About PEPITO

Recent advances in simulation (hardware and software) make its use more accessible, and engineering departments have changed from scarce and costly expert simulation, to more accessible engineering calculations carried out daily. Paradoxically, the proliferation of data and results dramatically increases the task of those who must use them and optimize them: the complexity is shifted from mastery of the simulation process to the optimization methodology for complete systems. The questions that now arise to an engineer are related to the factors in its study. Are the selected parameters relevant, does he have to consider interactions, or are the ranges large enough? Is the model accurate and reliable? Ultimately, how far does he have the means to take optimization?

The PEPITO project aims to experiment with a drastic approach using multi-physics and highly intensive computations, parameterized geometries and simulations, and design of experiments with a large number of factors. It also aims to seek for optima in a large dimension domain. This contribution to the development of sciences that treat and fully exploit the big data stream from intensive simulations will provide tools for the evaluation of parametric effects, for the exploration of large domains of solutions and for dealing with robustness and / or reliability.

The optimization process will concern turbomachinery cases proposed by industrial partners. It will rely on the construction of response surfaces (with uncertainty assessments) and on multi-objective research techniques, with or without constraints, in very large domains (up to 60 parameters).

Innovative methods are required for design of experiments, sensitivity analysis, dimension reduction, meta-modeling by kriging, co-kriging and neural networks, and in general terms for numerical experiments when they are costly, time consuming (in months) and require the use of high-performance computers.

Developments will be made in the field of (multiphysics) performance prediction for rotating machines. The problem of accuracy and reliability for simulations should be carefully studied, especially for irregular geometries based a set of independent parameters. CFD results will be post-processed for acoustic analytical models developed to assess source intensities. Finally, advances are expected in the calculation of eigenmodes in rotating domains, and fluid interaction will be taken into account for mechanics and vibro-acoustics.

Intensive simulation strategies will be explored by using fully automated processes and by seeking the optimum linking mesh, solver, parallelization and machine architecture (2.000.000 CPU hours on 125 processors for 2 years). Innovative techniques using parameterized simulations will be studied, and cost reductions will be assessed during the construction of databases with derivatives that contain added information.

Implementation of results from the previous fields of research are intended to demonstrate their complementarily for multiphysics optimization in large dimension domains for turbomachines.

A summary of academic work and feedback between statisticians and physicists will be proposed at the end of the project, and the industrial added value methods of simulation and optimization will also be evaluated in terms of cost, quality and time.