A method of quantitative validation based on model distortion

This method of model validation provides a numerical fidelity index, λ, for the model and also gives confidence ranges for predictions of control and failure transients. The paper describes the method in detail and describes specific examples drawn from the nuclear industry. The distortion method of model validation is based on analysing the differences between a rig or plant test record and the corresponding model prediction. An estimate is made of the distortion to the model that is necessary to make it follow the recorded transient precisely. This distortion is quantified in terms of parameter variations within the model and compared with the parameter variations that are acceptable from known approximations in the model. Hence, λ=(acceptable distortion)/(necessary distortion). When the fidelity criterion λ⩾1 is satisfied, the model distortion also provides a basis on which to estimate confidence limits for transient predictions. The examples given show this to be invaluable in assessing plant failure transients