Modélisation de la propagation des incertitudes des mesures sur l'aube d'une turbine hydraulique par Krigeage et simulations stochastiques

Abstract

This project aims to estimate the propagation on uncertainties for experimental measurements. Kriging, as spatial interpolation method used in geostatistics, has been used to predict missing measurements and to condition stochastic simulations. Conditional stochastic simulations have succeeded in crossing the smoothing caused by kriging and give as a result mapping estimates of missing measurement point. Spatial interpolation of the experimental measurements goal is to adjust the result of calculations performed by the finite elements simulations (ÉFs1). In fact, the code of these simulations is used just to calculate the static load on the turbine blade; it is unable to quantify the dynamic loads. The objective is to predict the experimental measurements (static & dynamic) and the spatial propagation of uncertainties. The work is based on variogram analysis, from simulations results by ÉFs. The major hypothesis stipulates that the real measure has the same structure of spatial variability in the results of simulations ÉFs. We have succeeded to quantify the spatial uncertainty on the notched specimen and an area of a blade hydraulic turbine. We have succeeded also to predict all missing measurements from a sample of about 1% of field of study measures. Statistical tests confirm the validity of our results with adjusted correlation coefficients exceeding 90% compliance for both case studies. Finally, this approach has shown the capacity to study 2D and 3D areas. In order to pursue this research, sampling optimization could be done by analysing the number and locations of measurements and in another part; the Bayesian approach may be useful with that uses our results as primary data.

Date	24 Nov 2014
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Antoine Tahan (Supervisor) & Martin Gagnon (Co-supervisor)