Optimization in Design of Electric Machines

Optimization in Design of Electric Machines:
Methodology and Workflow

Optimization is a very popular term in modern design of electrical machines and
devices in general. Due to the everlasting competition in the world markets, increased
cost of electrical energy and pressures for its conservation, design optimization of
electrical machines becomes more and more interesting and important. In other
words, mathematical optimization helps designers to push the existing invisible
design boundaries while using available materials and technology. The objective of
the optimization process is usually to minimize either the initial cost
of the machine or its lifetime cost including the cost of lost energy. Other objectives
such as mass minimization or efficiency maximization may be also appropriate in
some situations
It is very important to differ the exact mathematical optimization procedure from the
mere parameter variation. Many machine designers and scientists will use the word
”optimization” without being aware of its true background. One can quite often find
papers presented on conferences proclaiming optimal design, but actually describing
sensitivity analysis done on a single problem by varying one or few parameters with
heavy conclusion drawn at the end. This can be explained and understood through
the words of prof. TJE Miller (who is certainly aware of the true optimization) [2]: ”To
a WISE engineer, optimal design means a compromise between conflicting factors,
often producing an imperfect result from optimistic aspirations. Who would use a title
such as Compromises in the design of...? Optimal sounds better.... This paper is
written at a basic engineering level and makes no attempt to apply sophisticated
optimization theory.”
METHODOLOGY
Most of the requirements for electrical machine design
are in contradiction to each other (reduction in volume
or mass, increase in efficiency etc.). Therefore finding a
design that will satisfy all of them can be an
overwhelming task due to a large number of parameters
whose effects on the motor performance and quality of
the design are strongly coupled. There is an obvious
need for a systematic approach to decision making
based on an iterative scheme that would gradually lead
to an optimal motor design which satisfies all the
constraints imposed upon it and still fulfills its main task
to produce torque.
The design of a machine can be described by a
vector ~x of D variables stating dimensions, non-
dimensional parameters, current densities, types
of materials used etc. The design is subject to a
set of m constraints which may include
specifications arising from international technical
standards and electromagnetic, thermal,
mechanical or manufacturing constraints. The
goal of the design optimization is to make a
chosen objective function f(~x) reach its
minimum or maximum value while keeping other
technical indices within acceptable ranges
The result of the single-objective optimization is a single vector whose
parameters completely define a single machine design, and the result of
the multiobjective optimization is a population of non-dominant
solutions which belong to a Pareto optimal set. Since none of the vectors
dominate, they are all equally good solutions which provide invaluable
insight to the decision maker on how to choose the best design to satisfy
the performance criteria.
It can be concluded that optimization techniques in general require
different type of choices, such as the following :
1) type of optimization algorithm;
2) optimization variables (parameters), their type and
constraints;
3) constraint functions;
4) objective function(s);
5) parametrized model of the problem to be solved;
A. Optimization algorithm
There is a wide variety of optimization techniques which can be used for motor design.
Some of the techniques require providing a feasible starting point for the search process
to begin. Finding a feasible starting point that would lead to a global minimum of the
objective function is an almost impossible task. The complexity of electric machine
design is such that explicit methods of optimization, such as those dependent on making
certain derivatives equal to zero, are not feasible. The optimization techniques which do
not require a specific starting point represent a more flexible and attractive approach.
Therefore mostly metaheuristic techniques capable of solving global optimization
problems subject to non-linear constraint are used (Fig. 1). Metaheuristic algorithms,
however, do not strictly mathematically guarantee that the optimal solutions are ever
found, but there is a high possibility that a near optimal solution will be determined.
From designer and engineering point of view, it is a global optimum.
One of the most promising algorithms from the class of evolutionary algorithms widely
used in the field of electric machines is Differential Evolution (DE) – first introduced by
Price and Storn in 1995. Several authors have tested the algorithm using some well
known and difficult numerical test problems, and showed that it was capable of
outperforming other well known optimization algorithms. The algorithm was later
improved and named Generalized Differential Evolution (GDE) (extended DE for
constrained multiobjective optimization) by Lampinen.
Fig. 1. Overview of metaheuristic source: Wikipedia
In short, DE method works on a population (generation) which is a set of NP individuals
(members), where each individual presents one machine design. Initial population is
randomly initialized inside the boundary constraints. Candidate (trial) population is
obtained by crossover and mutation processes from the existing population. Next
generation is obtained by comparing the existing and canditate population by
choosing members that satisfy boundary functions and/or have better objective
function. Variety of other algorithms is used in electric machine design optimization:
Genetic Algorithm (GA) [18]–[21], Particle Swarm Optimization (PSO) [22]–[25],
Simulated Annealing (SA) [26] etc. Authors in [26] compared GA, SA and DE on design
optimization of permanent magnet motor and authors in [27] compared DE, GA and
PSO on design optimization of microstrip antennas. Both groups agree that DE
performance is the best. In [28], [29] PSO and GA were compared and PSO was found
computationally more effective with slightly better objective function value reached.
In [29], it is shown how PSO performs better then GA so some authors decided to use
Hybrid GA-PSO method [30]. Any ranking attempt between the different algorithms
is not truly appropriate since the performance is problem and case-dependent and
from engineering point of view, satisfying in all cases. Nevertheless, authors mostly
agree that DE achieves the best fitness values, i.e. the minimum objective function
value, usually with smaller number of evaluations. The second best-performing
algorithm is often PSO.
B. Definition of variables
• The variables of the optimization algorithm that compose the vector ~x are geometrical and other
quantitiesthat describe the outlook of the model or are derived from them. The most influential variables on
the target functions are usually identified using a sensitivity analysis tool [31], [32]. All the variables are
constrained in prescribed intervals, so called boundary constraints, which define the search space or the
design space. After reproduction in optimization algorithm, some variables of the newly created candidate
vectors may fall out of boundaries. These variables can be ”repaired” using random values generated within
the feasible range using the scheme proposed in [15]. Some authors [3], [4], [33] used model parameters
(stator bore diameter, depth of stator slot etc.) directly as optimization variables while some authors [1], [5],
[6], [8], [34], [35] used ratios of model parameters as optimization variables. It is better practice to choose
variables which are given as non-dimensional ratios of related geometrical parameters, for example ratio of
slot depth to difference between stator outer radius and stator inner radius, ratio of stator inner diameter to
stator outer diameter, ratio of tooth (or slot) width to slot pitch, ratio of magnet length to airgap length,
magnet pole arc relative to the pole pitch. Some other geometrical parameters can just be considered as
optimization variables directly within prescribed interval (outer stator diameter, stack length, slot current
density) or can be used as relative parameters (outer stator diameter to maximum outer stator diameter).
The usage of dimensionless variables, particularly as a ratio of the pole pitch, enables for example the
extension of results from the studied configuration to other motors with different number of poles [35]. An
example of parametrized IPM motor topology with geometrical design parameters is given in Fig. 2 and the
corresponding optimization variables are listed in Table I. The Differential Evolution algorithm, for example,
assumes that parameters in the population are continuous real numbers. However, in the motor design some
of the parameters, e.g. the number of pole pairs, can have only integer values. The example of discrete
variables are the standard wire diameters which can be used for the armature winding. The main difference
between integer and discrete variables is that although they both have a discrete nature, only discrete
variables can assume floating point values. The discrete variables can also be unevenly spaced. The original
DE algorithm was modified by Lampinen and Zelinka [15] in order to include mixedinteger- dicrete
parameters. Similar approach exists for GA [19].
Fig. 2. IPM motor topology with geometric design parameters

Optimization in Design of Electric Machines

Uploaded by

Copyright:

Available Formats

Optimization in Design of Electric Machines

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Optimization in Design of Electric Machines

Uploaded by

Copyright:

Available Formats

Optimization in Design of Electric Machines:

Methodology and Workflow

You might also like