Analyse statistique des défauts trouvés dans les joints soudés multicouches utilisés dans la fabrication de turbines hydroélectriques

Abstract

The evaluation of the fatigue resistance of the weld material in the hydroelectric turbine runners is carried out on test specimens. The question whether the welded joints and specimens used for testing can exhibit approximately similar fatigue strengths arises, since the processes are different. The joint is the result of multilayer welding with an addition of 410NiMo metal. The specimens are machined from a raw weld metal deposited in several layers on a substrate. This thesis suggests a method for comparing the materials of these two processes, with a view to verifying whether the two treatments have similar qualities. The proposed method is based on the best approximation models of the sizes of extreme defects found in materials. An experimental methodology by X-ray imaging for each case was implemented to investigate and quantify the defects. Analysis of 2D X-ray images of the wafers was used to quantify the defects in the material of the T-joint, due to the fact that these dimensions are relatively large and its shape has sharp edges complicating the 3D analysis. For the cylindrically shaped fatigue specimens, tomography produced 3D images of the specimens which were projected as 2D images for quantification of defects. The statistical analysis of the square root values of the projected areas of the collected pores was based on the theory of extreme values. Generalized extreme value and generalized Pareto distribution models were jointly explored. Three criteria were used to select a model to simulate the extreme values in each case. The winning model should be validated by statistical tests of Kolmogorov-Smirnov and Anderson-Darling, and judged to perform well by calculations of the square root of the mean squared errors and the coefficient of determination. Then, this one will have to carry out a very good approximation of the most extreme experimental values and finally to minimize the confidence intervals of the predictions. The high threshold GPD models at 91% of the data for the T-joint and GPD at 88.1% for the fatigue specimens were retained. The main conclusion of this study is that the two processes could be statically close considering the size of the large pores when the volume of solder material is less than 200 μm. On the other hand, the difference would be more significant for larger volumes.

Date	20 Aug 2021
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Philippe Bocher (Supervisor)