Improving fatigue evaluations of structures using in-service behavior measurement data

Conservative models and code practices are usually employed for fatigue-damage predictions of existing structures. Direct inservice behavior measurements are able to provide more accurate estimations of remaining-fatigue-life predictions. However, these estimations are often accurate only for measured locations and measured load conditions. Behavior models are necessary for exploiting information given by measurements and predicting the fatigue damage at all critical locations and for other load cases. Model-prediction accuracy can be improved using system identification techniques where the properties of structures are inferred using behavior measurements. Building upon recent developments in system identification where both model and measurement uncertainties are considered, this paper presents a new data-interpretation framework for reducing uncertainties related to prediction of fatigue life. An initial experimental investigation confirms that, compared with traditional engineering approaches, the methodology provides a safe and more realistic estimation of the fatigue reserve capacity. A second application on a full-scale bridge also confirms that using load-test data reduces the uncertainty related to remaining-fatigue-life predictions.

Uncontrolled Keywords

Remaining fatigue life; Model-based data interpretation; Population of models; Uncertainty; Conservatism