You are on page 1of 10

Chapter 5

2DMATEMA:
MATERIALS MANAGEMENT

THE MATERIALS IN FEM PROBLEMS


As evident from physical ingenuity, and more clearly from the mathematical
developments in the following Chapters 6 and 7, the numerical constants
associated to the materials (permeability, conducibility, permettivity, and so
on) have a very important role in the numerical analysis of ElectroMagnetic
fields. AMADEUS, obviously, takes care of this fact: there is an executable
file in AMADEUS package, 2DMATEMA.EXE, that has the only scope to
create and maintain the material databases of the user.
2DMATEMA, however, is not simply a database maintenance program, but
it has a number of mathematical and graphical features that perform
several very useful operations. By means of 2DMATEMA It is not only
possible to create a new material, to edit a previously defined material, to
view in a graphic way the characteristic of the permanent magnets. More
than all, 2DMATEMA is a very powerful program for the automatic
simulation of the non linear magnetisation curves, that often arise when
dealing with magnetic problems involving commercial iron. In the following
paragraphs, the operations performed by 2DMATEMA on B - H curves are
exposed in detail; follows a description of the operations of 2DMATEMA.

OPTIMUM SIMULATION OF B-H CURVES IN FEM ANALYSIS OF


MAGNETIC FIELDS

In FEM analysis of static electromagnetic fields where non-linear materials


are present, it is necessary to determine numerically the magnetic field H
from the knowledge of the magnetic flux density B as well as dH(B)/dB.
The same problem holds for dynamic simulations, where the hysteresis of
the material has to be taken into account. It is well known that a robust
software solution of non-linear problems requires a continuous function
H(B) as well as its derivative dH(B)/dB.
In general, from the data-sheet of the material a number of points of the
magnetisation curve, say N + 2, can be deduced: N is the number of the
Chapter 5 - 2DMATEMA: Materials Management

points internal to the interval, and two are the endpoints of the interval, i.e.
the points (B = 0, H = 0) and (B = Bmax, H = Hmax). Number N can be
very large: from each data-sheet one can determine the numerical values
of the flux density and of the magnetic field in 20 - 40 points and more.
B - H curves are well represented by cubic polynomial splines. This fact
depends on the numerical behaviour of the said curves, that exhibit parts
that are practically linear, and parts with rapidly varying curvature. A single-
interval polynomial representation of the curves is not feasible, and
algorithms such as Lagrange or Neville do not give good results. Also
exponential representations of the curve are rather poor, since they are
based on a limited number of parameters, and lead to great errors in the
values of H in the saturated zone. Cubic polynomials interpolate well both
straight lines and rapidly varying curves; the order the polynomial is always
the same, and there is not any constraint in the choice of the experimental
points.
For several reasons it is rather difficult to determine a set of cubic splines
for a great number of points. Firstly, obtaining a cubic spline interpolation is
linked to a matrix inversion: the accuracy of this operation decreases when
the dimension of the matrix increases. Secondly, the Newton iteration
process implies the necessity of determining the value of the magnetic field
when the numerical value of the magnetic flux density is given. When using
cubic splines, the number of operations to perform is always equal to six
multiplications and three sums for each numerical determination of H.
Since cubic splines is a multi-interval representation, it is preliminarly
necessary to determine the interval where the current flux density is
located: the higher the number of points, the greater the computer time to
determine the interval. Finally, a great number of intervals does not ensure
accuracy in determination of the optimising parameter that will be shown in
the following.
It is common practice to choose, for conventional materials, a number of
points between 7 and 11. There is no direct information about how many
points to choose neither where they should be located. General rules state
that, apart first and last point, one point is needed in the linear part of the
curve, three to five points in the so-said knee, and one to three points in
the saturated zone.
The problem of the optimum representation of B-H curves can therefore be
splitted in the following two sub-problems:

1. given a set of known magnetisation points, to find the optimum set of


cubic splines;
2. to find the optimum set of n magnetisation points, that is a sub-set of the
known N experimental points.

V-2
Chapter 5 - 2DMATEMA: Materials Management

In the following, N-set denotes the set of N known experimental (B, H)


points of the magnetisation curve (without considering the endpoints B = 0
and B = Bmax, and n-set is a subset of n of the N points (N ### n). N-set
and n-set are sorted for increasing values of B.
As a final remark, the numerical determination of H from B is not the main
task in FEM analysis of electromagnetic fields. In each iteration of the
Newton process relevant to a geometry with NE elements in non-linear
materials, exactly NE numerical values of H from B must be calculated.
Utilisation of an n-set rather than an N-set allows to save exactly NE (N - n)
cycles of the type IF - THEN. The possible time saving is obviously limited,
especially if compared with the total computer time of a single Newton
iteration. These considerations are no longer valid if the spline
approximation is performed inside the FEM program itself; when
magnetisation curves change along with FEM simulation (this is the case of
the hysteresis curves in dynamic simulations) it is highly desirable to have
an optimal cubic spline representation of the curve.

CUBIC SPLINE INTERPOLATION


The algorithm is well known, and it is summarised here only for
completeness of the discussion. Cubic splines are based on the hypothesis
that between two known points, say (Bi, Hi) and (Bi+1, Hi+1), the function
H(B) behaves as a cubic polynomial:

H(B) = ai B3 + bi B2 + ci B + di , Bi B Bi+1 (5.1)

There is no constraint in the choice of the location of the points: no matter


how the interval is large in B-direction and/or in H-direction, the hypothesis
is that into each interval the unknown function can be represented by the
polynomial expression in eqn (1). Obviously, parameters ai, bi, ci, di
change from an interval to the other. Since a n-set is defined by n + 2
known experimental points, the number of intervals is n + 1, and the
number of unknowns to determine is 4 (n + 1). A number of linear
equations can be written in order to determine the unknowns. Firstly, into
each interval the given endpoints should belong to the interpolating H(B)
curve, and two linear equations can be written:

ai Bi3 + bi Bi2 + ci Bi + di = Hi
(i = 1, n + 1) (5.2)
ai Bi3+ 1 + bi Bi2+ 1 + ci Bi + 1 + di = Hi + 1

V-3
Chapter 5 - 2DMATEMA: Materials Management

Moreover, the first and the second derivative must be continuous at the
interface between two intervals. It is possible to write 2 n more linear
equations:

3 ai Bi2+ 1 + 2bi Bi + 1 + ci = 3 ai + 1Bi2+ 1 + 2bi + 1Bi + 1 + ci + 1


(i = 1, n ) (5.3)
6 ai Bi + 1 + 2bi = 6 ai + 1Bi + 1 + 2bi + 1

The number of linear equations is 4 n + 2. Two more equations have to be


written in order to define the linear problem. A natural choice is to stipulate
the numerical value of the first derivative in the two endpoints of the whole
interval:

3 a1B12 + 2b1B1 + c1 = H ' ( B1 )


(5.4)
3 an + 1Bn2+ 2 + 2bn + 1Bn + 2 + cn + 1 = H ' ( Bn + 2 )

The linear problem is completely defined: the solution of a system of 4 (n +


2) linear algebraic equations leads to the numerical knowledge of the cubic
spline parameters.

Optimisation of the Cubic Spline Interpolation by Line Minimisation


and Numerical Determination of the Derivative in the Last
Experimental Point

It has been shown that a n-set does not define a unique cubic polynomial
approximation, but two more parameters have to be given: the numerical
values of the first derivative of H(B) in the first and in the last experimental
points. Numerical evaluation of the first parameter is quite simple, since in
the point (B1, H1) = (0, 0) the curve is practically linear, and it is possible to
determine this parameter as a finite difference:

dH H H1
H '( B1) = 2 (5.5)
dB B = B B2 B1
1

Numerical evaluation of the second parameter is rather complicated, and


computer tests confirm that this quantity plays a dominant rule into the
approximation process. This is due to the fact that, in the neighboroughs of
the last point (Bn+2, Hn+2) = (Bmax, Hmax), there are not only great
variations of H in opposition to small variations of B, but also the derivative
dH(B)/dB varies greatly. As a matter of fact, the numerical value of
H'(Bn+2) has to be determined algorithmically rather than evaluated from

V-4
Chapter 5 - 2DMATEMA: Materials Management

the data sheets of the materials. In the neighboroughs of B = Bmax, an


approximation of the type

dH H n + 2 H n +1
H '( Bn + 2 ) = (5.6)
dB B = B Bn + 2 Bn +1
n+2

is not longer correct.


The application of cubic polynomial spline technique to a real case is
shown in Fig. 5.1, where the procedure has been applied to the
determination of the curve H(B) for Armco M4T28 oriented electrical steel
(n = 7). The numerical values of the first derivative in the endpoints were
calculated by means of Eqs. (5.6) and (5.7).
This curve does not look very well. It is not monotonic in H; first and
second derivative change their signs several times. It is not only an
aesthetic problem: Newton process will slower, and convergence is not
ensured.
The problem to solve consists in communicating to the computer that the
curve is not attractive enough. On the other hand, attractiveness can be
seen as a sum of nice details, and therefore we will look for an integral
quantity.

dH(B)/dB

H H(B)

Fig. 5.1 H(B) curve for a Armco M4T28 material interpolated using cubic
splines (9 points, n = 7). The derivative of the curve for Bmax is
evaluated as (H(9) - H(8)) / (B(9) - B(8)). Circles denote
experimental points, thick line is H(B), slim line is dH(B)/dB.

V-5
Chapter 5 - 2DMATEMA: Materials Management

The reference is the well known B - H curve in all textbooks, where the first
and second derivative never change their sign: the ideal curve never
changes curvature. From a computer point of view, it is possible to
translate this fact by stating that the integral of the negative part of the
second derivative should be minimum (if second derivative is never less
than zero, the integral of the negative part is obviously zero). A new
function can be defined, the integral of the negative part of the second
derivative. This function depends only on the derivative of the magnetic
field with respect to the flux density in B = Bn+2:

Bn + 2
d2H

2
dB = f dH
dB

dB

B = Bn + 2
(5.8)
0

Behaviour of the H(B) curve is greatly improved if this quantity is reduced.


Analytical minimisation of the function is impossible in practice, but the
problem can be solved numerically by means of one dimensional
minimisation method (Brent' method) where bracketing and golden section
search techniques are used.

dH(B)/dB

B H(B)

Fig. 5.2 H(B) curve for a Armco M4T28 material interpolated using cubic
splines (9 points, n = 7): the derivative of the curve for Bmax is
obtained by the described optimisation technique. Circles denote
experimental points, thick line is H(B), slim line is dH(B)/dB.

V-6
Chapter 5 - 2DMATEMA: Materials Management

In Fig. 5.2 the B-H curve for the same material as in Fig. 5.1 is shown after
the application of the optimisation algorithm: both H(B) and dH(B)/dB are
smooth and regular.

THE MENUS OF 2DMATEMA


When 2DMATEMA is invoked from DOS, the conventional graphic screen
of AMADEUS appears, and the user has to accept Copyright rules.
The first operation is the choice of the file that contains the parameters of
the materials: also this operation is performed in a totally standard way,
very similar to the choice of the project described in Chapter 2.
Once the file has been chosen, the materials contained in the file are listed:
the user has to previously determine the type of action he wants to
perform, and subsequently he has to choose the material on which the
operation has to be performed.

view the material is chosen, and the characteristic of the


material are shown on the screen

add a new material is added to the database of the materials

edit a material previously inserted in the current file is


chosen, and subsequently edited

delete a material is chosen and subsequently deleted from the


current file of materials

continue the program is terminated and the control returns to


DOS.

AMADEUS can work with the following type of materials:

LINEAR, ISOTROPIC

LINEAR, NON ISOTROPIC,

PERMANENT MAGNETS

NON LINEAR, ISOTROPIC

V-7
Chapter 5 - 2DMATEMA: Materials Management

In the following, the parameters of these different types of materials are


discussed, as well as the physical dimensions that AMADEUS will consider
the input quantities.

LINEAR ISOTROPIC MATERIALS


name of the material chosen by the user, up to 20 characters

type of the material linear

directionality isotropic

permeability [H / m]

conducibility [S m]

permettivity [F / m]

LINEAR NON ISOTROPIC MATERIALS


name of the material chosen by the user, up to 20 characters

type of the material linear

directionality anisotropic

easy axis permeability [H / m]

transverse axis permeability [H / m]

easy axis direction (degrees) the angular shift between the


direction of the easy axis and a conventional x - y
reference frame. Please note that this number can by
altered by a quantity of 180 degrees without any effect.

conducibility [S m]

permettivity [F / m]

V-8
Chapter 5 - 2DMATEMA: Materials Management

PERMANENT MAGNETS
name of the material chosen by the user, up to 20 characters

kind of material permanent magnet

type of the material linear

directionality anisotropic, axial or radial

easy axis direction (degrees) the angular shift between the


direction of the easy axis and a conventional x - y
reference frame. Respect to non isotropic materials, by
adding to this number a quantity of 180 degrees the
magnetisation is in practice reversed.

conducibility [S m]

permettivity [F / m]

number of points the number of points defined by the user to define the B -
H curve of the current materials.

When dealing with permanent magnets, 2DMATEMA shows in the lower


left part of the screen the B - H chart of the material; in the lower right part
the B - H coordinates of the describing points are listed.

NON LINEAR MATERIALS


name of the material chosen by the user, up to 20 characters

type of the material non linear

directionality isotropic

conducibility [S m]

V-9
Chapter 5 - 2DMATEMA: Materials Management

permettivity [F / m]

number of points the number of points defined by the user to define the B -
H curve of the current materials.

When dealing with non linear materials, 2DMATEMA shows in the lower
left part of the screen the B - H chart of the material; in the lower right part
the B - H coordinates of the describing points are listed.

V - 10

You might also like