When system identification methodologies are used to interpret measurement data taken from structures, uncertainty dependencies are in many cases unknown due to

This paper presents a new algorithm called PGSL-Probabilistic Global Search Lausanne. PGSL is founded on the assumption that optimal solutions can be identified through focusing search around sets of good solutions. Tests on benchmark problems having multi-parameter non-linear objective functions revealed that PGSL performs better than genetic algorithms and advanced algorithms for simulated annealing in 19 out of 23 cases studied. Furthermore as problem sizes increase, PGSL performs increasingly better than these other approaches. Empirical evidence of the convergence of PGSL is provided through its application to Lennard-Jones cluster optimisation problem. Finally, PGSL has already proved to be valuable for engineering tasks in areas of design, diagnosis and control

Structural Health Monitoring (SHM) has the potential to provide quantitative and reliable data on the real condition of structures, observe the evolution of their behaviour and detect degradation. This paper presents two methodologies for model-free data interpretation to identify and localize anomalous behaviour in civil engineering structures. Two statistical methods based on i) moving principal component analysis and ii) robust regression analysis are demonstrated to be useful for damage detection during continuous static monitoring of civil structures. The methodologies are tested on numerically simulated elements with sensors for a range of noise in measurements. A comparative study with other statistical analyses demonstrates superior performance of these methods for damage detection. Approaches for accommodating outliers and missing data, which are commonly encountered in structural health monitoring for civil structures, are also proposed. To ensure that the methodologies are scalable for complex structures with many sensors, a clustering algorithm groups sensors that have strong correlations between their measurements. Methodologies are then validated on two full-scale structures. The results show the ability of the methodology to identify abrupt permanent changes in behavior.

Civil engineering structures are difficult to model accurately and this challenge is compounded when structures are built in uncertain environments. As consequence, their real behavior is hard to predict; such difficulties have important effects on the reliability of damage detection. Such situations encourage the enhancement of traditional approximate structural assessments through in-service measurements and interpretation of monitoring data. While some proposals have recently been made, in general, no current methodology for detection of anomalous behavior from measurement data can be reliably applied to complex structures in practical situations. This paper presents two new methodologies for model-free data interpretation to identify and localize anomalous behavior in civil engineering structures. Two statistical methods i) moving principal component analysis and ii) moving correlation analysis have been demonstrated to be useful for damage detection during continuous static monitoring of civil structures. The algorithms are designed to learn characteristics of time series generated by sensor data during a period called the initialization phase where the structure is assumed to behave normally. This phase subsequently helps identify those behaviors which can be classified as anomalous. In this way the new methodologies can effectively identify anomalous behaviors without explicit (and costly) knowledge of structural characteristics such as geometry and models of behavior. The methodologies have been tested on numerically simulated elements with sensors at a range of damage severities. A comparative study with wavelet and other statistical analyses demonstrates superior performance for identifying the presence of damage.

Context: Model-based data-interpretation techniques are increasingly used to improve the knowledge of complex system behavior. Physics-based models that are identified using measurement data are generally used for extrapolation to predict system behavior under other actions. In order to obtain accurate and reliable extrapolations, model-parameter identification needs to be robust in terms of variations of systematic modeling uncertainty introduced when modeling complex systems. Approaches such as Bayesian inference are widely used for system identification. More recently, error-domain model falsification (EDMF) has been shown to be useful for situations where little information is available to define the probability density function (PDF) of modeling errors. Model falsification is a discrete population methodology that is particularly suited to knowledge intensive tasks in open worlds, where uncertainty cannot be precisely defined.Objective: This paper compares conventional uses of approaches such as Bayesian inference and EDMF in terms of parameter-identification robustness and extrapolation accuracy.Method: Using Bayesian inference, three scenarios of conventional assumptions related to inclusion of modeling errors are evaluated for several model classes of a simple beam. These scenarios are compared with results obtained using EDMF. Bayesian model class selection is used to study the benefit of posterior model averaging on the accuracy of extrapolations. Finally, ease of representation and modification of knowledge is illustrated using an example of a full-scale bridge.Results: This study shows that EDMF leads to robust identification and more accurate predictions than conventional applications of Bayesian inference in the presence of systematic uncertainty. These results are illustrated with a full-scale bridge. This example shows that the engineering knowledge necessary to perform parameter identification and remaining-fatigue-life predictions of a complex civil structure is easily represented by the EDMF methodology. Conclusion:Model classes describing complex systems should include two components: (1) unknown physical parameters that are identified using measurements; (2) R. Pasquier and I. F. C. Smith. Robust system identification and model predictions in the presence of systematic uncertainty, in Advanced Engineering Informatics, vol. 29, num. 4, p. 1096Informatics, vol. 29, num. 4, p. -1109Informatics, vol. 29, num. 4, p. , 2015Informatics, vol. 29, num. 4, p. . doi:10.1016Informatics, vol. 29, num. 4, p. /j.aei.2015 that cannot be represented only as uncertainties related to physical parameters. In order to obtain accurate predictions, both components need to be included in the model-class definition. This study indicates that Bayesian model class selection may lead to over-confidence in certain model classes, resulting in biased extrapolation.