CBL - Campus del Baix Llobregat

Projecte llegit

Títol: Models substitutius per a modelat de dades aeronàutiques

Estudiants que han llegit aquest projecte:

Director/a: MORA SERRANO, FRANCISCO JAVIER

Departament: DECA

Títol: Models substitutius per a modelat de dades aeronàutiques

Data inici oferta: 23-06-2015     Data finalització oferta: 23-02-2016


Estudis d'assignació del projecte:
    Tipus: Individual
     
    Lloc de realització:
    UPC
        Departament: CIMNE
     
            Supervisor/a extern: Dr. Jordi Pons
            Institució/Empresa: CIMNE
            Titulació del Director/a: Dr. Enginyer Industrial
     
    Paraules clau:
    xarxes neuronals, Kriging, models d'aproximació, turbopropulsors, aeronàutica
     
    Descripció del contingut i pla d'activitats:
    El pla de treball és:
    El projecte pretén estudiar com aproximar un conjunt de dades
    per obtenir-ne una funció d'aproximació que faciliti el càlcul
    de futurs escenaris. El treball es dividirà en les següents
    parts:
    - Estudi de l'estat de l'art en models basats en Xarxes
    Neuronals, i models de Kriging. L'objectiu és entendre com els
    dos models ajusten les dades per poder proveir de valors
    aproximats per a qualsevol escenari.
    - Generació del model basat en Xarxes Neuronals, i si és
    factible també generar el model basat en Kriging.
    - Aplicació a un cas amb dades reals per al predisseny de
    motors d'aviació i dades per el càlcul de les toleràncies de
    fabricació dels àleps dels turbopropulsors. Buscar informació
    sobre aquestes dues aplicacions .
    - Comparativa entre els dos models i treure'n conclusions.

     
    Overview (resum en anglès):
    To make a simulation can take some hours or even days, therefore, is more efficient to design and execute a limited number of computer approximations to know the behaviour of the data and be able to get the rest of the results that have not been calculated in the previous simulation. This way is much faster, although not as accurate.

    The data approximation methods use a set of known values from some points to interpolate other points located in the same area of study, which value is not known.

    The interpolation method by kriging to predict the response at any point of the design is suitable for this study because it’s very accurate, also it allows to calculate the error, even though it requires some previous knowledge of the data to be treated, does not require very complex calculations.

    Taking all these ideas present, in this work, we have studied a few data to obtain an approximation function that facilitates the calculation of future research. Understanding how to adjust the data to calculate approximate values for any situation.

    We have created a model for computer executions that performs calculations with kriging interpolation for spherical, exponential and Gaussian semivariogram models, and it shows the results numerically and visually with some diagrams.

    After performing different tests with different data obtained from some fields such as geology or medicine, it has been proved that the model makes the process very balanced in terms of accuracy of the results and the time of calculation used.

    Finally, we have applied to a case with real data for the preliminary design of aircraft engines, specifically to calculate the manufacturing tolerances of the blades of the turboprop.

    At last, we have named future research lines with this study and the program created as a starting point.

    © CBLTIC Campus del Baix Llobregat - UPC