\t")
row.append("\tcorrection_scaling[-]\t")
row.append("\tstep_size[-]\t")
row.append("\tt_iteration[s]\t")
row.append("\tt_total[s]")
historyWriter.writerow(row)
# Define initial design (initial design corresponds to a zero shape update)
# Note that we assume an incremental design variable in vertex morphing
X = {}
for node in self.opt_model_part.Nodes:
X[node.Id] = [0.,0.,0.]
# Miscellaneous working variables for data management
initial_f = 0.0
previous_f = 0.0
previous_c = 0.0
# Start optimization loop
for opt_itr in range(1,self.config.max_opt_iterations+1):
# Some output
print("\n>===================================================================")
print("> ",time.ctime(),": Starting optimization iteration ",opt_itr)
print(">===================================================================\n")
# Start measuring time needed for current optimization step
start_time = time.time()
# Set controller to evaluate objective and constraint
self.controller.initialize_controls()
self.controller.get_controls()[only_F_id]["calc_value"] = 1
self.controller.get_controls()[only_C_id]["calc_value"] = 1
# Set controller to evaluate objective and constraint gradient
self.controller.get_controls()[only_F_id]["calc_gradient"] = 1
self.controller.get_controls()[only_C_id]["calc_gradient"] = 1
# Initialize response container
response = self.controller.create_response_container()
# Start analyzer according to specified controls
iterator = str(opt_itr) + ".0"
self.analyzer( X, self.controller.get_controls(), iterator, response )
# Check if constraint is active
for func_id in self.config.constraints:
if(self.config.constraints[func_id]["type"] == "eq"):
constraints_given = True
elif(response[func_id]["value"]>0):
constraints_given = True
else:
constraints_given = False
break
# Store gradients on the nodes of the model_part
self.store_grads_on_nodes( response[only_F_id]["gradient"], response[only_C_id]["gradient"] )
# Compute unit surface normals at each node of current design
self.geom_utils.compute_unit_surface_normals()
# Project received gradients on unit surface normal at each node to obtain normal gradients
self.geom_utils.project_grad_on_unit_surface_normal( constraints_given )
# Compute mapping matrix
self.mapper.compute_mapping_matrix()
# Map sensitivities to design space
self.mapper.map_sensitivities_to_design_space( constraints_given )
# # Adjustment of step size
# if( opt_itr > 1 and response[only_F_id]["value"]>previous_f):
# self.config.step_size = self.config.step_size/2
correction_scaling = [False]
if(constraints_given):
self.opt_utils.compute_projected_search_direction( response[only_C_id]["value"] )
self.opt_utils.correct_projected_search_direction( response[only_C_id]["value"], previous_c, correction_scaling )
self.opt_utils.compute_design_update()
else:
self.opt_utils.compute_search_direction_steepest_descent()
self.opt_utils.compute_design_update()
# Map design update to geometry space
self.mapper.map_design_update_to_geometry_space()
# Compute and output some measures to track objective function
delta_f_absolute = 0.0
delta_f_relative = 0.0
print("\n> Current value of objective function = ",response[only_F_id]["value"])
if(opt_itr>1):
delta_f_absolute = 100*(response[only_F_id]["value"]-initial_f)/initial_f
delta_f_relative = 100*(response[only_F_id]["value"]-previous_f)/initial_f
print("\n> Absolut change of objective function = ",round(delta_f_absolute,6)," [%]")
print("\n> Relative change of objective function = ",round(delta_f_relative,6)," [%]")
# Compute and output some measures to track function
print("\n> Current value of constraint function = ",round(response[only_C_id]["value"],12))
# Take time needed for current optimization step
end_time = time.time()
time_current_step = round(end_time - start_time,2)
time_optimization = round(end_time - self.opt_start_time,2)
print("\n> Time needed for current optimization step = ",time_current_step,"s")
print("\n> Time needed for total optimization so far = ",time_optimization,"s")
# Write design in GID format
self.gid_io.write_results(opt_itr, self.opt_model_part, self.config.nodal_results, [])
# Write design history to file
with open(self.config.design_history_directory+"/"+self.config.design_history_file, 'a') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append(str(opt_itr)+"\t")
row.append("\t"+str("%.12f"%(response[only_F_id]["value"]))+"\t")
row.append("\t"+str("%.2f"%(delta_f_absolute))+"\t")
row.append("\t"+str("%.6f"%(delta_f_relative))+"\t")
row.append("\t"+str("%.12f"%(response[only_C_id]["value"]))+"\t")
if not response[only_C_id]["reference_value"]:
row.append("\t"+str("-\t"))
else:
percentage_of_reference = 100*(response[only_C_id]["value"]/response[only_C_id]["reference_value"])
row.append("\t"+str("%.6f"%(percentage_of_reference)))
row.append("\t"+str("%.12f"%(correction_scaling[0]))+"\t")
row.append("\t"+str(self.config.step_size)+"\t")
row.append("\t"+str("%.1f"%(time_current_step))+"\t")
row.append("\t"+str("%.1f"%(time_optimization)))
historyWriter.writerow(row)
# Check convergence (Further convergence criterions to be implemented )
if(opt_itr>1):
# Check if maximum iterations were reached
if(opt_itr==self.config.max_opt_iterations):
print("\n> Maximal iterations of optimization problem reached!")
break
# Check for relative tolerance
if(abs(delta_f_relative)<self.config.relative_tolerance_objective):
print("\n> Optimization problem converged within a relative objective tolerance of ",self.config.relative_tolerance_objective,"%.")
break
# Update design
X = self.get_design()
# Store values of for next iteration
previous_f = response[only_F_id]["value"]
previous_c = response[only_C_id]["value"]
# Store initial objective value
if(opt_itr==1):
initial_f = response[only_F_id]["value"]
# --------------------------------------------------------------------------
def store_grads_on_nodes(self,objective_grads,constraint_grads={}):
# Read objective gradients
eucledian_norm_obj_sens = 0.0
for node_Id in objective_grads:
# If deactivated, nodal sensitivities will not be assigned and hence remain zero
if(self.opt_model_part.Nodes[node_Id].GetSolutionStepValue(SENSITIVITIES_DEACTIVATED)):
continue
# If not deactivated, nodal sensitivities will be assigned
sens_i = Vector(3)
sens_i[0] = objective_grads[node_Id][0]
sens_i[1] = objective_grads[node_Id][1]
sens_i[2] = objective_grads[node_Id][2]
self.opt_model_part.Nodes[node_Id].SetSolutionStepValue(OBJECTIVE_SENSITIVITY,0,sens_i)
# When constraint_grads is defined also store constraint sensitivities (bool returns false if dictionary is empty)
if(bool(constraint_grads)):
eucledian_norm_cons_sens = 0.0
for node_Id in constraint_grads:
# If deactivated, nodal sensitivities will not be assigned and hence remain zero
if(self.opt_model_part.Nodes[node_Id].GetSolutionStepValue(SENSITIVITIES_DEACTIVATED)):
continue
# If not deactivated, nodal sensitivities will be assigned
sens_i = Vector(3)
sens_i[0] = constraint_grads[node_Id][0]
sens_i[1] = constraint_grads[node_Id][1]
sens_i[2] = constraint_grads[node_Id][2]
self.opt_model_part.Nodes[node_Id].SetSolutionStepValue(CONSTRAINT_SENSITIVITY,0,sens_i)
# --------------------------------------------------------------------------
def get_design(self):
# Read and return the current design in the corresponding mode
X = {}
if(self.config.design_output_mode=="relative"):
for node in self.opt_model_part.Nodes:
X[node.Id] = node.GetSolutionStepValue(SHAPE_UPDATE)
elif(self.config.design_output_mode=="total"):
for node in self.opt_model_part.Nodes:
X[node.Id] = node.GetSolutionStepValue(SHAPE_CHANGE_ABSOLUTE)
elif(self.config.design_output_mode=="absolute"):
for node in self.opt_model_part.Nodes:
X[node.Id] = [node.X,node.Y,node.Z]
else:
sys.exit("Wrong definition of design_output_mode!")
return X
print("Stage 2, STEP = ", step)
fluid_solver.Solve()
PrintDrag(drag_list, drag_file_output_list, fluid_model_part, time)
for node in fluid_model_part.Nodes:
vel = node.GetSolutionStepValue(VELOCITY)
press = node.GetSolutionStepValue(PRESSURE)
node.SetSolutionStepValue(PRIMAL_VELOCITY,vel)
node.SetSolutionStepValue(PRIMAL_PRESSURE,press)
# Adjoint solution
adjoint_solver.Solve()
print("adjoint problem solved")
adjoint_solver.ComputeSensitivity()
print("sensitivity computed")
gid_io.write_results(
time,
fluid_model_part,
ProjectParameters.nodal_results,
ProjectParameters.gauss_points_results)
gid_io.finalize_results()
for i in drag_file_output_list:
i.close()
benchmarking.Output(TestNode.GetSolutionStepValue(SHAPE_SENSITIVITY_X), "Test SHAPE_SENSITIVITY_X", None, 0.001)
benchmarking.Output(TestNode.GetSolutionStepValue(SHAPE_SENSITIVITY_Y), "Test SHAPE_SENSITIVITY_Y", None, 0.001)
optimized_model_part = ModelPart("optimized_model_part")
optimized_model_part.AddNodalSolutionStepVariable(NORMAL)
restart_file_name = "Small_Cantilever_Restart_File_21"
model_part_io = ModelPartIO(restart_file_name)
model_part_io.ReadModelPart(optimized_model_part)
# Extract volume
threshold = 0.2
extracted_volume_model_part = ModelPart("extracted_volume_model_part")
TopologyExtractorUtilities().ExtractVolumeMesh(optimized_model_part, threshold, extracted_volume_model_part)
# Write extracted volume
gid_io = GiDOutput("extracted_volume_model_part", VolumeOutput, GiDPostMode, GiDMultiFileFlag, GiDWriteMeshFlag, GiDWriteConditionsFlag)
gid_io.initialize_results(extracted_volume_model_part)
gid_io.write_results(1, extracted_volume_model_part, nodal_results, gauss_points_results)
gid_io.finalize_results()
# Extract surface
extracted_surface_model_part = ModelPart("extracted_surface_model_part")
TopologyExtractorUtilities().ExtractSurfaceMesh(extracted_volume_model_part, extracted_surface_model_part)
# Write extracted surface
gid_io_2 = GiDOutput("extracted_surface_model_part", VolumeOutput, GiDPostMode, GiDMultiFileFlag, GiDWriteMeshFlag, GiDWriteConditionsFlag)
gid_io_2.initialize_results(extracted_surface_model_part)
gid_io_2.write_results(1, extracted_surface_model_part, nodal_results, gauss_points_results)
gid_io_2.finalize_results()
# Smooth extracted surface
smoothing_relaxation_factor = 1
smoothing_iterations = 5
smoothed_surface_model_part = ModelPart("extracted_surface_model_part")
class VertexMorphingMethod:
# --------------------------------------------------------------------------
def __init__(self,opt_model_part,config,analyzer):
# For GID output
self.gid_io = GiDOutput(config.design_surface_name,
config.VolumeOutput,
config.GiDPostMode,
config.GiDMultiFileFlag,
config.GiDWriteMeshFlag,
config.GiDWriteConditionsFlag)
# Add variables needed for shape optimization
opt_model_part.AddNodalSolutionStepVariable(NORMAL)
opt_model_part.AddNodalSolutionStepVariable(NORMALIZED_SURFACE_NORMAL)
opt_model_part.AddNodalSolutionStepVariable(OBJECTIVE_SENSITIVITY)
opt_model_part.AddNodalSolutionStepVariable(CONSTRAINT_SENSITIVITY)
opt_model_part.AddNodalSolutionStepVariable(SHAPE_UPDATE)
opt_model_part.AddNodalSolutionStepVariable(SHAPE_CHANGE_ABSOLUTE)
opt_model_part.AddNodalSolutionStepVariable(IS_ON_BOUNDARY)
opt_model_part.AddNodalSolutionStepVariable(BOUNDARY_PLANE)
opt_model_part.AddNodalSolutionStepVariable(SHAPE_UPDATES_DEACTIVATED)
opt_model_part.AddNodalSolutionStepVariable(SENSITIVITIES_DEACTIVATED)
# Read and print model part (design surface)
buffer_size = 1
model_part_io = ModelPartIO(config.design_surface_name)
model_part_io.ReadModelPart(opt_model_part)
opt_model_part.SetBufferSize(buffer_size)
print("\nThe following design surface was defined:\n\n",opt_model_part)
# Set configurations
self.config = config
# Set analyzer
self.analyzer = analyzer
# Set response functions
self.objectives = config.objectives
self.constraints = config.constraints
print("> The following objectives are defined:\n")
for func_id in config.objectives:
print(func_id,":",config.objectives[func_id],"\n")
print("> The following constraints are defined:\n")
for func_id in config.constraints:
print(func_id,":",config.constraints[func_id],"\n")
# Create controller object
self.controller = Controller( config );
# Model parameters
self.opt_model_part = opt_model_part
# Add toolbox for vertex morphing
max_nodes_affected = 10000 # Specification required by spatial (tree) search
# Defines maximum nodes that may be considered within the filter radius
self.vm_utils = VertexMorphingUtilities( self.opt_model_part,
self.config.domain_size,
self.objectives,
self.constraints,
config.filter_size,
max_nodes_affected)
# --------------------------------------------------------------------------
def optimize(self):
print("\n> ==============================================================================================================")
print("> Starting optimization using the following algorithm: ",self.config.optimization_algorithm)
print("> ==============================================================================================================\n")
# Start timer and assign to object such that total time of opt may be measured at each step
self.opt_start_time = time.time()
# Initialize design output in GID Format
self.gid_io.initialize_results(self.opt_model_part)
# Call for for specified optimization algorithm
if(self.config.optimization_algorithm == "steepest_descent"):
self.start_steepest_descent()
elif(self.config.optimization_algorithm == "augmented_lagrange"):
self.start_augmented_lagrange()
elif(self.config.optimization_algorithm == "penalized_projection"):
self.start_penalized_projection()
else:
sys.exit("Specified optimization_algorithm not implemented!")
# Finalize design output in GID formad
self.gid_io.finalize_results()
# Stop timer
opt_end_time = time.time()
print("\n> ==============================================================================================================")
print("> Finished optimization in ",round(opt_end_time - self.opt_start_time,1)," s!")
print("> ==============================================================================================================\n")
# --------------------------------------------------------------------------
def start_steepest_descent(self):
# Flags to trigger proper function calls
constraints_given = False
# Get Id of objective
only_F_id = None
for F_id in self.objectives:
only_F_id = F_id
break
# Initialize file where design evolution is recorded
with open(self.config.design_history_file, 'w') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append("itr")
row.append("\tf")
row.append("\tdf_absolute[%]")
row.append("\tdf_relative[%]")
historyWriter.writerow(row)
# Define initial design (initial design corresponds to a zero shape update)
# Note that we assume an incremental design variable in vertex morphing
X = {}
for node in self.opt_model_part.Nodes:
X[node.Id] = [0.,0.,0.]
# Miscellaneous working variables for data management
initial_f = 0.0
previous_f = 0.0
# Start optimization loop
for opt_itr in range(1,self.config.max_opt_iterations+1):
# Some output
print("\n>===============================================")
print("> Starting optimization iteration ",opt_itr)
print(">===============================================\n")
# Start measuring time needed for current optimization step
start_time = time.time()
# Set controller to evaluate objective
self.controller.initialize_controls()
self.controller.get_controls()[only_F_id]["calc_func"] = 1
# Set to evaluate objective gradient if provided
if(self.objectives[only_F_id]["grad"]=="provided"):
self.controller.get_controls()[only_F_id]["calc_grad"] = 1
# Initialize response container
response = self.controller.create_response_container()
# Start analyzer according to specified controls
iterator = str(opt_itr) + ".0"
self.analyzer( X, self.controller.get_controls(), iterator, response )
# Store gradients on the nodes of the model_part
self.store_grads_on_nodes( response[only_F_id]["grad"] )
# Compute unit surface normals at each node of current design
self.vm_utils.compute_unit_surface_normals()
# Project received gradients on unit surface normal at each node to obtain normal gradients
self.vm_utils.project_grad_on_unit_surface_normal( constraints_given )
# Compute filtered gradients (do backward mapping)
self.vm_utils.filter_gradients( constraints_given )
# Compute search direction
self.vm_utils.compute_search_direction_steepest_descent()
# Compute design variable update (do one step in the optimization algorithm)
self.vm_utils.update_design_variable( self.config.step_size_0 )
# Compute shape update (do forward mapping)
self.vm_utils.update_shape()
# Compute and output some measures to track changes in the objective function
delta_f_absolute = 0.0
delta_f_relative = 0.0
print("\n> Current value of objective function = ",response[only_F_id]["func"])
if(opt_itr>1):
delta_f_absolute = 100* ( response[only_F_id]["func"]/initial_f - 1 )
delta_f_relative = 100*( response[only_F_id]["func"]/previous_f - 1 )
print("\n> Absolut change of objective function = ",round(delta_f_absolute,6)," [%]")
print("\n> Relative change of objective function = ",round(delta_f_relative,6)," [%]")
# Write design history to file
with open(self.config.design_history_file, 'a') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append(str(opt_itr)+"\t")
row.append("\t"+str("%.12f"%(response[only_F_id]["func"]))+"\t")
row.append("\t"+str("%.2f"%(delta_f_absolute))+"\t")
row.append("\t"+str("%.6f"%(delta_f_relative))+"\t")
historyWriter.writerow(row)
# Write design in GID format
self.gid_io.write_results(opt_itr, self.opt_model_part, self.config.nodal_results, [])
# Check convergence
if(opt_itr>1):
# Check if maximum iterations were reached
if(opt_itr==self.config.max_opt_iterations):
print("\n> Maximal iterations of optimization problem reached!")
break
# Check for relative tolerance
if(abs(delta_f_relative)<self.config.relative_tolerance_objective):
print("\n> Optimization problem converged within a relative objective tolerance of ",self.config.relative_tolerance_objective,".")
break
# # Check if value of objective increases
# if(response[only_F_id]["func"]>previous_f):
# print("\n> Value of objective function increased!")
# break
# Update design
X = self.get_design()
# Store values of for next iteration
previous_f = response[only_F_id]["func"]
# Store initial objective value
if(opt_itr==1):
initial_f = response[only_F_id]["func"]
# Take time needed for current optimization step
end_time = time.time()
print("\n> Time needed for current optimization step = ",round(end_time - start_time,1),"s")
print("\n> Time needed for total optimization so far = ",round(end_time - self.opt_start_time,1),"s")
# --------------------------------------------------------------------------
def start_augmented_lagrange(self):
# README!!!
# Note that the current implementation assumes that only one scalar objective & constraint is present
# Flags to trigger proper function calls
constraints_given = True
# Get Id of objective
only_F_id = None
for F_id in self.objectives:
only_F_id = F_id
break
# Get Id of constraint
only_C_id = None
for C_id in self.constraints:
only_C_id = C_id
break
# Initialize the optimization algorithm
self.vm_utils.initialize_augmented_lagrange( self.constraints,
self.config.penalty_fac_0,
self.config.gamma,
self.config.penalty_fac_max,
self.config.lambda_0 )
# Initialize file where design evolution is recorded
with open(self.config.design_history_file, 'w') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append("itr\t")
row.append("\tsub_itr\t")
row.append("\tl\t")
row.append("\tdl_relative[%]\t")
row.append("\tf\t")
row.append("\tdf_absolute[%]\t")
row.append("\tpenalty_fac\t")
row.append("\tC["+str(only_C_id)+"]: "+str(self.constraints[only_C_id]["type"])+"\t")
row.append("\tlambda["+str(only_C_id)+"]\t")
historyWriter.writerow(row)
# Define initial design (initial design corresponds to a zero shape update)
# Note that we assume an incremental design variable in vertex morphing
X = {}
for node in self.opt_model_part.Nodes:
X[node.Id] = [0.,0.,0.]
# Miscellaneous working variables for data management
initial_f = 0.0
previous_l = 0.0
# Start primary optimization loop
for opt_itr in range(1,self.config.max_opt_iterations+1):
# Some output
print("\n>===============================================")
print("> Starting optimization iteration ",opt_itr)
print(">===============================================\n")
# Start measuring time needed for current optimization step
start_time = time.time()
# Solve optimization subproblem
for sub_opt_itr in range(1,self.config.max_sub_opt_iterations+1):
# Some output
print("\n>===============================================")
print("> Starting suboptimization iteration ",sub_opt_itr)
print(">===============================================\n")
# Start measuring time needed for current suboptimization step
subopt_start_time = time.time()
# Set controller to evaluate objective and constraint
self.controller.initialize_controls()
self.controller.get_controls()[only_F_id]["calc_func"] = 1
self.controller.get_controls()[only_C_id]["calc_func"] = 1
# Set controller to evaluate objective and constraint gradient if provided
if(self.objectives[only_F_id]["grad"]=="provided"):
self.controller.get_controls()[only_F_id]["calc_grad"] = 1
if(self.constraints[only_C_id]["grad"]=="provided"):
self.controller.get_controls()[only_C_id]["calc_grad"] = 1
# Initialize response container
response = self.controller.create_response_container()
# Start analyzer according to specified controls
iterator = str(opt_itr) + "." + str(sub_opt_itr)
self.analyzer( X, self.controller.get_controls(), iterator, response )
# Evaluate Lagrange function
l = self.vm_utils.get_value_of_augmented_lagrangian( only_F_id, self.constraints, response )
# Store gradients on the nodes of the model_part
self.store_grads_on_nodes( response[only_F_id]["grad"], response[only_C_id]["grad"] )
# Compute unit surface normals at each node of current design
self.vm_utils.compute_unit_surface_normals()
# Project received gradients on unit surface normal at each node to obtain normal gradients
dJdX_n = self.vm_utils.project_grad_on_unit_surface_normal( constraints_given )
# Compute filtered gradients (do backward mapping)
self.vm_utils.filter_gradients( constraints_given )
# Compute search direction
self.vm_utils.compute_search_direction_augmented_lagrange( self.constraints, response )
# Compute design variable update (do one step in the optimization algorithm)
self.vm_utils.update_design_variable( self.config.step_size_0 )
# Compute shape update (do forward mapping)
self.vm_utils.update_shape()
# Compute and output some measures to track changes in the objective function
delta_f_absolute = 0.0
delta_l_relative = 0.0
print("\n> Current value of Lagrange function = ",round(l,12))
if(sub_opt_itr>1):
delta_f_absolute = 100* ( response[only_F_id]["func"]/initial_f - 1 )
delta_l_relative = 100*( l/previous_l - 1 )
print("\n> Relative change of Lagrange function = ",round(delta_l_relative,6)," [%]")
# We write every major and every suboptimization iteration in design history
with open(self.config.design_history_file, 'a') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append(str(opt_itr)+"\t")
row.append("\t"+str(sub_opt_itr)+"\t")
row.append("\t"+str("%.12f"%(l))+"\t")
row.append("\t"+str("%.6f"%(delta_l_relative))+"\t")
row.append("\t"+str("%.12f"%(response[only_F_id]["func"]))+"\t")
row.append("\t"+str("%.2f"%(delta_f_absolute))+"\t")
row.append("\t"+str("%.2f"%(self.vm_utils.get_penalty_fac()))+"\t")
row.append("\t"+str("%.12f"%(response[only_C_id]["func"]))+"\t")
row.append("\t"+str("%.12f"%(self.vm_utils.get_lambda(only_C_id)))+"\t")
historyWriter.writerow(row)
# Write design in GID format
write_itr = float(iterator)
self.gid_io.write_results(write_itr, self.opt_model_part, self.config.nodal_results, [])
# Check convergence (We ensure that at least 2 subiterations are done)
if(sub_opt_itr>3):
# Check if maximum iterations were reached
if(sub_opt_itr==self.config.max_sub_opt_iterations):
print("\n> Maximal iterations of supoptimization problem reached!")
break
# Check for relative tolerance
if(abs(delta_l_relative)<self.config.relative_tolerance_sub_opt):
print("\n> Optimization subproblem converged within a relative objective tolerance of ",self.config.relative_tolerance_sub_opt,".")
break
# Check if value of lagrangian increases
if(l>previous_l):
print("\n> Value of Lagrange function increased!")
break
# Update design
X = self.get_design()
# Store value of Lagrange function for next iteration
previous_l = l
# Store initial objective value
if(opt_itr==1 and sub_opt_itr==1):
initial_f = response[only_F_id]["func"]
# Take time needed for current suboptimization step as well as for the overall opt so far
subopt_end_time = time.time()
print("\n> Time needed for current suboptimization step = ",round(subopt_end_time - subopt_start_time,1),"s")
print("\n> Time needed for total optimization so far = ",round(subopt_end_time - self.opt_start_time,1),"s")
# Check Convergence (More convergence criterion for major optimization iteration to be implemented!)
# Check if maximum iterations were reached
if(opt_itr==self.config.max_opt_iterations):
print("\n> Maximal iterations of optimization problem reached!")
break
# Update lagrange multipliers and penalty factor
self.vm_utils.udpate_augmented_lagrange_parameters( self.constraints, response )
# Take time needed for current optimization step
end_time = time.time()
print("\n> Time needed for current optimization step = ",round(end_time - start_time,1),"s")
# --------------------------------------------------------------------------
def start_penalized_projection(self):
# README!!!
# Note that the current implementation assumes that only one scalar objective & constraint is present
# Flags to trigger proper function calls
constraints_given = True
# Get Id of objective
only_F_id = None
for F_id in self.objectives:
only_F_id = F_id
break
# Get Id of constraint
only_C_id = None
for C_id in self.constraints:
only_C_id = C_id
break
# Initialize file where design evolution is recorded
with open(self.config.design_history_file, 'w') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append("itr")
row.append("\tf")
row.append("\tdf_absolute[%]")
row.append("\tdf_relative[%]")
row.append("\tC["+str(only_C_id)+"]: "+str(self.constraints[only_C_id]["type"])+"\t")
historyWriter.writerow(row)
# Define initial design (initial design corresponds to a zero shape update)
# Note that we assume an incremental design variable in vertex morphing
X = {}
for node in self.opt_model_part.Nodes:
X[node.Id] = [0.,0.,0.]
# Miscellaneous working variables for data management
initial_f = 0.0
previous_f = 0.0
# Start optimization loop
for opt_itr in range(1,self.config.max_opt_iterations+1):
# Some output
print("\n>===============================================")
print("> Starting optimization iteration ",opt_itr)
print(">===============================================\n")
# Start measuring time needed for current optimization step
start_time = time.time()
# Set controller to evaluate objective and constraint
self.controller.initialize_controls()
self.controller.get_controls()[only_F_id]["calc_func"] = 1
self.controller.get_controls()[only_C_id]["calc_func"] = 1
# Set controller to evaluate objective and constraint gradient if provided
if(self.objectives[only_F_id]["grad"]=="provided"):
self.controller.get_controls()[only_F_id]["calc_grad"] = 1
if(self.constraints[only_C_id]["grad"]=="provided"):
self.controller.get_controls()[only_C_id]["calc_grad"] = 1
# Initialize response container
response = self.controller.create_response_container()
# Start analyzer according to specified controls
iterator = str(opt_itr) + ".0"
self.analyzer( X, self.controller.get_controls(), iterator, response )
# Scale constraint if specified
response[only_C_id]["func"] = response[only_C_id]["func"]*self.config.constraint_scaling
if(self.config.constraint_scaling!=1.0):
for node_Id in response[only_C_id]["grad"]:
response[only_C_id]["grad"][node_Id] = response[only_C_id]["grad"][node_Id]*self.config.constraint_scaling
# Check if constraint is active
for func_id in self.config.constraints:
if(self.config.constraints[func_id]["type"] == "eq"):
constraints_given = True
elif(response[func_id]["func"]>0):
constraints_given = True
else:
constraints_given = False
break
# Store gradients on the nodes of the model_part
self.store_grads_on_nodes( response[only_F_id]["grad"], response[only_C_id]["grad"] )
# Compute unit surface normals at each node of current design
self.vm_utils.compute_unit_surface_normals()
# Project received gradients on unit surface normal at each node to obtain normal gradients
self.vm_utils.project_grad_on_unit_surface_normal( constraints_given )
# Compute filtered gradients (do backward mapping)
self.vm_utils.filter_gradients( constraints_given )
# Compute search direction
if(constraints_given):
self.vm_utils.compute_search_direction_penalized_projection( response[only_C_id]["func"] )
else:
self.vm_utils.compute_search_direction_steepest_descent()
# Compute design variable update (do one step in the optimization algorithm)
self.vm_utils.update_design_variable( self.config.step_size_0 )
# Compute shape update (do forward mapping)
self.vm_utils.update_shape()
# Compute and output some measures to track changes in the objective function
delta_f_absolute = 0.0
delta_f_relative = 0.0
print("\n> Current value of objective function = ",response[only_F_id]["func"])
if(opt_itr>1):
delta_f_absolute = 100* ( response[only_F_id]["func"]/initial_f - 1 )
delta_f_relative = 100*( response[only_F_id]["func"]/previous_f - 1 )
print("\n> Absolut change of objective function = ",round(delta_f_absolute,6)," [%]")
print("\n> Relative change of objective function = ",round(delta_f_relative,6)," [%]")
print("\n> Current value of constraint function = ",round(response[only_C_id]["func"],12))
# Write design history to file
with open(self.config.design_history_file, 'a') as csvfile:
historyWriter = csv.writer(csvfile, delimiter=',',quotechar='|',quoting=csv.QUOTE_MINIMAL)
row = []
row.append(str(opt_itr)+"\t")
row.append("\t"+str("%.12f"%(response[only_F_id]["func"]))+"\t")
row.append("\t"+str("%.2f"%(delta_f_absolute))+"\t")
row.append("\t"+str("%.6f"%(delta_f_relative))+"\t")
row.append("\t"+str("%.12f"%(response[only_C_id]["func"]))+"\t")
historyWriter.writerow(row)
# Write design in GID format
self.gid_io.write_results(opt_itr, self.opt_model_part, self.config.nodal_results, [])
# Check convergence (Further convergence criterions to be implemented )
if(opt_itr>1):
# Check if maximum iterations were reached
if(opt_itr==self.config.max_opt_iterations):
print("\n> Maximal iterations of optimization problem reached!")
break
# Update design
X = self.get_design()
# Store values of for next iteration
previous_f = response[only_F_id]["func"]
# Store initial objective value
if(opt_itr==1):
initial_f = response[only_F_id]["func"]
# Take time needed for current optimization step
end_time = time.time()
print("\n> Time needed for current optimization step = ",round(end_time - start_time,1),"s")
print("\n> Time needed for total optimization so far = ",round(end_time - self.opt_start_time,1),"s")
# --------------------------------------------------------------------------
def store_grads_on_nodes(self,objective_grads,constraint_grads={}):
# Read objective gradients
eucledian_norm_obj_sens = 0.0
for node_Id in objective_grads:
# If deactivated, nodal sensitivities will not be assigned and hence remain zero
if(self.opt_model_part.Nodes[node_Id].GetSolutionStepValue(SENSITIVITIES_DEACTIVATED)):
continue
# If not deactivated, nodal sensitivities will be assigned
sens_i = Vector(3)
sens_i[0] = objective_grads[node_Id][0]
sens_i[1] = objective_grads[node_Id][1]
sens_i[2] = objective_grads[node_Id][2]
self.opt_model_part.Nodes[node_Id].SetSolutionStepValue(OBJECTIVE_SENSITIVITY,0,sens_i)
# When constraint_grads is defined also store constraint sensitivities (bool returns false if dictionary is empty)
if(bool(constraint_grads)):
eucledian_norm_cons_sens = 0.0
for node_Id in constraint_grads:
# If deactivated, nodal sensitivities will not be assigned and hence remain zero
if(self.opt_model_part.Nodes[node_Id].GetSolutionStepValue(SENSITIVITIES_DEACTIVATED)):
continue
# If not deactivated, nodal sensitivities will be assigned
sens_i = Vector(3)
sens_i[0] = constraint_grads[node_Id][0]
sens_i[1] = constraint_grads[node_Id][1]
sens_i[2] = constraint_grads[node_Id][2]
self.opt_model_part.Nodes[node_Id].SetSolutionStepValue(CONSTRAINT_SENSITIVITY,0,sens_i)
# --------------------------------------------------------------------------
def get_design(self):
# Read and return the current design in the corresponding mode
X = {}
if(self.config.design_output_mode=="relative"):
for node in self.opt_model_part.Nodes:
X[node.Id] = node.GetSolutionStepValue(SHAPE_UPDATE)
elif(self.config.design_output_mode=="total"):
for node in self.opt_model_part.Nodes:
X[node.Id] = node.GetSolutionStepValue(SHAPE_CHANGE_ABSOLUTE)
elif(self.config.design_output_mode=="absolute"):
for node in self.opt_model_part.Nodes:
X[node.Id] = [node.X,node.Y,node.Z]
else:
sys.exit("Wrong definition of design_output_mode!")
return X
#mount the search structure
locator = BinBasedFastPointLocator2D(base_model_part)
locator.UpdateSearchDatabase()
#locator.UpdateSearchDatabaseAssignedSize(0.01)
#construct the utility to move the points
bfecc_utility = BFECCConvection2D(locator)
base_model_part.CloneTimeStep(0.00)
substepping = 10.0
final_time = 10.0 # math.pi
t = 0.0
dt = 0.05 #1
step = 0
while (t < final_time):
t = t + dt
base_model_part.CloneTimeStep(t)
print("time = ", t)
bfecc_utility.BFECCconvect(base_model_part, TEMPERATURE, VELOCITY,
substepping)
gid_io_base.write_results(t, base_model_part,
ProjectParameters.nodal_results,
ProjectParameters.gauss_points_results)
step += 1
gid_io_base.finalize_results()
class SIMPMethod:
# --------------------------------------------------------------------------
def __init__(self, opt_model_part, config, analyzer):
# Set Topology Optimization configurations
self.config = config
# For GID output
self.gid_io = GiDOutput(config.GiD_output_file_name,
config.VolumeOutput, config.GiDPostMode,
config.GiDMultiFileFlag,
config.GiDWriteMeshFlag,
config.GiDWriteConditionsFlag)
# Set analyzer
self.analyzer = analyzer
# Set response functions
self.objectives = config.objectives
self.constraints = config.constraints
print("\n::[Initializing Topology Optimization Application]::")
print(" The following objectives are defined:")
for func_id in config.objectives:
print(" ", func_id, ":", config.objectives[func_id], "\n")
print(" The following constraints are defined:")
for func_id in config.constraints:
print(" ", func_id, ":", config.constraints[func_id])
# Create controller object
self.controller = Controller(config)
# Model parameters
self.opt_model_part = opt_model_part
# Initialize element variables
for element_i in opt_model_part.Elements:
element_i.SetValue(E_MIN, config.E_min)
element_i.SetValue(PENAL, config.penalty)
element_i.SetValue(X_PHYS, config.initial_volume_fraction)
element_i.SetValue(X_PHYS_OLD, config.initial_volume_fraction)
element_i.SetValue(
E_0, opt_model_part.Properties[config.simp_property].GetValue(
YOUNG_MODULUS))
element_i.SetValue(
YOUNG_MODULUS, opt_model_part.Properties[
config.simp_property].GetValue(YOUNG_MODULUS))
# Only happens if continuation strategy is activated (Initialization of penalty factor)
if (self.config.continuation_strategy == 1):
for element_i in self.opt_model_part.Elements:
element_i.SetValue(PENAL, 1)
# Add toolbox for topology filtering utilities
self.filter_utils = TopologyFilteringUtilities(
opt_model_part, self.config.filter_radius,
self.config.max_elements_in_filter_radius)
# Add toolbox for topology updating utilities
self.design_update_utils = TopologyUpdatingUtilities(opt_model_part)
# Add toolbox for I/O
self.io_utils = IOUtilities()
# --------------------------------------------------------------------------
def optimize(self):
print(
"\n> =============================================================================================================="
)
print("> Starting optimization using the following algorithm: ",
self.config.optimization_algorithm)
print(
"> =============================================================================================================="
)
# Start timer and assign to object such that total time of opt may be measured at each step
self.opt_start_time = time.time()
# Initialize the design output in GiD format and print initial 0 state
self.gid_io.initialize_results(self.opt_model_part)
self.gid_io.write_results(0, self.opt_model_part,
self.config.nodal_results,
self.config.gauss_points_results)
# Call for the specified optimization algorithm
if (self.config.optimization_algorithm == "oc_algorithm"):
self.start_oc_algorithm()
else:
raise TypeError(
"Specified optimization_algorithm not implemented!")
# Finalize the design output in GiD format
self.gid_io.finalize_results()
# Stop timer
opt_end_time = time.time()
print(
"\n> =============================================================================================================="
)
print("> Finished optimization in ",
round(opt_end_time - self.opt_start_time, 1), " s!")
print(
"> =============================================================================================================="
)
# --------------------------------------------------------------------------
def start_oc_algorithm(self):
# Get Id of objective & constraint
only_F_id = None
only_C_id = None
for F_id in self.objectives:
only_F_id = F_id
break
for C_id in self.constraints:
only_C_id = C_id
break
# Initialize variables for comparison purposes in Topology Optimization Tool
pmax = self.config.penalty # Maximum penalty value used for continuation strategy
Obj_Function = None
Obj_Function_old = None
Obj_Function_initial = None
Obj_Function_relative_change = None
Obj_Function_absolute_change = None
# Print the Topology Optimization Settings that will be used in the program
print("\n::[Topology Optimization Settings]::")
print(" E_min: ", self.config.E_min)
print(" Filter radius: ", self.config.filter_radius)
print(" Penalty factor: ", self.config.penalty)
print(" Rel. Tolerance: ", self.config.relative_tolerance)
print(" Volume Fraction:", self.config.initial_volume_fraction)
print(" Max. number of iterations:", self.config.max_opt_iterations)
if (self.config.restart_write_frequency <
self.config.max_opt_iterations):
if (self.config.restart_write_frequency == 1):
print(" Make a restart file every iteration")
elif (self.config.restart_write_frequency > 1):
print(" Make a restart file every",
self.config.restart_write_frequency, "iterations")
else:
print(" No restart file will be done during the simulation")
else:
print(" No restart file will be done during the simulation")
# Start optimization loop
for opt_itr in range(1, self.config.max_opt_iterations + 1):
# Some output
print(
"\n> =============================================================================================="
)
print("> Starting optimization iteration ", opt_itr)
print(
"> ==============================================================================================\n"
)
# Start measuring time needed for current optimization step
start_time = time.time()
# Initialize response container
response = self.controller.create_response_container()
# Set controller to evaluate objective & constraint
self.controller.initialize_controls()
self.controller.get_controls()[only_F_id]["calc_func"] = 1
self.controller.get_controls()[only_C_id]["calc_func"] = 1
# Set to evaluate objective & constraint gradient if provided
if (self.objectives[only_F_id]["grad"] == "provided"):
self.controller.get_controls()[only_F_id]["calc_grad"] = 1
if (self.constraints[only_C_id]["grad"] == "provided"):
self.controller.get_controls()[only_C_id]["calc_grad"] = 1
# RUN FEM: Call analyzer with current X to compute response (global_strain_energy, dcdx)
self.analyzer(self.controller.get_controls(), response, opt_itr)
# Filter sensitivities
print("\n::[Filter Sensitivities]::")
self.filter_utils.ApplyFilter(self.config.filter_type,
self.config.filter_kernel)
# Update design variables ( densities ) --> new X by:
print("\n::[Update Densities]::")
self.design_update_utils.UpdateDensitiesUsingOCMethod(
self.config.optimization_algorithm,
self.config.initial_volume_fraction,
self.config.grey_scale_filter, opt_itr, self.config.q_max)
# Print of results
print("\n::[RESULTS]::")
Obj_Function = response[only_F_id]["func"]
C_Function = response[only_C_id]["func"]
print(" Obj. function value = ",
math.ceil(Obj_Function * 1000000) / 1000000)
print(" Const. function value = ",
math.ceil(C_Function * 1000000) / 1000000)
if opt_itr == 1:
Obj_Function_initial = Obj_Function
if opt_itr > 1:
Obj_Function_relative_change = (
Obj_Function - Obj_Function_old) / Obj_Function_initial
print(
" Relative Obj. Function change =",
math.ceil(
(Obj_Function_relative_change * 100) * 10000) / 10000,
"%")
Obj_Function_absolute_change = (
Obj_Function - Obj_Function_initial) / Obj_Function_initial
print(
" Absolute Obj. Function change =",
math.ceil(
(Obj_Function_absolute_change * 100) * 10000) / 10000,
"%")
Obj_Function_old = Obj_Function
# Write design in GiD format
self.gid_io.write_results(opt_itr, self.opt_model_part,
self.config.nodal_results,
self.config.gauss_points_results)
# Continuation Strategy
if (self.config.continuation_strategy == 1):
print(
" Continuation Strategy for current iteration was ACTIVE")
if opt_itr < 20:
for element_i in self.opt_model_part.Elements:
element_i.SetValue(PENAL, 1)
else:
for element_i in self.opt_model_part.Elements:
element_i.SetValue(
PENAL, min(pmax, 1.02 * element_i.GetValue(PENAL)))
else:
print(
" Continuation Strategy for current iteration was UNACTIVE"
)
# Write restart file every selected number of iterations
restart_filename = self.config.restart_output_file.replace(
".mdpa", "_" + str(opt_itr) + ".mdpa")
if (self.config.restart_write_frequency > 0):
if (opt_itr % self.config.restart_write_frequency == False):
print("\n::[Restart File]::")
print(" Saving file at iteration", opt_itr)
self.io_utils.SaveOptimizationResults(
self.config.restart_input_file, self.opt_model_part,
restart_filename)
# Check convergence
if opt_itr > 1:
# Check if maximum iterations were reached
if (opt_itr == self.config.max_opt_iterations):
end_time = time.time()
print("\n Time needed for current optimization step = ",
round(end_time - start_time, 1), "s")
print(" Time needed for total optimization so far = ",
round(end_time - self.opt_start_time, 1), "s")
print(
"\n Maximal iterations of optimization problem reached!"
)
self.io_utils.SaveOptimizationResults(
self.config.restart_input_file, self.opt_model_part,
restart_filename)
break
# Check for relative tolerance
if (abs(Obj_Function_relative_change) <
self.config.relative_tolerance):
end_time = time.time()
print("\n Time needed for current optimization step = ",
round(end_time - start_time, 1), "s")
print(" Time needed for total optimization so far = ",
round(end_time - self.opt_start_time, 1), "s")
print(
"\n Optimization problem converged within a relative objective tolerance of",
self.config.relative_tolerance)
self.io_utils.SaveOptimizationResults(
self.config.restart_input_file, self.opt_model_part,
restart_filename)
break
# Set X_PHYS_OLD to X_PHYS to update the value for the next simulation's "change percentage"
for element_i in self.opt_model_part.Elements:
element_i.SetValue(X_PHYS_OLD, element_i.GetValue(X_PHYS))
# Take time needed for current optimization step
end_time = time.time()
print("\n Time needed for current optimization step = ",
round(end_time - start_time, 1), "s")
print(" Time needed for total optimization so far = ",
round(end_time - self.opt_start_time, 1), "s")
locator.UpdateSearchDatabase()
#locator.UpdateSearchDatabaseAssignedSize(0.01)
#construct the utility to move the points
bfecc_utility = BFECCConvection2D(locator)
base_model_part.CloneTimeStep(0.00)
substepping = 10.0
final_time = 10.0# math.pi
t=0.0
dt = 0.05 #1
step = 0
while( t < final_time):
t = t+dt
base_model_part.CloneTimeStep(t)
print("time = ",t)
bfecc_utility.BFECCconvect(base_model_part,TEMPERATURE,VELOCITY,substepping)
gid_io_base.write_results(
t,
base_model_part,
ProjectParameters.nodal_results,
ProjectParameters.gauss_points_results)
step +=1
gid_io_base.finalize_results()