Bring a Field’s Data Locally to Improve Performance#

Reducing the number of calls to the server is key to improving performance. Using the as_local_field option brings the data from the server to your local machine where you can work on it. When finished, you send the updated data back to the server in one transaction.

Import necessary modules:

from ansys.dpf import core as dpf
from ansys.dpf.core import examples
from ansys.dpf.core import operators as ops

Create a model object to establish a connection with an example result file and then extract:

model = dpf.Model(examples.download_multi_stage_cyclic_result())


DPF Model
Modal analysis
Unit system: MKS: m, kg, N, s, V, A, degC
Physics Type: Mecanic
Available results:
     -  displacement: Nodal Displacement
     -  stress: ElementalNodal Stress
     -  elastic_strain: ElementalNodal Strain
     -  structural_temperature: ElementalNodal Temperature
DPF  Meshed Region:
  3595 nodes
  1557 elements
  Unit: m
  With solid (3D) elements
DPF  Time/Freq Support:
  Number of sets: 6
Cumulative     Frequency (Hz) LoadStep       Substep        Harmonic index
1              188.385357     1              1              0.000000
2              325.126418     1              2              0.000000
3              595.320548     1              3              0.000000
4              638.189511     1              4              0.000000
5              775.669703     1              5              0.000000
6              928.278013     1              6              0.000000

Create the Workflow#

Maximum principal stress usually occurs on the skin of the model. Computing results only on this skin reduces the data size.

# Create a simple workflow computing the principal stress on the skin
# of the model.

skin_op = ops.mesh.external_layer(model.metadata.meshed_region)
skin_mesh = skin_op.outputs.mesh()

Plot the mesh skin:

05 use local data

Compute the stress principal inveriants on the skin nodes only:

stress_op = ops.result.stress(data_sources=model.metadata.data_sources)

principal_op = ops.invariant.principal_invariants_fc(stress_op)
principal_stress_1 = principal_op.outputs.fields_eig_1()[0]
principal_stress_2 = principal_op.outputs.fields_eig_2()[0]
principal_stress_3 = principal_op.outputs.fields_eig_3()[0]

Manipulate Data Locally#

This example goes over the fields, keeping the largest invariant value by node if the averaged value of invariants is large enough. Exploring data allows you to customize it to meet your needs.

node_scoping_ids = principal_stress_1.scoping.ids
threshold = 300000.0

field_to_keep = dpf.fields_factory.create_scalar_field(
    len(node_scoping_ids), location=dpf.locations.nodal

with field_to_keep.as_local_field() as f:
    with principal_stress_1.as_local_field() as s1:
        with principal_stress_2.as_local_field() as s2:
            with principal_stress_3.as_local_field() as s3:
                for i, id in enumerate(node_scoping_ids):
                    d1 = abs(s1.get_entity_data_by_id(id))
                    d2 = abs(s2.get_entity_data_by_id(id))
                    d3 = abs(s3.get_entity_data_by_id(id))
                    if (d1 + d2 + d3) / 3.0 > threshold:
                        d = max(d1, d2, d3)
                        f.append(d, id)

Plot Result Field#

Plot the result field on the skin mesh:

05 use local data

Plot Initial Invariants#

Plot the initial invariants on the skin mesh:

  • 05 use local data
  • 05 use local data
  • 05 use local data

Total running time of the script: ( 0 minutes 5.718 seconds)

Gallery generated by Sphinx-Gallery