||Decision support through multiple models and probabilistic search
||Robert-Nicoud Y, Raphael B, Smith I
||"A large number of candidate behaviour models may exist for existing civil engineering structures such as bridges. Finding the right model for explaining a given set of observations is a difficult task. Traditionally, modelling assumptions are made without adequate justifications and verifications. Manually constructing multiple models and comparing their behaviour with measurements is arduous and hence, we are developing support tools for engineers. Techniques of model composition and model reuse are used for systematically constructing and evaluating multiple models. In this paper, experiments in model construction for the Lutrive bridge in Switzerland and their results are reported. The Lutrive bridge was constructed in 1972 using the cantilever method with central hinges. It is found that the bridge continues to creep significantly even after twenty-seven years of construction. However, load tests indicate that the bridge possesses unusually high rigidity and earlier theoretical models (constructed manually) gave results that were different from displacement measurements by as much as 100%. It was necessary to evaluate different modelling possibilities in order to obtain reasonable correlation with measurement data. The approach we have used is summarised in the following steps: ? Cases consisting of candidate models are constructed manually.? Spaces of behaviour represented by each case are defined.? The total solution space which is a union of the spaces represented by individual cases is searched using a new probability-based algorithm.The following conclusions were drawn from the above approach:? If a model contains enough number of parameters as observation points it may be possible to get an exact match by tuning values of parameters. However, certain models are capable of representing only a few modes of behaviour and no combination of parameter values might exist that explain the observed behaviour. ? Model creation in a domain such as structural engineering requires considerable amount of skill. Storing cases consisting of complete models is a means of reusing such expertise. ? Model construction is a search problem. Evolutionary and stochastic search techniques produce good results when used in combination with cases."
|Year of publication:
Robert-Nicoud Y, Raphael B, Smith I (2000).
Decision support through multiple models and probabilistic search. Construction Information Technology 2000. Taking the construction industry into the 21st century.; ISBN 9979-9174-3-1; Reykjavik, Iceland, June 28 - 30, 2000 (ISSN: 2706-6568),