PHIMECA (France)

Abstract. Uncertainty estimation is a key issue in nuclear crisis situations. Probabilistic methods for taking uncertainties into account in assessments are often costly in terms of the number of simulations and computation time. This is why emulation methods, which enable rapid estimation of numerical model outputs, represent a promising solution. However, in the context of radioactive dispersion modeling, existing emulators are mostly limited to scalar outputs. In a crisis context, decisions are often based on dose maps, which are mathematically represented by high-dimensional data. In this study, we use the auto-associative model method to reduce the dimension of dose results and then predict these reduced representations using kriging. We also compare this prediction method with others used by the French Nuclear Safety and Radiation Protection Authority (ASNR) to predict the consequences of a nuclear accident.

International audience

Metrology is extensively used in the manufacturing industry to determine whether the dimensions of parts are within their tolerance interval. However, errors cannot be avoided. If the metrology experts are actually aware of it, and currently able to identify the different sources that contribute to making errors, very few research has been made in this area to develop metrology methods accounting for such errors. The probability density function of the error is here assumed to be given as an input. This work deals with a batch of measures and its statistical properties. The first proposed method aims to correct the effects of the errors to the distribution that characterize the entire batch. Then a second method tries to estimate for each single measure, the dimension that is being the most likely given by a measure, after the error is deducted. It is based on the output knowledge of the first method and integrates it with Bayesian statistics. Only Gaussian distributions are considered in the paper. Their relevance is shown through one example applied on simulated data.

In this paper, damage indicators are presented for bridges based on measured data collected under uncontrolled traffic conditions. The test bridge considered is a small stiff concrete culvert with low dynamics, making it suboptimal for monitoring with accelerometers. Largely static strain responses are measured, with the variability in the vehicle weights and configurations addressed by handling the data in statistical terms. Continuously recorded temperature data are used in an alternative and innovative way to validate damage indicators. Temperature change causes stiffness change in concrete structures. It is posited that temperature can be used as a proxy for bridge damage as, while the effects are clearly different, both cause changes in flexural rigidity and therefore in bridge response to load. It is demonstrated here, using field data, that the damage indicators are sensitive to temperature change. Strain and temperature data for a 2-year period are obtained for a culvert bridge in Slovenia. The defined damage indicators are calculated for the strain data and validated using the simultaneously recorded temperature data. Two ways of using the proposed damage indicators together with the information about temperature for bridge structural health monitoring (SHM) are also presented: one is including the mean and standard deviation of the damage indicators and a compensation for the temperature effect and another to obtain directly the modified Z score, which is used for identifying outliers, and draw a 2D risk map having damage indicator and temperature on the two axes.

Most advanced mathematical methods for the analysis of numerical model cannot cope with functional outputs of dynamic Modelica models. Principal component analysis is a well established method for dimension reduction, and can be used to tackle this issue. It relies however on a linear hypothesis that limits its applicability. We illustrate on a case study how the non linear method of autoassociative model overcomes this shortcoming and provides physically interpretable data representations.

This paper presents a bilinear two-dimensional probabilistic fracture mechanics model for fatigue crack growth in welded steel details. The model takes besides membrane stresses also the influence of misalignments and bending stresses in the welded detail into account. The distribution of the membrane stresses are estimated using a generic bridge structure and traffic measurements. The optimal reliability level for a welded detail in a bridge subjected to fatigue are estimated by cost benefit-analysis taking into account the risk of human lives through the Life Quality Index. Since the optimal reliability level is higher than the estimated reliability level, it is investigated how planned and reliability-based inspections techniques can be used to update the reliability level.