APPROXIMATION OF PARAMETER UNCERTAINTY IN NONLINEAR OPTIMIZATION-BASED PARAMETER-ESTIMATION SCHEMES

Theoretical models are often developed using experimental data to estimate the model parameters, but an evaluation is not always made of the model's reliability or robustness. Errors can exist in the theoretical model for many reasons, including inadequate model formulation and poor parameter estimates. This paper investigates the approximation of second-order statistics (confidence regions) of Parameters estimated using nonlinear optimization schemes. An optimal set of parameters is calculated to minimize the difference between theoretical model predictions and experimental data. This metric is formulated in terms of a weighted least squares objective function. A truss structure was analyzed to evaluate nonlinear optimization and confidence region approximation techniques. It was found that since the truss model was linear and parameter interaction was small, the confidence region approximations were reliable.