Improving Knowledge of Structural System Behavior through Multiple Models

A system identification and model updating methodology that accounts for factors influencing the reliability of identification is proposed. An important aspect of this methodology is the generation of a population of candidate models. This paper presents an analysis of error sources that are used to define model populations. A case study illustrates the need for such an approach even when a single conservative model has been appropriate for design. Data mining techniques such as principal component analysis and k-means clustering combined to interpret model predictions. These methods are useful for estimating the dependability of system identification.