Method version 1.0.0

Explicit models. Repeatable calculations.

The lab provides idealized effective-property models and tools for evaluating predictions you supply. These methods do not determine material properties, opening reductions, laminate strength or failure mechanisms from geometry.

Wall: four compliances in series

For a small-displacement cantilever under a horizontal point force at its top, the model adds panel bending, panel shear, base sliding and base rotation. The effective panel properties and base stiffnesses must be supplied and justified.

H is in mm; EI in kN mm²; GA in kN; Ks in kN/mm; and Kθ in kN mm/rad. GA includes the shear correction. Each term has units mm/kN. This model neglects second-order effects, opening stress concentrations, interaction between mechanisms and damage. Actuator displacement can contain test-rig movement not represented by wall-top displacement.

Beam: central point load

A simply supported beam carrying one point load at midspan uses the sum of bending and shear displacement. Effective flexural and shear rigidities must be consistent with the member and loading direction.

L is in mm; EI in kN mm²; and effective GA in kN. This expression does not apply unchanged to four-point bending, distributed loading or different supports.

Optional force cap

A user-specified positive force limit caps the predicted positive force. It is an idealized elastic-perfectly-plastic backbone, not a cyclic constitutive law. It does not model hysteresis, pinching, stiffness degradation, fracture or post-peak softening. A cap fitted to measured capacity must be reported as calibrated.

Comparing predictions

Import observations with specimen and campaign identifiers, measured outcomes and model predictions in a single consistent unit. Define calibration campaigns explicitly. Every observation from one calibration campaign stays in the calibration group; it cannot also enter held-out evaluation.

Signed error is predicted minus observed. Mean absolute error and root mean square error retain the outcome unit. Percentage errors use the absolute observed value as denominator; a zero observed value has no relative error. Missing predictions are reported through each model’s coverage. Compare coverage before comparing error scores. A held-out campaign label does not prove that its data were absent from prior model development.

Planning proposed tests

Candidate priority uses the spread between supplied model predictions divided by the absolute value of their arithmetic mean. Ranking describes disagreement, not accuracy or scientific importance. It uses no measured result for the proposed test and does not certify feasibility. Check candidate constraints and replication needs before choosing an experiment.

Source-linked packages and versions

Research packages contain the current public specimen snapshot, its campaign, source locators, licence and available reduced curve. Each snapshot has a SHA-256 content identifier. Save it locally to retain that version. The current download route can change as curation progresses. A data package with no curated model says so explicitly.

Contributions and access

Approved academic accounts can submit structured research records. The owner reviews submissions, and approval is separate from public release. Published contribution metadata is credited to its contributor; review status describes editorial handling and does not certify scientific validity.

Open the workspace · Download the Python reference runner · Browse the experimental sources