The aim of this study is to compare a new and some existing finite elementmodel error localization methods. The methods are applied to two problems. First,fundamental properties of the error localization methodsare studied on asimple sprung mass system. Inthe second problem a three-bay frame structure is studied.Here the analytical results of a finite element analysisis taken as substitute for measured data. The model differences between thismodel and a perturbed model are then found by use of error localization methods. When data from a known finite element model take place as substitute for test data, the cause of the differences between the data sets are known. A so-calledconsistent para meterization, i.e. a parameterization of the quantities known to be in error, is then possible. The error localization methods are compared for both consistent and inconsistent parameterization. A pre-error localization is made. It is based on the finiteelement model's properties. Candidatemodel parameters, possibly in error, for which the experimental data are not informative, are rejected. Non-identifiable parameters are also rejected. Quantification of data information richness and identifiability with newly developed index numbers support the pre-error localization.
Four error localization methods are compared. Two of these are developed by Lallement and Piranda. These are the so called Balancing of Eigenvalue Equation and Best Subspace Methods. The third is developed by Link and Santiago and is the Substructure Energy Function Method. A new localization method, using gradient and Hessian information of the error criterion function, constitute the fourth method.
Linderholt, former Larsson.