Integration et optimisation multidisciplinaire d'une turbomachine au stade preliminaire de conception.

摘要

This thesis presents the integration and optimization of an advanced turbomachinery design. A strategy proposes a three-step optimization procedure to cope with such a complex multi-disciplinary problem. This method is based on a sensitivity analysis on variables for selection and a specific process on the multi-disciplinary analyses couplings.; The sensitivity analyses allow to select input variables from multi-disciplinary analyses. The purpose is to establish (i) a ranked list of variables according to their relative influence on the outputs, and (ii) a relevant selection of these variables, to be incorporated into the global strategy. The proposed sensitivity analysis method is based on latin-hypercube sampling and a response surface model. This model is a second-order approximation of the sampled input and output variables obtained by a least-square fitting method. The sensitivity of an output variable to an input variable is obtained from the weights of the model. For the ranking and the selection of the variables, two criteria have been established. The first criterion is a threshold to decide if the variable has enough influence or not. The second defines the simplification objective: a maximum number of variables to be managed per subset. Then, some input variables are selected for each considered output variable. To take all output variables into account, lists are compared between the outputs to built a final list for each of the three optimization steps.; The coupling processing is made by conducting analyses on specific links to temporarily suspend them from the multi-disciplinary analysis (MDA), without any relative comparison to other links.; The study shows that the sensitivity analysis performs a relevant selection of influent variables. Indeed, an optimization that uses only the most important variables, enables a global weight reduction comparable with a classic optimization process that uses all the variables, while maintaining the engine performance. The proposed method for coupling processing did not help to suspend the most problematic links. Finally, the proposed strategy performs well on a problem that has specific characteristics. The PMDO problem from Pratt & Whitney Canada does not perfectly fit with these characteristics. However, some optimizations that uses the proposed sensitivity analysis, are able to find turbomachine geometries that reduce the overall weight within a very competitive timescale. (Abstract shortened by UMI.)