Calibrate test data

When you have selected the data and the model, you must calibrate to create the fitted curves.

To fit test data to the models you have selected:
  1. Enter coefficient details for each model in the table to the left of the plot area.
    1. Optional: Edit the Value column to set the initial value of a coefficient.
    2. Enter values into both the Upper bound and Lower bound columns to apply constraints to a coefficient.
    3. Select the check box in the Fixed column to convert a coefficient into a constant with value = Value.
    Note:
    • If you want to use the Levenburg-Marquardt fitting algorithm, the Upper bound and Lower bound for each coefficient are optional.
    • If you want to use the Genetic Algorithm, you must enter an Upper bound and Lower bound for all coefficients.
  2. Click Calibrate.
  3. For each model you want to calibrate:
    1. Ensure the Calibrate? check box is selected.
    2. Select a fitting algorithm from the Algorithm Type list.
    3. Optional: Edit the fitting parameters associated with that algorithm, if required. The default values provided by the Ansys Material Calibration Service are as follows:
      Algorithm Type Defaults
      Levenburg-Marquardt
      • Number of Iterations = 50
      • Residual Change Tolerance = 0.0
      • Coefficient Change Tolerance = 0.0
      • Least Squared Error Norm = Normalized
      Genetic Algorithm
      • Number of Generations = 50
      • Population Size = 300
  4. Click OK to begin calibration.
Your test data and model parameters are sent to the Ansys Material Calibration Service. When the fitted curves are returned, they are displayed in the Plot area in the middle of the application workspace, and the fitted coefficient values are displayed in the Models panel to the left of the plot (in place of their initial values). Each fit is associated with a residual, which is reported alongside the fitted coefficients.
Note: The residual, calculated using the least squares method, quantifies the error between the input test data and the fitted curve. The fitting algorithm iteratively refines the model coefficients to minimize this error. A residual close to zero indicates an excellent fit, while a large positive value suggests a poor fit.

After calibrating your data, you can fit the data again using the current coefficients as initial values (this will overwrite the current values with the new results).