Creep Analysis Verification & Validation
Verification and validation of creep analysis — constitutive model calibration, benchmark problems, comparison with test data, convergence verification, sensitivity analysis and model checking against service experience.
Verification vs validation
Verification and validation are distinct but complementary processes. Verification confirms that the FEA model is solving the equations correctly — that the constitutive model is properly implemented, the constants are correctly entered, the mesh and time stepping are adequate, and the numerical solution is accurate. Validation confirms that the model represents the real physical behaviour — that the constitutive model captures the relevant creep mechanisms, the material constants are correct for the service material, and the predicted deformation and life match the observed behaviour. Verification is a mathematical and numerical check; validation is a physical check. Both are necessary — a model that is verified but not validated may be solving the wrong equations correctly, and a model that is validated but not verified may be getting the right answer for the wrong reasons.
Constitutive model calibration and checking
The constitutive model calibration is the first step in both verification and validation. The model constants are fitted to the material test data (verification that the model can reproduce the data). The fitted model is then checked against data that was not used in the fitting (validation that the model predicts the material behaviour at other conditions). The calibration and the checking should be documented — the data used, the fitting method, the fitted constants, the residuals and the validation results. If the model does not reproduce the calibration data, the calibration or the model form is wrong. If the model does not predict the validation data, the model form is inadequate or the constants are not valid outside the calibration range. The constitutive model checking is described in detail in the creep model verification article.
Benchmark problems
Benchmark problems are standard cases with known solutions (analytical or from independent FEA codes) that are used to verify the FEA implementation. For creep, benchmark problems include: a thick cylinder under internal pressure with creep (the steady-state stress distribution is known analytically for a Norton law), a beam in bending with creep (the stress redistribution is known), a two-bar structure with different stresses (the load redistribution is known). The FEA should reproduce the benchmark solutions within a specified tolerance. The benchmark problems verify the creep model implementation, the time stepping, the mesh and the overall FEA setup. If the FEA does not match the benchmark, the source of the discrepancy should be identified and resolved. The benchmark problems should be documented as part of the verification.
Comparison with test data
The validation of the creep analysis against test data is the most important check. The test data may be from material tests (creep curves, rupture tests) or from component tests (a creep test of a full-size or model component). The material test comparison verifies the constitutive model at the material level. The component test comparison validates the full analysis chain — the constitutive model, the FEA, the boundary conditions and the geometry — at the component level. The comparison should be quantitative — the predicted strain, displacement or life should be compared to the measured values, and the difference should be within an acceptable tolerance. If the comparison is poor, the source of the discrepancy should be investigated (the material data, the boundary conditions, the mesh, the model form). The comparison and the discrepancy analysis should be documented.
Convergence verification
The convergence verification confirms that the FEA results are independent of the numerical parameters — the mesh, the time step and the iterative tolerance. The mesh convergence is checked by refining the mesh in the critical region and comparing the results. The time-step convergence is checked by reducing the time step and comparing the results. The iterative tolerance is checked by tightening the tolerance and comparing the results. If the results do not change with refinement, the solution is converged. The convergence verification should be documented with the baseline and the refined results and the tolerance. The convergence verification is described in detail in the convergence and stability article.
Sensitivity analysis and model checking
The sensitivity analysis (described in the companion article) identifies the dominant parameters and quantifies the uncertainty. The sensitivity analysis is part of the validation — it confirms that the model is not overly sensitive to parameters that should not dominate, and it identifies where the model is most uncertain. The model checking against service experience is the ultimate validation — if the predicted life or deformation is compared to the observed behaviour of similar components in service, the model is validated at the real-world level. The service experience comparison may be qualitative (the model predicts the correct failure location and mechanism) or quantitative (the model predicts the correct life within a factor of 2–3). The service experience comparison is the strongest validation but is only available for mature designs with service history. The verification and validation should be documented as part of the analysis, and the level of verification and validation achieved should be communicated with the results.
Verification and validation are both necessary for a defensible creep analysis. Verification confirms that the model is solving the equations correctly; validation confirms that the model represents the real behaviour. Document both — the benchmark results, the test data comparison, the convergence verification and the service experience check. A model that has not been verified and validated is not suitable for safety-critical decisions.
Implementation verification and regression benchmarks
Verification should demonstrate not only that the FE mesh and time step are adequate, but also that the selected constitutive law is implemented correctly in the actual software configuration being used. Small regression models are valuable for this purpose: a uniaxial constant-stress creep test, a relaxation test, a temperature change and, where relevant, a multiaxial or user-material benchmark. Expected responses can be generated analytically or from an independently checked reference calculation. These tests should be rerun when solver version, material subroutine or parameter set changes. Maintaining such 'golden' benchmarks provides configuration control and can detect implementation errors before they enter a complex component model. It also separates constitutive implementation verification from component-model verification, which are different questions.
Validation hierarchy and evidence matrix
Validation evidence should be organised by how directly it challenges the intended prediction. Material coupon data validate constitutive behaviour; feature tests can validate local redistribution or weld response; component tests and service measurements challenge the integrated thermal-structural model. No single comparison validates every aspect. A useful evidence matrix lists each important model assumption or output, the available validation evidence, its relevance and any residual gap. Where direct validation is impossible, the assessment should state which parts are verified only by benchmark or sensitivity and what conservatism is used to manage the gap. This is stronger than a generic statement that the model was 'validated against test' because it makes clear exactly what the evidence supports.