Building a SimAI Model

This example demonstrates how to configure a SimAI model with inputs, outputs, global coefficients, and domain of analysis, then start the model training process.

Before you begin

• Complete “Creating a SimAI Project and Uploading Training Data” to create a project with training data.

• Ensure all training data in your project are ready (processed successfully).

• Know the names of surfaces and boundary conditions in your training data.

• Determine which global coefficients you want to calculate.

Import necessary libraries

import ansys.simai.core as asc
from ansys.simai.core.data.model_configuration import (
    DomainOfAnalysis,
    GlobalCoefficientDefinition,
    ModelConfiguration,
    ModelInput,
    ModelOutput,
)

Configure your settings

Update these variables with your specific settings:

ORGANIZATION_NAME = "<your_organization>"  # Replace with your organization name
PROJECT_NAME = "<your_project_name>"  # Replace with your project name

Initialize the client and get the project

Connect to SimAI:

simai_client = asc.SimAIClient(organization=ORGANIZATION_NAME)

Retrieve the project by name:

project = simai_client.projects.get(name=PROJECT_NAME)
print(f"Using existing project: {PROJECT_NAME}")

Configure model inputs

Define which surfaces and boundary conditions the model will use as inputs. Replace the placeholder names with the actual names from your training data:

model_input = ModelInput(
    surface=["<your_surface_name>"],  # List of surface names to use as input
    boundary_conditions=["<your_BC_name>"],  # List of boundary condition names
)

Configure model outputs

Define which surfaces the model will predict. These are typically the same surfaces as the inputs, but you can specify different ones:

model_output = ModelOutput(
    surface=["<your_surface_name>"],  # List of surface names to predict
)

Define global coefficients

Global coefficients are scalar values calculated from the prediction results. They can be used to extract key performance indicators from your simulations.

In this example, we calculate the integral of a field called “Photometric”. You can define multiple global coefficients with different formulas.

Available locations: "points" or "cells".

global_coefficients = [
    GlobalCoefficientDefinition(
        name="<your_global_coefficient_name>",  # Name for this global coefficient
        formula="integral(Photometric)",  # Formula to calculate the global coefficient value
        gc_location="points",  # Calculate at points (or at "cells")
    )
]

Define domain of analysis

The domain of analysis specifies the spatial extent that the model will analyze. Values can be specified as absolute dimensions or relative to minimum values.

Format: reference_type, min_value, max_value

• reference_type: "relative_to_min", "relative_to_max", "relative_to_center” or "absolute".

• min_value: minimum dimension value.

• max_value: maximum dimension value.

You can find appropriate values by analyzing your training data in the SimAI platform.

doa = DomainOfAnalysis(
    length=("relative_to_min", 0.102, 1.219),  # Length bounds
    width=("relative_to_min", 0.102, 1.219),  # Width bounds
    height=("relative_to_min", 0.134, 1.61),  # Height bounds
)

Create model configuration

Combine all the configuration elements into a ModelConfiguration object.

Build presets determine the training duration:

• debug: Quick training for testing.

• 1_day: Short training.

• 2_days: Standard training (recommended).

• 7_days: Maximum training for best accuracy.

mdl_conf = ModelConfiguration(
    project=project,  # Project containing the training data
    build_preset="1_day",  # Duration of the build
    build_on_top=False,  # Start training from scratch
    input=model_input,  # Model input configuration
    output=model_output,  # Model output configuration
    global_coefficients=global_coefficients,  # Global coefficient definitions
    domain_of_analysis=doa,  # Domain of analysis bounds
)

Verify and build the model

Before building, check if the project meets all requirements for training:

if project.is_trainable():
    print("Project is trainable. Starting model build...")
    new_model = simai_client.models.build(mdl_conf)
    print(f"Model build started with ID: {new_model.id}")
    print(
        "The model will take some time to train. You can monitor its progress in the SimAI platform."
    )
else:
    print("Project is not trainable. Please check the following:")
    print("- All training data must be processed successfully.")
    print("- The project must have enough training data.")
    print("- All required surfaces and boundary conditions must exist in the training data.")

Next steps

Once your model is trained, you can:

• Monitor the training progress in the SimAI platform.

• Run predictions on new geometries: Running Physics Predictions.

• Evaluate model performance using validation data.