ADS-79D-HDBK Downloaded from https://www.everyspec.com In general, the validation schemes can be classed into three categories: 1) Complete overlap, 2) Partial overlap, and 3) No overlap. The first category represents a vast majority of validation cases for most engineering issues. This validation scheme engages a broad validation domain which covers the entire application domain. It is the prevalent situation in validation since the entire region of application has been validated and additional validation data beyond the application domain provides the means for future extension of applicability of the computational model.
Sometimes, extensive validations covering the entire application domain may not always be feasible. As depicted in Figure K-4 (b), the domain of validation addresses only a portion of the application domain. This validation scheme represents an engineering situation where a comprehensive validation is infeasible or prohibitive and therefore the focus of validation is on the application region of primary interest. In the third validation scheme, shown in Figure K-4 (c), there is no overlap between validation domain and application domain. This scenario highlights a rare engineering situation in which direct validation of the applicability of a model is not possible.
As the only alternate, indirect validations are conducted to confirm the basic assumptions of the model and inference from the validation domain is made using both physics based models and statistical methods. Although the aforementioned validation scheme may seem to lack rigor, it may be the only option for validation of some complicated engineering issues. A typical example of using the inference-based approach to validate a predictive model is to validate an estimate of probability of failure or quantitative risk assessment. In the aircraft industry, highly reliable components are expected. Therefore, direct evidence for extremely small probability failures are very difficult to obtain.
To validate probabilistic prediction for a specific component, indirect evidence such as limited fielded data for an entire fleet as well as data for similar components are collected and pooled to form the basis for further statistical inference. Once the statistical confidence of the pooled database is shown to demonstrate similar behavior, validation of the probabilistic prediction can be made accordingly. As discussed by Oberkampf, et.al the need to perform extrapolation beyond a model‘s validation domain reinforces the need for models to be judged on the basis of achieving the right answers for the right reasons in the intended validation regime.
65 Validation, as one of the vital steps to qualify Maintenance Credits, is generally carried out through physical experiments. A variety of validation methodologies exist. Validation testing can be performed by either physical or numerical experiments. Physical experiments are an ideal way to conduct validation testing. Various forms of physical experiments have been used for validation, including sub-component/component test, sub-system test, seeded fault test, and fielded system level test. Each of these forms has its own advantages and disadvantages.
Selection of a specific form of physical experiment depends on the intended scope of study, requirements of model accuracy, criticality of potential failure mode under the investigation, budget and schedule of the validation program. Clearly, performing fielded system level test increases the complexity and model fidelity. But this also decreases feasibility of validation and causes potential delay of implementation.
If feasible and affordable, physical experiments offer great opportunities to generate direct evidence to compare simulation prediction against physical 65 SAND 2003 – 3769, Verification, Validation, and Predictive Capability in Computational Engineering and Physics, Sandia National Laboratories, 2003. 197