|
" A Data-Driven Design Process Including Multiphysics for Synchronous AC Machines Using High-Performance Computing "
Mohammadi, Mohammad Hossain
Alister Lowther, David
Document Type
|
:
|
Latin Dissertation
|
Language of Document
|
:
|
English
|
Record Number
|
:
|
1107801
|
Doc. No
|
:
|
TLpq2456877541
|
Main Entry
|
:
|
Alister Lowther, David
|
|
:
|
Mohammadi, Mohammad Hossain
|
Title & Author
|
:
|
A Data-Driven Design Process Including Multiphysics for Synchronous AC Machines Using High-Performance Computing\ Mohammadi, Mohammad HossainAlister Lowther, David
|
College
|
:
|
McGill University (Canada)
|
Date
|
:
|
2019
|
student score
|
:
|
2019
|
Degree
|
:
|
Ph.D.
|
Page No
|
:
|
213
|
Abstract
|
:
|
Over the past century, electric machines have been used in many applications and sizes, ranging from washing machines and other home appliances to large pumps and fans within the industrial sector. Typical procedures for designing them have advanced from analytical formulations to the more recent finite element analysis for running physics simulations. This latter tool allowed motor designers to arrive at design solutions operating close to reality with minimal need of manufacturing hundreds if not thousands of electric machines. Despite the simulation benefits, only the electromagnetic performances are generally considered during a design process. Other physics, such as structural, acoustic and thermal, are usually ignored and only verified for a selected design. This assumption could lead to suboptimal solutions due to the tradeoffs among physical phenomena. Another issue is the increase in simulation time while incorporating multiphysics simulations in the design process. Depending on the modeling complexity, each motor simulation could take minutes or even hours to solve which can be problematic when thousands of designs are to be analyzed for different physics and operating points. Also, previous works often neglect using the simulated data to understand the underlying relationships among design performances and variables. In an optimization problem, only the final set of optimal solutions are analyzed which does not necessarily provide information on how they were achieved for re-use. To address these issues, this thesis proposes a multiphysics design process for synchronous AC machines using a data-driven approach. Each stage of the proposed process is explained using different case studies of a synchronous reluctance machine with a varying number of slots and rotor barriers. Upon setting the initial specifications, thousands of motor geometries are simulated using electromagnetics, structural, acoustics, and thermal analyses in days instead of months with the help of a high-performance computing system. A new methodology known as barrier mapping is then introduced which relates the design spaces of multiple-barrier rotors and systematically reduces their simulation time. Finally, the acquired multiphysics datasets are statistically analyzed for all their performances and variables before recommending various optimal designs for different priorities. Extracting design knowledge and guidelines can help a motor designer arrive at a more informed choice when analyzing results and selecting an optimal design. While this thesis focuses on electric machines, the presented multiphysics design process is applicable to any physical device.
|
Subject
|
:
|
Computer engineering
|
|
:
|
Electrical engineering
|
| |