Performance Assessment and Prognosis for Civil Infrastructure Based on Model Falsification Reasoning

Infrastructure facilities are important to the economic growth and quality of life of modern societies. Much infrastructure has been built principally in the second half of the 20th century and several thousand structures are now attaining their design lives in Switzerland alone. Due to limited funds available for replacing all infrastructure that may have become weak, evaluating and maintaining existing infrastructure are inevitable tasks for structural engineers. Evaluation of infrastructure is often based on conservative models and simple code guidelines due to the lack of understanding of real behavior. When conservative provisions require either strengthening or replacement, less costly actions may be justified when measurements are performed. However, even if assessments are accurate at measured locations, this information cannot generally be extrapolated to predict, for example, at other locations and for other loads. Structural mechanics models are required to predict at locations where measurements are not available. In order to provide reliable performance assessment and prognosis, model-based data-interpretation techniques that include modeling and measurement uncertainties are needed to link structural response to properties using information acquired through measurement. This activity is called structural identification. Model-based data interpretation for complex structures is an ill-defined inverse task that is carried out in open-world conditions. Performance assessments in these contexts lead to multiple possibilities for models of real behavior. The number of possibilities may be reduced by acquiring additional knowledge of the structural behavior. Knowledge is acquired gradually by new information obtained using data-interpretation methods. Through these methods, engineers test their hypotheses with in-situ observations, including measurement data. The experience and judgment of engineers are of utmost importance. This thesis proposes an iterative structural identification framework that guides engineers in the process of knowledge acquisition and data interpretation in order to provide good performance assessments and prognoses. Six tasks are necessary for supporting engineers; modeling, in-situ inspection, monitoring, model falsification, performance assessment and prognosis. Such tasks are carried out through a process of model-class and model-instance population generation and falsification. In addition, two complementary population-based prognosis methodologies for the evaluation of fatigue reserve capacity are proposed that outperform current conservative estimations. Finally, a new metric based on expected utility that assists measurement-system design is proposed. This metric is able to quantify the usefulness of monitoring actions and estimate points where overinstrumentation is likely. These proposals are tested and verified on five cases, four full-scale structures and one simple beam. The framework is robust to systematic modeling uncertainties and incomplete knowledge of error interdependencies between measurement locations. Multi-path iterations of activities are compatible with the way knowledge evolves as evaluation proceeds. This leads to good support for decision-making related to many aspects of structural health management.

Smith, Ian
Lausanne, EPFL
Other identifiers:
urn: urn:nbn:ch:bel-epfl-thesis6756-6

Note: The status of this file is: EPFL only

 Record created 2015-08-18, last modified 2018-03-17

Download fulltext

Rate this document:

Rate this document:
(Not yet reviewed)