Solution Verification Linked to Model Validation, Reliability, and Confidence

Logan, R W ; Nitta, C K

Lawrence Livermore National Laboratory

关键词: Calibration;    99 General And Miscellaneous//Mathematics, Computing, And Information Science;    Documentation;    Implementation;    Sensitivity;   

DOI  :  10.2172/923117
RP-ID  :  UCRL-TR-207693
RP-ID  :  W-7405-ENG-48
RP-ID  :  923117

美国|英语

来源: UNT Digital Library

【 摘 要 】

The concepts of Verification and Validation (V&V) can be oversimplified in a succinct manner by saying that 'verification is doing things right' and 'validation is doing the right thing'. In the world of the Finite Element Method (FEM) and computational analysis, it is sometimes said that 'verification means solving the equations right' and 'validation means solving the right equations'. In other words, if one intends to give an answer to the equation '2+2=', then one must run the resulting code to assure that the answer '4' results. However, if the nature of the physics or engineering problem being addressed with this code is multiplicative rather than additive, then even though Verification may succeed (2+2=4 etc), Validation may fail because the equations coded are not those needed to address the real world (multiplicative) problem. We have previously provided a 4-step 'ABCD' quantitative implementation for a quantitative V&V process: (A) Plan the analyses and validation testing that may be needed along the way. Assure that the code[s] chosen have sufficient documentation of software quality and Code Verification (i.e., does 2+2=4?). Perform some calibration analyses and calibration based sensitivity studies (these are not validated sensitivities but are useful for planning purposes). Outline the data and validation analyses that will be needed to turn the calibrated model (and calibrated sensitivities) into validated quantities. (B) Solution Verification: For the system or component being modeled, quantify the uncertainty and error estimates due to spatial, temporal, and iterative discretization during solution. (C) Validation over the data domain: Perform a quantitative validation to provide confidence-bounded uncertainties on the quantity of interest over the domain of available data. (D) Predictive Adequacy: Extend the model validation process of 'C' out to the application domain of interest, which may be outside the domain of available data in one or more planes of multi-dimensional space. Part 'D' should provide the numerical information about the model and its predictive capability such that given a requirement, an adequacy assessment can be made to determine of more validation analyses or data are needed.