How do you validate and verify CAE models for multiphysics and multidisciplinary applications?

Different communities have different interpretations. I am particularly inclined to adopt the ASME definitions: Verification: The process of determining that a computational model accurately represents the underlying mathematical model and its solution. Validation: The process of determining the degree to which a model is an accurate representation of the real world from the perspective of the intended uses of the model. Consider that depending on the context, V&V will unfold differently. It is likely that a software engineer will spend more time (or all the time!) in verification, while the end-user of a commercial software will probably be more concerned about validation. An engineer using an in-house code will often look at both equally.

It’s realistic and confidence in the model. In my thoughts every model should be verified and validated to build more confidence in modeling methods and this ensures realistic results every time you build models. Best practices, testing methods, modeling methods ensures both to be achieved.

In my experience with CAE, validation and verification represent crucial facets of model integrity. Validation focuses on the "Accuracy" of predictions, aligning them with real-world data through iterative adjustments, ensuring the model's reliability. While verification can be used as a measure of "Precision", assessing how well the model generates consistent results despite changes in its features, such as sensitivity analysis based on mesh variations. Prioritizing both accuracy and precision is imperative before the model is deemed ready for operational use, as they collectively underpin the model's credibility and efficacy. At times, CAE engineers prioritize precision for consistent performance and robust design.

Generate point clouds by incorporating variability into the model by replacing all discrete parameters with a range and distribution and run 100 to 500+ runs randomly sampling the parameters. Conduct an equivalent runner of tests with instrumentation measuring parameters. Compare the model point clouds with the test point clouds. If the clouds overlap with the same shape and density, the model can be considered valid. Recommend starting with a model and test that match deterministically. This process was used to improve car safety through crash simulation. Number of crash tests have been reduced with crash tests used to validate models rather than explore design space. Models can then be run thousands of times look for outliers.

In validating CAE models for multiphysics and various applications, I've found that the approach varies with complexity. For simpler models, comparisons against analytical solutions or mathematical models suffice. Moderately complex models benefit from semi-analytical solutions or reliable experimental data. In cases where models are highly complex and conducting experiments is prohibitively expensive, a tiered approach is essential. Initially, validation may involve component-level testing, where each element is validated individually against experimental data or conventional numerical solutions. Once confidence is established at this level, system-level testing can be pursued, utilising available experimental models as reference points.

One means of verification is a similar method as described for validation. As long as the same physics is being modeled, different software tools can be used to generate point clouds that can be compared. A back of the envelope hand calculation should be done as an initial verification check.

Recommend starting with a box and arrow process chart showing the different analyses in the design process. Different analyses often require different parameters. How parameters are communicated and translated are key to integrating a multi-disciplinary math model. This step requires engineering expertise with the disciplines being integrated.