Méthodologie de validation des modèles numériques utilisés lors de la prédiction des contraintes sur les aubes de turbines hydroélectriques

Abstract

The aim of the present research is to provide a consistent method for validation of numerical models used to estimate the mechanical stresses, particularly suited to hydroelectric turbines sector. Indeed, the simulation models are designed to reflect the physical reality of parts or complex systems in the most faithful way possible. When a mechanical or fluid model is proposed, it is necessary to quantify its uncertainty and compare with experimental data. Once the work done, it is possible to compare different models between them, and to validate or not for a given context, known a priori. To do this, different quantitative approaches are possible. However, in order to produce quality work, producing reliable results, it is essential to adopt a rigorous and robust process that leaves the least possible interpretation to the analyst. In this context, the use of the Area Metric is proposed to quantify the differences between simulation results and experimental measurements. This method has many advantages compared to classical methods of validation from traditional statistics which do not meet most of the recommendations of the ASME V&V. This is based on the distributions of simulation and experimental data to quantify the adequacy of the numerical model from the physical reality as measured by an experimental device which has itself uncertainty. We have to stress that the result of the application of this method makes no unequivocal conclusion about the validity of the model. The area metric only quantifies the level of adequacy. The validity criterion is determined by the analyst who will draw conclusions based on the level of adequacy. This method is applied on a complex model : a hydroelectric turbine wheel for which in situ measurements by strain gauges were available. Two different fluid models are thus compared. Moreover, an improvement of this method is proposed to overcome the limitations encountered to validate a model at an overall level, not just locally. This method can be applied to all kinds of problems when it is necessary to validate a numerical model. Indeed, the versatility of the area metric can be used to very heavy experimental measurements (eg. replications) or when they are few.

Date	30 Sept 2016
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Antoine Tahan (Supervisor) & Martin Gagnon (Co-supervisor)