Degree Type

Dissertation

Date of Award

2019

Degree Name

Doctor of Philosophy

Department

Mechanical Engineering

Major

Mechanical Engineering

First Advisor

Song-Charng Kong

Abstract

A novel approach to incorporating Machine Learning into optimization routines is presented. An approach which combines the benefits of ML, optimization, and meta-model searching is developed and tested on a multi-modal test problem; a modified Rastragin's function. An enhanced Particle Swarm Optimization method was derived from the initial testing. Optimization of a diesel engine was carried out using the modified algorithm demonstrating an improvement of 83% compared with the unmodified PSO algorithm. Additionally, an approach to enhancing the training of ML models by leveraging Virtual Sensing as an alternative to standard multi-layer neural networks is presented. Substantial gains were made in the prediction of Particulate matter, reducing the MMSE by 50% and improving the correlation R^2 from 0.84 to 0.98. Improvements were made in models of PM, NOx, HC, CO, and Fuel Consumption using the method, while training times and convergence reliability were simultaneously improved over the traditional approach.

Copyright Owner

Aaron M Bertram

Language

en

File Format

application/pdf

File Size

207 pages

Available for download on Tuesday, January 21, 2020

Share

COinS