Loading...

Identification Of Surrogate Models for Hybrid Distributed Parameters Systems Using Machine Learning Algorithms

Taghizadeh, Mohammad Javad | 2024

1 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 56908 (06)
  4. University: Sharif University of Technology
  5. Department: Chemical and Petroleum Engineering
  6. Advisor(s): Bozorgmehry Boozarjomehry, Ramin
  7. Abstract:
  8. In various industries, particularly in the process industries, computational fluid dynamics (CFD) stands as the predominant method for simulating distributed systems. In these methods, discretization of system geometry and partial differential equations is necessary, resulting in a system of algebraic or ordinary differential equations, or a combination thereof. The significant computational demands arising from the extensive number of equations derived from dynamic system simulations highlight the necessity for substantial processing and computing power. The objective of this project is to reduce the computational load associated with solving these equations. It focuses on utilizing machine learning to access surrogate models, in special cases, for predicting the behavior of various components within the system. These surrogate models aim to achieve accuracy comparable to that attained through numerical methods, yet within significantly truncated time. The primary objective of this project is to develop a systematic framework that leverages machine learning to train surrogate models. These surrogate models are specifically tailored for data derived from computational fluid dynamics. Within this project, two distinct machine learning subsets have been used: principal component analysis for dimensionality reduction, and deep learning via neural networks for predicting the states of the system. This methodology has been validated for predicting fluid velocity across five distinct scenarios. These scenarios include cavity flow, flow around a circular cylinder, flow around a rectangular cube, draining the fluid from the tank, and water boiling. The integration of principal component analysis (PCA) and neural network techniques facilitates convergence during neural network training, diminishes the risk of overfitting, and expedites the training of alternative models
  9. Keywords:
  10. Machine Learning ; Computational Fluid Dynamics (CFD) ; Distributed Parameter System ; Distributed System ; Cavity Flow ; Principal Component Analysis (PCA)

 Digital Object List

 Bookmark

No TOC