Code Monkey home page Code Monkey logo

dd-fm's Introduction

Data Driven Fluid Mechanics (Notas de estudo)

Capítulo 1

Cylinder Wake: Benchmark para Navier Stokes Incompressível

$\nabla \cdot \mathbf{u} = 0$

$\partial_t \mathbf{u} + \mathbf{u} \cdot \nabla \mathbf{u} = -\nabla p + \frac{1}{Re} \Delta \mathbf{u}$

$\Omega_{DNS} = \{ (x,y) \in \mathbb{R}: x^2 + y^2 \geq 1/4 \cap -10 \leq x \leq 40 \cap |y| \leq 10 \} $

Mean Field Theory 🔍

$\mathbf{u} = \mathbf{u_D} + \mathbf{u'}$

$\mathbf{u_D}(\mathbf{x}, t) = \mathbf{u_S}(\mathbf{x}) + a_{\Delta}(t)\mathbf{u}_{\Delta}(\mathbf{x})$

Shift mode $\mathbf{u}_{\Delta}$: $\mathbf{u}_{\Delta} = (\mathbf{u}_0 - \mathbf{u_S}) / \lVert \mathbf{u}_0 - \mathbf{u_S} \rVert_{\Omega}$

Amplitude $a_{\Delta}$: $a_{\Delta} = (\mathbf{u}-\mathbf{u_S}, \mathbf{u}_{\Delta})$

Fluctuation Energy $\mathcal{K}$: $\mathcal{K} = \lVert \mathbf{u'} \rVert_{\Omega}^{2}/2$

Feature Extraction 🔍

  • Proximity Maps with Classical Multidimensional Scaling (CMDS) $\gamma^m \in \mathbb{R}^2$: Transforma features de alta dimensão em duas dimensões (Parecido com LLE)

    $\displaystyle E = \sum^{M}_{m=1} \sum^{M}_{n=1} [ \lVert \mathbf{u}^m - \mathbf{u}^n \rVert - \lVert \mathbf{\gamma}^m - \mathbf{\gamma}^n \rVert ]^2$

    Para remover efeito translacional (centralizar o mapa):

    $\displaystyle \sum^{M}_{m=1} \mathbf{\gamma}^m = 0$

    Para remover efeito rotacional $\mathbf{\gamma}^0$ é tomado como o máximo

  • Manifold Extraction with Local Linear Embedding (LLE) $\gamma^m \in \mathbb{R}^N$: Aproxima os snapshots por combinações lineares dos $K$ vizinhos próximos criano um manifold aproximado de menor dimensão

    $\displaystyle \mathbf{u}^m \approx \sum^{K}_{k=1} w_{mk} \mathbf{u}^{i_{k}^{m}}$, e consequentemente $\displaystyle \mathbf{\gamma}^m \approx \sum^{K}_{k=1} w_{mk} \mathbf{\gamma}^{i_{k}^{m}}$

    Com $i_{k}^{m}$ sendo os indices dos $K$ vizinhos próximos e $w_{mk}$ pesos não negativos e o número de features $N$ escolhido até a convergência para o erro fixado

Low-Dimensional Representations (Autoencoder): Transformação do espaço de $M$ snapshots para um espaço reduzido de dimensão $N$

  • Encoder $G(\mathbf{u}^{m})$: $\mathbf{u}^{m} \mapsto \mathbf{a}^{m} := G(\mathbf{u}^{m})$
  • Decoder $H(\mathbf{a}^{m})$: $\mathbf{a}^{m} \mapsto \mathbf{u}^{m} := H(\mathbf{a}^{m})$
  • $G$ e $H$ são escolhidos de forma a minimizar o quadrado do erro acumulado entre a aproximação e os snapshots (in-sample):

    $\displaystyle E_{in} = \frac{1}{M} \sum^{M}_{m=1} \lVert \mathbf{\hat{u}}^m - \mathbf{u}^m \rVert_{\Omega}^2$

  • Autoencoder 1 - Proper Orthogonal Decomposition (POD): POD é Um Autoencoder linear otimizado para encontrar subespaços otimizados
    • POD Encoder $G$: $G(\mathbf{u}_{i}) = \mathbf{a}_{i} = (\mathbf{u} - \mathbf{u}_{0}, \mathbf{u}_i)_{\Omega}$ ❓ (Não entendi o $u$ sem indice)
    • POD Decoder $H$: $\displaystyle \mathbf{\hat{u}}(\mathbf{x}) = \mathbf{u}_0(\mathbf{x}) + \sum^{M}_{i=1} \mathbf{a}_{i}\mathbf{u}_{i}(\mathbf{x})$
  • Autoencoder 2 - K-Means:

dd-fm's People

Contributors

rafaelcmoura avatar

Watchers

 avatar

Recommend Projects

  • React photo React

    A declarative, efficient, and flexible JavaScript library for building user interfaces.

  • Vue.js photo Vue.js

    🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.

  • Typescript photo Typescript

    TypeScript is a superset of JavaScript that compiles to clean JavaScript output.

  • TensorFlow photo TensorFlow

    An Open Source Machine Learning Framework for Everyone

  • Django photo Django

    The Web framework for perfectionists with deadlines.

  • D3 photo D3

    Bring data to life with SVG, Canvas and HTML. 📊📈🎉

Recommend Topics

  • javascript

    JavaScript (JS) is a lightweight interpreted programming language with first-class functions.

  • web

    Some thing interesting about web. New door for the world.

  • server

    A server is a program made to process requests and deliver data to clients.

  • Machine learning

    Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.

  • Game

    Some thing interesting about game, make everyone happy.

Recommend Org

  • Facebook photo Facebook

    We are working to build community through open source technology. NB: members must have two-factor auth.

  • Microsoft photo Microsoft

    Open source projects and samples from Microsoft.

  • Google photo Google

    Google ❤️ Open Source for everyone.

  • D3 photo D3

    Data-Driven Documents codes.