Introduction To Finite Element Method Dr. R. Krishnakumar Department of Mechanical Engineering Indian Institute of Technology, Madras Lecture - 29
Introduction To Finite Element Method Dr. R. Krishnakumar Department of Mechanical Engineering Indian Institute of Technology, Madras Lecture - 29
Introduction To Finite Element Method Dr. R. Krishnakumar Department of Mechanical Engineering Indian Institute of Technology, Madras Lecture - 29
Dr. R. Krishnakumar
Department of Mechanical Engineering
Indian Institute of Technology, Madras
Lecture - 29
In the last class, we were talking about contact and we were really getting into the
modeling aspects of the problem which we had carried out.
There are lot of things that we learnt as far as boundary conditions were concerned
and we realized that things like contact is very, very important in modeling,
especially in mechanical engineering and that the results may not be correct, if you
make assumptions, drastic assumptions in order to avoid contact. We will come back
to this contact and I am going to right now and I am going to discuss more in detail
about contact nonlinear finite elements. But before we do that, let us have a look at
one of the things which we left out. If you remember, I had asked about the axisymmetric element, which we just left out and we entered into certain discussions on
contact, it is good. We will come back to that and we will just finish that part of the
axi-symmetric element, sort of niceties regarding the axi-symmetric element, before
we come and again talk about the nonlinear finite element analysis, where contact is
going to play a major role.
Let us just have a look at this. You know, all of you have seen it, I have written this
equation before in one of the earlier classes.
You know, for example, how the strain displacement relationship is given for an axisymmetric case. We had already discussed what an axi-symmetric case is? We said
that there should be an axis of symmetry and we had already discussed about the
stress state, strain state and so on, in the beginning of this course. I hope you all
remember that or else have a look at your notes, but nevertheless we had done this
before. What I want to just point out is that there is a small change, when you go to
axi-symmetric element with respect to this ones and zeros matrix.
You remember that we had called this matrix as what? H matrix, with ones and zeros
matrix or this is one of the earlier matrices that we had seen. In the earlier version of
this matrix, we had how many rows and columns? 4 by 4. Now, because of the fact
that epsilontheta, epsilontheta which has a 1 by r in it, there is a small change in this
matrix for axi-symmetric element. Actually, how does the axi-symmetric element
look like? The axi-symmetric element looks something like this.
Suppose this is the axis and say this is the complete body which we are analyzing.
Say, there is a cylinder, a hollow cylinder with a radius say r, r1 and r2. Suppose we
are analyzing like this and we are dividing this into number of elements and one such
element happens to be here, suppose we are dividing into a number of elements and
so on, the type of element we are using is axi-symmetric element. Please note this
carefully, though we are discretizing it in two dimensional regime or two dimensional
plane or in a plane with say r and z, r and z plane, as the two axis, r and z being the
two axes which define this plane, this element is actually a ring element. It is actually
a ring which goes to the other side. So, the element is formulated, it goes like that.
The element is formulated taking into account this complete ring. That is the first
point.
Where do we bring this in? We will see that in a minute. First point is that, though we
are discretizing it as an axi-symmetric element, in the sense that in the r z coordinate
system, what we are considering as an element is actually a ring element. That is the
ring element. Is that clear? Now, we will come to that in a minute as to where we are
going to put it, but concentrate on this right now that this H matrix what we have has
now one more column added to it because of this epsilontheta and epsilontheta is u by r,
you remember that we had done this before, u by r.
This is because of the change in the radius. As the radius moves out, remember that
that is what we said that there is a change in the dimension of a ring and hence there
is u by r that comes into play and that is epsilon theta and hence you see immediately
that there is one more column that is added here as zero 1 by r zero zero. In other
words, there is a small change in this one zeroes matrix which we have seen. This is
similar to u comma x u comma y w comma x w comma y, that we have seen already.
The only addition now is this u, to take care of this u by r.
What is the next step? Remember the next step. What we did was to replace this by an
isoparametric formulation or in other words to replace r and z by our natural
coordinate systems psi and eta. This u comma r u comma z and so on is now replaced
by means of u comma psi through that gamma matrix; remember that gamma matrix.
Again remember that in the plane strain or plane stress axi-symmetric problem, we
had only a gamma matrix to be just j inverse. This is 2 by 2 then zero zero 2 by 2; 4
by 4 matrix.
Here there is only one addition that we have 1 here. Is that clear? The calculations are
very similar, exactly the same. There is no change. Only thing is that these matrices
6
now change a bit. Is it clear? Have a look at that. If there is any question, just have a
look at this and this. If there is any question, I will answer that. What is now B
matrix? How do you get the B matrix? Remember how we get the B matrix? Yes,
now what I do is I substitute; from here I substitute onto this equation. This I
substitute it from this equation. When I substitute it there, then I get here, for
example, let me remove that and put say along with it, let me call this as H matrix just
for brevity that gamma, gamma I had defined here already and then now what
happens to this u comma psi u comma eta w comma psi w comma eta and u, they go
here.
Is that clear? Now, what is the next step? Next step is this u comma psi u comma eta
and all those things have to be replaced by N1 comma psi and N1 comma eta and so
on. Is that clear? That is what we did. You remember that that is what we did. But, the
only difference now is that, what is the difference? I have a u here. So, my matrix will
have another row with N1 N2 and so on. Is it clear? What will be the size of that
matrix? Correct; 5 by 8, beautiful. So, that matrix will be a 5 by 8 matrix. You know
how to substitute it, straight away. My B matrix will be this multiplied by this
multiplied by the substitution for this which is a 5 by 8 matrix and that would be my
B matrix. Is that clear? So, u comma psi u comma eta all these things would be
replaced by N1 comma psi zero N2 comma psi. I am not going to write it, because it is
exactly what you have written in you previous classes. So, that will be my B matrix.
Having done that what is my next step? I will just substitute it in my K. What do I do
there?
This is now substituted into B transpose D B dv. I have to now get this in
isoparametric form. How do I get that? How do I get that? I now substitute it for dv. I
am substituting for dv. How do I substitute for dv?
In order to analyze this problem, first I take a small part of it as r dtheta, so, r dtheta.
dv is r dtheta into dx dy. So, r dtheta dx dy or in other words, dr dz now becomes J
9
dpsi deta. This is dr into dz into perpendicular to that r dtheta. What will happen to
my integral sign here?
That will be my minus 1 to plus 1; another fellow minus 1 to plus 1, another chap
from zero to, yeah, 2 pi, fantastic. Do you understand? Zero to 2 pi. Is it clear? dtheta,
this is independent of dtheta. So, that fellow can come out, so, 2 pi can come out.
10
So, you can have 2 pi out and then you have an equation of this form. Now, the only
question is r. r is the radius. Where do we get this radius now? r is the radius, where
do we get, how do we get this? Remember that we are going to integrate it with
respect to or we are going to find out the values at the gauss points and so we are
going to get this values at the gauss points and so we need r.
We have to substitute for r at the gauss points, at these four gauss points, say for
example, if I am going to use 2 by 2 rule, I am going to get these radiuses at the gauss
points. How do I get the radius at the gauss point? The rule is very simple, because I
am using an isoparametric element and the shape of the element is what? Interpolated
with respect to the nodes or in other words the coordinates are interpolated from the
nodal values through the shape functions defined in terms of psi and eta. So, problem
becomes very simple. I need to get the shape functions at these gauss points and say
that r at gauss points can be written as N1r1 plus N2r2 plus N3r3 plus N4r4.
11
This r is nothing but sigma Niri. Because psi and eta values of the gauss point are
known, substitute it here. You will get sigma Niri. You do the same thing. Please note
that similar thing is done in the case of force and other things as well. For example, if
there is a pressure that is acting, then the pressure will be acting say internally. 1530
12
The pressure will be acting internally and you have to now integrate it throughout the
ring. Again, there will be one 2 pi term, which will come out. Is it not and in a similar
fashion even if you have a line load, a load that is acting, that load makes the problem
axi-symmetric only if that load acts throughout the line like that. Is that clear? What
does it mean? It means that this 2 pi term which we have here, this 2 pi term is
common to the stiffness matrix, to all the force terms, because we are going to
integrate the force throughout; zero to 2 pi, we are going to integrate. Is it clear? For
example, if you are looking at pressure, N transpose p into dS. What is dS in this
case? Say, length of the element into r dtheta, integrated from zero to pi; so, r l it will
become, so, 2 pi will go out. Everywhere the 2 pi will come, because we will take an
infinitesimal element, an infinitesimal volume and then integrate it from zero to 2 pi.
Is that clear? Everywhere, we will integrate it from zero to 2 pi. Go and have a look.
Deliberately I am not writing it. I want you to write it, because go and take every term
and integrate like this. You will get one 2 pi term standing outside. So, 2 pi will get
cancelled.
What is the significance? Why am I saying this? This becomes very important
because, when you give an input to many of the packages, you have to be very careful
as to what that package asks as input, because some of the packages would require the
total load that is acting on that line, some of the packages would require the total load
that is acting on the line, some of the packages would require the load acting per unit
length and so on. It is very important that you check up what the load is that this
software requires. What do I mean by that? What I mean to say is that, suppose there
is a load that is acting at this point.
13
Suppose there is a load that is acting at that point. Then, actually this load acts
throughout the ring. Is that clear? Throughout the ring. When it acts throughout the
ring which passes through those point, that point there, this point, it is important to
realize whether the software asks for total load or load per unit length or how does it
handle that load? This usually varies from package to package. That is an important
input that you have to give or else you will make a mistake. If it gives total load and it
asks for unit length or something like that and normalize it, at 2 pi we will eliminate
it, then it would be a problem. He would multiply by 2 pi, then that would be a
problem. That is very important as to what it asks.
The other thing that you have to be very careful in axi-symmetric element is the r z
coordinate system. Usually the x is r and the y is z. The softwares usually define this,
the softwares usually define this as to which is r and which is z. You cannot go
against it, then the problem is an entirely different problem.
14
For example, this particular problem you cannot do it by making the ring like this and
saying that I will take this as r and this as z, that is not allowed in most softwares.
They have a fixed way of defining x and y. Is it clear and one more thing that comes
out of this formulation is that since the pressure, actually the pressure is now rotated,
the force is rotated, this pressure that is acting is rotated about 360 degrees. The ring
is in equilibrium, automatically. Physically if you see, the ring is in equilibrium
because of the force that acts.
What does it mean? It means that for an axi-symmetric element you need not specify
a boundary condition in r, in r direction, r degree of freedom you need not specify a
boundary condition. Why, because I am rotating this. On the other side also, as the
ring comes to the other side, I will have a force that will be acting here as well,
completely, radially force will be acting throughout the ring.
15
Hence that ring is under automatic equilibrium. You need not specify a boundary
condition in the radial direction. In other words, suppose I have a problem where I
have this particular tube, a hollow cylinder with internal pressures defined.
when I integrated, would automatically equilibrate one side and the other side
accurately. The formulation will take care of the rigid body motion in the r direction.
After all why do you want to specify a boundary condition? We want to avoid a rigid
body motion. If you the specify one f and another f, then you do not avoid rigid body
motion. In this case, you definitely avoid rigid body motion in the r direction. It
cannot just expand as it likes. The movement in the r direction is not equivalent to a
rigid body motion. That should be very, very clear or in other words, when I say that
this fellow moves like that if it had been just a plane strain problem or a plane stress
problem, if I move from here, this position to another position, this position to
another position in the x direction, then it is equivalent to a rigid body motion. Clear?
On the other hand, if I change this from here to say here, suppose I say I change this
location from here to another location here; suppose I change, this whole block I
move it like this, can I move it as a rigid body? No, because I am now increasing the
radius. As I increase the radius, obviously, there is going to be strains. So, it is no
more a rigid body motion. Hence there is no boundary condition that is required in the
r direction and the pressures are also equilibrated on either side. On the other hand,
there is a small warning. This is a very practical thing that when you apply this
17
boundary condition, you need not worry about it. On the other hand, suppose I have a
solid cylinder, instead of a hollow cylinder; if I have a solid cylinder instead of a
hollow cylinder, I have a situation like this.
We can argue in a similar fashion and say that the center, this part need not be fixed,
is that clear, the radius, the radius part in the radial direction, of course in the r
direction. But, usually it is recommended that you fix the nodes at r is equal to zero in
the r direction that is r is equal to zero line, that axi-symmetric line is the r is equal to
zero line and it is important that you fix this r is equal to zero line, because usually
what happens is that the computers have a fixed number of digits of accuracy and
beyond which the computer is not very accurate; fixed number of digits. The
computer is not accurate. This depends upon so many things.
18
which are sitting here would have moved a bit and creating as if there is a hole there.
Actually it is not a physical motion. There is nothing that has happened there, but
nevertheless, because of the computer accuracy some sort of a hole seems to have
been created at the center. If this happens, especially if you are working in a
company, you will be hard pressed to explain to your boss why a hole has been
created. He is immediately going to ask, oh, the material has failed at the center, hole
is created. Then he is going to ask why there should be a hole at the centre that is
created like this and so on, because this has happened to me before and then it is very
difficult to explain it to people. Just to have a nice result, one of the reasons, you fix
this nodes at r is equal to zero. Is this clear?
Please note that the failures are not, are not, shown as a hole in finite element
analysis. This I told .. within quotes who does not understand what is happening.
The problem is that the failure has to be interpreted by you. Failure is not shown as a
crack or as a displacement as two pieces and so on. It is only the stress which is going
to come into picture for a linear case, which you have to interpret according to the
failure criterion. Hence the finite element analysis will not show this kind of holes
that are created. Nevertheless, in other words compatibility becomes very important.
What are compatibility equations or compatibility conditions? Compatibility means
that you do not create a hole. Though theoretically there are elements, in the last class
I will talk about that, incompatible elements, but they are what I would call some
crimes that we commit.
Some of the crimes, in fact there is an author called Strang. He called them as sundry
variational crimes. We commit some crimes. If a mathematician looks at it, he would
say that it is a crime but as engineers sometimes we make an approximation and make
things work. There are some approximations that have been done. So, because of that
we enter into what is called as incompatibility, incompatibility elements or
incompatibility modes. But on the whole many of the elements are compatible or they
are not there to produce a hole. Then there is a mistake. We are treating this as a
19
We had already defined or we had started our discussion on contact. We said that
there is a special element called contact element. See, previously we used to have
what is called as gap element, but nowadays this gap element is being replaced by
what are called as contact elements, contact elements, in a two dimensional case.
20
We will extend this to 3D in a minute. In two dimensional case, this contact element
is defined by three nodes, is defined by three nodes.
These nodes are two, what are called as master nodes say 1 and 2 and a slave node
say 3. Now, how do we define this? What is the picture I have drawn here? It is
nothing but a part of discretization of say a block sitting on a, say a rigid surface.
21
I define as master nodes, the nodes belonging to the rigid body. Say for example, I
have 4 nodes. Forget about these nodes for a minute. Suppose I have 4 nodes, then
this node and this node are the rigid bodys nodes or in other words master nodes. So,
master node defines a surface, defines a surface, in this case a line, because it is 2D;
defines a surface or a line into which a slave node cannot penetrate, into which a
slave node cannot penetrate. What are the slave nodes?
22
All these nodes sitting here are the slave nodes, all the nodes there, sitting there are
the slave nodes. Let me number them, say for example, let me say that this is 1 2 3 4.
Let me say that this node is say 20 and 21. Say, just for illustration purposes, let us
say that the master nodes on either side is, say 20, 21. Sometimes people call this 20,
21 as a master surface or rigid surface. The contact in this case is between a
deformable body and a rigid body. Is that clear? It is between a deformable body and
a rigid body. My major aim in writing an algorithm is to see to it that these 1 2 3 4
nodes which belong to the deformable body, which I called as slave nodes should not
penetrate the master surface, which is defined by 20 and 21.
So, number 1, when you use any contact algorithm you will be asked to define a
master and a slave. Is that clear? What are these contact elements in this case?
Contact elements will be formed between the slave nodes, note this carefully, slave
nodes and the rigid surface. How many contact elements will be there? There will be
4 contact elements or in other words stiffness of these contact elements will be
formed with 1 20 21, 2 20 21, 3 20 21 and 4 20 21. In other words, note this carefully
that the contact element is a node to surface contact element; it is a node to surface
contact element and not a surface-to-surface contact element.
23
Suppose you are working with a package like Ansys, say. The package will ask you
to pick up two surfaces, target surface. One he will call as master surface and another
is what he will call as slave surface. When you pick up, actually you will be defining
a surface. In most packages, you will be defining a surface. Afterwards if you go and
look at the input file, for example in Ansys, though you would have defined it as two
surfaces, you will see that he would have split it up and he would have written some
say, in this case, in this case what he would have done in say a package like Ansys, is
that he would have defined 1 2 3 4 this surface as one of the surface and 20 21 as
another surface. You would have defined only two surfaces.
You may think that it is some surface to surface contact algorithm and that the
algorithm will not allow this surface to penetrate into this surface. No; it is not like
that. All these algorithms are node to surface contact, node to surface contact.
In fact if you had seen inside, go and see the data file, you would see that he would
have defined 1 20 21, 2 20 21 and so on. Is that clear? That is the first point that they
are node to surface algorithms. The problem in contact algorithm is that the
algorithms stiffness matrix is written in such a fashion that the slave nodes are not
24
allowed to penetrate. But, unfortunately the master nodes have no such restrictions.
Master nodes are allowed to penetrate. Is that clear? Before we go to that point, how
does or how does this whole system of contact element operate? It is very simple
actually, in the sense that, I am not going to the mathematical derivations, because the
theory that is required is quite high and hence I am not going into the mathematical
derivations of the stiffness matrix. That is not required also, unless you are going to
work in contact.
The only thing I want you to know is that, the size of the stiffness matrix you can
very easily understand, that the ds, what are ds? Understand that they are the
degrees of freedom. The degrees of freedom that are associated with a slave node is 6
in number.
(Refer Slide Time: 40:00)
Suppose, I say 1 2 and 3, then u1v1 u2v2 and u3v3 and hence K matrix is now
correspondingly, what? 6 by 6 and hence there is a force also. Correspondingly there
are 6 forces. In other words, in other words, contact element can also take into
account friction. There is friction which can also be taken into account contact
element and that is the reason why I have displacement in the x direction as well, so,
friction. There is a normal stiffness Kn as well as shear stiffness Ks. Normal stiffness
25
comes from penalty; both of them come from penalty and the shear also take into
account friction and so on. Point number 2 is that contact elements are capable of
taking into account friction as well. Is that clear?
Please note that when I define the contact element, I need to, I need to make the
algorithm recognize which is the segment in which it is operating. It is very important
to realize that the algorithm inside a software would search and find out which is the
segment of the master element or the rigid body that it is operating. What is it? Let
me explain that more deeply.
Due to some reasons, instead of saying 20 and 21, I can say that, I can divide the
master into say two elements, say I say 20 21 and say 23. One of the jobs of the
algorithm in any of the softwares that you may use is to find out whether this 1 say 2
3 4, these guys sit or come under the influence of which of the master segments?
There are now two master segments. Remember that 20 21, 21 23. There is what is
called as, in software they call this as contact searching algorithm, contact searching
algorithm. The contact searching algorithm will find out whether one sits in 20 21 or
in 21 23, so that the contact element is formed between 1 20 and 21; 3 is formed
26
between 3 21 and 23. Note that carefully, 3 will be formed between 3 21 and 23 and
so on. That is nice, you know.
You may think that is very simple. But, suppose this block has started moving due to
application of your load.
Suppose you can move the block. There is absolutely no problem with friction,
because the stiffness matrix is such that the singularity will be removed and you can
move the block. There will be a contact force, there will be a friction force. But if the
force is such that, there will be some normal forces here of course and then suppose
you are moving this block. It is possible the This is possible only if you have a
friction or contact algorithm. Now what will happen? That 2 chap, this fellow who
was under the influence of 20 21 may move to the other segment, 21 23.
Initially though you define, that is why you define it as a complete surface. You have
to be very careful to see to it that you define all the surfaces, for which there is
likelihood for the master segment to come into contact later during loading. Is it
27
clear? The next step is for you to find out whether the segment, which segment it falls
and so on. Now what happens?
Now, this particular list u1 u2 u3, initially for example for 2, it would have been u20 v20
u21 v21 and u2 v2; it would have been like that. But when this 2 moves to the segment
21 23, this would become now u21 u23, sorry v23 and so on; u21 v21 u23 v23 and again u2
v2. Is it clear? The corresponding stiffness, where he assembles would also be
different.
28
This will not happen, please note this would not have happened in our earlier finite
element formulation, because if this had been say 1 2 3 sorry 4, 5 6 or sorry yeah, 5 6
and I defined an element 1 2 6 5, I define it as 1 2 6 5, that would be 1 2 6 5. When I
assemble it, I will not have any problem. Here, note that the master and the slave,
slave may remain the same or the master segments would change. What is the
practical lesson behind it? Practical lesson is that you exhaust as a master segment,
for a particular node all the segments that are possible in its history, as you load it as a
master segment. In other words, again explaining it practically what it means is that if
I just give 20 21 for the node 2 and forget about 21 23 and if it starts moving, when
this fellow goes here into this segment, the fellow will go down straight away,
because you would not have given 21 23 as a master segment. These are niceties,
which we have look at the software which you define or which you are using in order
to do that. Yeah, any question?
Fantastic; so, when the slave node and the master node coincide there are bound to be
certain small problems, but when we write the code we see to it that there is some sort
of a small tolerance that we give, to put it either in this segment or that segment. But
when you write the code this is a problem. Many times, many times this would be,
29
this would be a problem. There are a lot of places where a node would penetrate.
What you say for example, it is a very good problem. It happens in many softwares
also. The software people who write the software should take care of it. Let us say
that, where is my duster? Yeah; let me take the same problem. Let us say that I am
doing an axi-symmetric problem.
This is the axis of symmetry, that is a axis of symmetry. This you say, what does it
mean? It means, it is a ring. The ring is now sitting on a block. This is a master node,
that is a master node many times and there is a node, corresponding node, sitting in
the axi-symmetric part of it. If both the nodes have the same coordinates, have the
same coordinates, this node, this slave node will start penetrating inside. Many times
this would happen and hence people who write softwares will adjust this. This is only
a beginning. There are lots of issues here in contact. We will again look at contact in
the next class and also certain nonlinear algorithms.
30