Files in this item

FilesDescriptionFormat

application/pdf

application/pdfGAO-DISSERTATION-2021.pdf (31MB)Restricted to U of Illinois
(no description provided)PDF

Description

Title:Study of multi-component spray and combustion and their optimization with machine learning in internal combustion engines
Author(s):Gao, Suya
Director of Research:Lee, Chia-Fon F
Doctoral Committee Chair(s):Lee, Chia-Fon F
Doctoral Committee Member(s):Stephani, Kelly; Wang, Xinlei; West, Matthew
Department / Program:Mechanical Sci & Engineering
Discipline:Mechanical Engineering
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):Internal Combustion Engines
Modeling and Simulation
Vaporization
Atomization
Optimization
Machine Learning
Abstract:As energy resources dry up and environmental pollution becomes severe, clean and highly efficient internal combustion engine operation has become the need of the hour. Understanding the physical process of the fuel spray evolution is imperative in predicting and improving the performance of applications of modern fuel injection systems. Once fuel is directly injected into the combustion chamber, the proceedings, such as liquid fuel vaporization and formation of the fuel-air mixture, would conspicuously influence the onset of combustion, combustion efficiency and thus the emissions. Studies in this dissertation focus on two phenomena in internal combustion engines: the liquid fuel spray evolution and fuel-air mixture preparation. The computational methodologies which specialized in physics-based and data-driven model development and simulations are utilized in order to predict the performances of internal combustion engines and to seek better solutions for clean energy, higher efficiency, and lower emission design. In this dissertation, the objectives included are: 1) a comprehensive model considering preferential vaporization of a complex fuel mixture using discrete approach; 2) a unified model for spray atomization under the consideration of flash boiling; 3) a new approach of emission reduction in the compression-ignition engine is introduced. By controlling the fuel-air mixing process the burning sequence of the mixture is shifted from traditional ignition behavior; 4) applications of artificial intelligence techniques to help accelerating the model development and engine design processes. Fuel droplet vaporization rates and corresponding flow evolutions are important because they directly influence the onset of combustion, the efficiency of combustion, thus emissions. Therefore, prediction of the real fuel vaporization is the first step of this research project. An efficient and precise multi-component fuel droplet vaporization model has been developed in this work using a discrete approach. A quasi-1D frame for liquid droplet is constructed such that the evolution of the non-uniform mass and energy distribution for liquid droplets can be accurately described by the surface state and bulk mean state. The second objective is to develop a comprehensive model for spray atomization under the consideration of flash boiling. To develop a comprehensive multi-component fuel flash boiling model, three main processes, bubble nucleation, bubble growth, and breakup, are considered. The effect of superheating on breakup is built into a classic aerodynamic breakup model such that the unified breakup model is feasible to be used over all possible engine operation conditions. The new vaporization and breakup model are integrated into a CFD program (KIVA-3V) to further conduct the evaluation of the spray superheating effects. A new approach of NOx reduction in compression-ignition engines is introduced using a fuel that exhibits the “inverted 𝜙-sensitivity” (IPS) behavior in this work. From 0-D calculations, ethanol, which is also a clean and renewable fuel, is selected as the test fuel for satisfying the criteria considered in this study. Engine conditions that cover from low to high load operations are tested with 3-D simulations. Characteristics of ethanol vapor/air mixture are analyzed to reveal the mechanism of inverted 𝜙 -sensitivity fuel in reducing emissions. Optimal operations for compression-ignition engines with IPS fuel are conducted using the genetic algorithm to exaggerate the superior effect of inverted 𝜙-sensitivity on engine performance. Advanced modeling methods utilizing machine learning are developed to substitute the high-fidelity CFD simulation and coupled with optimization algorithms to accelerate the design processes and increase predictive capability of models. The underlying patterns and relations between engine control parameters and performances (emissions and efficiency) are revealed by data-driven models. Such compact information processes are assembled to construct a stacked machine learning model. Instead of using CFD models, this machine learning model is called as a function at the genetic algorithm interface for optimization procedures. This work provides the fundamental understandings of liquid fuel spray and corresponding mixture evolutions and combustion processes in internal combustion engines. New technologies for better atomization process and engine emission reduction are provided. Applications of artificial intelligence demonstrate the possibilities of implementing the advanced computational knowledge and addressing the cross-disciplinary work to engine studies.
Issue Date:2021-11-12
Type:Thesis
URI:http://hdl.handle.net/2142/113960
Rights Information:Copyright 2021 Suya Gao
Date Available in IDEALS:2022-04-29
Date Deposited:2021-12


This item appears in the following Collection(s)

Item Statistics