Calibration Workflow

How to enter data, choose an objective and solver, run calibration, and interpret diagnostics.

Click Calibration in the top menu bar to open the selected material’s data and diagnostics. The window can be moved, resized, or minimised while charts and menus remain usable. Closing it retains your data and settings, and a running fit continues. Reopen Calibration to see its progress; use Cancel to stop the fit.

Data Type

Mode	Columns	Notes
Minor/Major Principal Stress	sigma3, sigma1	Available except for Barton-Bandis.
Normal/Shear Stress	sigma_n, tau	Available for all criteria; required for Barton-Bandis.

Target Parameters

Criterion	Calibratable Parameters
Mohr-Coulomb	cohesion, frictionAngle; tensile strength remains fixed.
Generalized Hoek-Brown	sigmaCi, GSI, mi, disturbanceFactor.
Power Curve	a, b, c, d.
Barton-Bandis	phiR, JRC, JCS; normal/shear data only.

Table Operations

Paste tab-separated or comma-separated spreadsheet values; decimal and scientific notation retain invalid drafts for correction.
Select cells by dragging or Shift-clicking. Row numbers select rows; numeric headers select columns. Use arrow keys, Tab, Enter, Cut, Copy, Paste, Delete and Select All as in the JStereonet orientation table.
Drag a numeric column's edge to resize it. Keyboard navigation and clipboard operations work across virtualized rows.
Ctrl+Home and Ctrl+End jump to the first and last cells. Up/Down and Enter move between rows; Left/Right continue to edit the cursor position within a text cell. Delete clears a selected range.
Show on plots, the leftmost toggle, controls visibility on workspace and Calibration stress plots. Its header shows or hides all rows. This setting is saved with the project and supports Undo/Redo.
Fit independently includes or excludes a measurement from calibration. Hidden points can still be fitted; visible excluded points use hollow markers.
Use the footer or row context menu to insert and delete rows; the footer also moves selected rows. Deletion leaves at least one row. Clear Data resets the table after confirmation.
Each material supports up to 50,000 rows. The table expands with the available window height. Long tables scroll internally beneath fixed headers and footer actions.

Running Calibration

Calibration requires at least one more included, valid row than the number of selected target parameters. All fitting runs in a background worker. Progress is reported by phase, duplicate runs are disabled, and Cancel stops and recreates the worker. Successful completion reports the actual solver, iterations, fitted parameters, raw RMSE, and any exact lower or upper bounds reached. A stopped fit distinguishes an iteration limit, an evaluation limit, or an objective that could not become finite; no parameters are applied in those failure cases.

Successful runs create a preview without changing material parameters, workspace curves or Results. Review the two stress plots on the right, then click Accept to apply the fitted values in one undoable change. The window stays open. Both plots show strength curves; measured points appear only on the plot matching the entered data type.

Preview curves automatically extend as needed to leave approximately 5% margin past the largest visible test input, within the criterion's valid domain. Larger configured ranges are retained. Visible points excluded from Fit still determine this extent; hidden points do not. This also applies before fitting and after Accept, without changing the material's saved failure range or workspace plot limits.

Unaccepted previews remain available when closing/reopening Calibration or switching materials during the current project session. They are not saved in project files or autosave. Editing fit inputs invalidates the preview; changing point visibility or material name/colour does not. Starting another run replaces the previous preview. Closing a running fit does not accept its result.

Advanced Calibration

New sessions default to Auto, absolute residuals, and L2 loss. The Advanced settings… popup exposes the following explicit alternatives. Settings are saved separately for each material.

Setting	Definition and Behaviour
Auto	Uses five deterministic starts. Smooth L2 problems use bounded Levenberg-Marquardt; L1, Huber, and sampled/interpolated modes use pattern search.
Levenberg-Marquardt	Finite-difference, normalized bounded fit for L2 loss only. Selecting a non-L2 loss switches the solver visibly back to Auto.
Pattern search	Derivative-free bounded search suitable for L1, Huber, and sampled/interpolated predictions.
Differential evolution	Deterministic bounded global population search followed by local polishing. It uses more evaluations but is less dependent on the starting parameters.
Absolute residual	Predicted minus measured, in stress units. This is the default.
Relative residual	(predicted minus measured) divided by max(abs(measured), floor). A positive floor is required and is never inserted automatically.
Log-ratio residual	Natural log of predicted divided by measured. Every included measured value and every accepted prediction must be positive.
L2 / L1 / Huber	Squared, absolute, or standard Huber loss. Auto Huber threshold is 1.345 × 1.4826 × MAD from a preliminary L2 fit, with RMSE used when MAD is zero.

JLab minimizes mean loss. Mean and sum have the same optimum for a fixed set of rows, while the mean is easier to compare between datasets with different row counts. Changing residual scale resets a custom Huber threshold to Auto because the threshold units change.

Expand Calibration summary and residuals below the preview charts for fitted parameters, solver details, predicted-versus-measured and residual-versus-input plots, and row-level values. The disclosure starts closed to keep the normal laptop layout free of outer vertical scrolling. The Raw/Objective selector switches the residual view. Raw metrics include RMSE, MAE, median and maximum absolute error, and R². Excluded valid rows receive predictions and residuals but do not affect the objective or raw metrics. Relevant input, target, solver, objective, mode, data, or exclusion changes invalidate diagnostics until calibration is run again.

Criterion	Parameter Bounds Used During Fit
Mohr-Coulomb	cohesion 0-500 MPa; friction angle 0-89 deg.
Hoek-Brown	sigmaCi 1-500 MPa; GSI 5-100; mi 0.1-50; D 0-1.
Power Curve	a 0-500; b 0.05-5; c 0-500 MPa; d 0-500 MPa.
Barton-Bandis	phiR 0-70 deg; JRC 0-20; JCS 0.001-1000 MPa.