Frequency Domain Model (In)Validation

ABSTRACT Classical validation methods accept or reject a model as a valid representation of a plant. However this quantization has several problems: i) Models are neither good nor bad but have a certain valid frequency range. ii) This gives no guess on the failure reasons or the way the model could be improved. iii) In iterative identification and control approaches undermodeling is usually present. This fact makes difficult to apply traditional model validation schemes. We present a new validation procedure that overtakes this problems by performing the validation in the frequency domain. Hence the validation/invalidation process is frequency dependent allowing to know for which frequencies the model is not validated. This is useful for control design purpouses (e.g. bandwidth selection) and identification issues such as experiment design and model order selection.