Model Correlation and Calibration
暂无分享,去创建一个
Vector of predicted changes in parameters to update the model Δf Vector of differences between test frequencies and model frequencies COVΔp Covariance matrix for parameters m Number of frequencies n Number of parameters ABSTRACT Analytic model validation assesses the usefulness of a model for its intended purpose. Validation of the model should be based on a blind prediction of test results, so that the predictive capability is demonstrated. However, there is certainly a place in the validation process for exercising the model in correlation and calibration before making the final blind validation prediction. In many cases, the initial model deviates from a useful state for unknown or unquantified reasons. Then model correlation exercises are performed. The term "correlation" for structural dynamics comes from the initial one to one correlation of the modes of the model with modes from a modal test. But correlation really includes more than just this initial comparison. Correlation exercises can uncover unintended errors or incorrect assumptions and simplifications in the thousands of details important to model development. Calibration, on the other hand, is designed to improve estimates on specific uncertain parameters. Sensitivity analysis is examined as a method for calibration. Sensitivity analysis is an inferential process that has definite limits, and three tools for understanding those limits are provided. In some cases, calibration does not provide sufficient model improvement, and additional correlation exercises are performed after the calibration. Often correlation and calibration are most effective when applied to subsystems of the model to isolate specific errors.
[1] Thomas L. Paez,et al. Validation of mathematical models : an overview of the process. , 2005 .
[2] Randall L. Mayes,et al. A tool to identify parameter errors in finite element models , 1997 .
[3] J. D. Collins,et al. Statistical Identification of Structures , 1973 .