forums.silverfrost.com

jjgermis · Posted: Fri May 18, 2012 12:11 pm Post subject: Using OpenGL to plot a mesh

In the past I used Matlab or Octave for visualisation of the meshes [see 3]. However, this often is a waste of time since it takes ages to display the meshes (graphics) - my laptop at home is not up to date. Using OpenGL is amazingly (very) fast. Moreover, I can use it under Windows and Linux [see 4]. The example below shows a simple mesh.

Questions
1.) How can I make the mesh visible, i.e. a black outline for each element;
2.) assure that the aspect ratio is update automatically when the user resize.

DanRRight · Posted: Sun May 20, 2012 3:21 pm Post subject:

1) For visible mesh add one line (i think after CALL glMatrixMode(GL_PROJECTION) )

call glPolygonMode( GL_FRONT_AND_BACK, GL_LINE )

2) to return back fill view from mesh view call

call glPolygonMode( GL_FRONT_AND_BACK, GL_FILL )

instead. I do that using different mouse buttons (i use right mouse for mesh view and left mouse for fill view) or a %rb button with the key (see "Mesh View" on the panel in MESH.FOR example). There also could be a way to plot both simultaneously, otherwise you will need to plot twice -first the fill, then the line mesh

3) For manipulation of view you better need to read and interpret mouse position and mouse clicks. Add callback function which is doing that like CBmouseGL01 in my mesh example

i=winio@('%pv%^og[double,depth16]&',640,600,CBmouseGL01)

4) For resizing you will need just to change z if i am not forgotten - in the demo it is done by pushing middle mouse button and simultaneously moving the mouse. see what is done there after this line

if(reasonMouseGL01.EQ.'MOUSE_middle_CLICK'.or.iflag_mouseGL01.eq.16) then

5) to visualize your own mesh you do not need even to change anything in MESH.FOR demo, you just need to read your x,y,z the way code understands it when you drag and drop the input file for it. I removed drag and drop option just for simplicity, but will add it. You can hard wire that with your own programming additions meantime since you've already done principal part of the job, everything else is just the evolutionary beautifications

The way i currently setup input in drag and drop is to define (x,y) for each node and z for its value
x1,yA,zD
x1,yB,zE
x1,yC,zF
.......
x2,yK,zP
x2,yL,zQ
x2,yM,zR
.......
x3,yS,zT
.....
That's because x in my case is always time slice which is equal for all y,z in a given moment

If you tell me how you do the output for your mesh i will probably make view arbitrary meshes possible when will have some free time in a month or so (reviews, reports, and conferences time! I will do that just for fun because i know the result will look a-m-a-z-i-n-g). For example if you define each (x,y) for the node of the triangle and one value z for all of 3 nodes of a triangle it may look like this
x1a,y1a,x2a,y2a,x3a,y3a,z1a
x1b,y1b,x2b,y2b,x3b,y3b,z1b
.......

or some different way when you define x,y,z for each node

x1a,y1a,z1a,x2a,y2a,z2a,x3a,y3a,z3a
......

jjgermis · Posted: Mon May 21, 2012 12:28 pm Post subject:

Hi Dan

Thanks for the tipps. I found the part in the MESH.FOR where the z-values are plotted (as polygons). This means I have to replace this part with my own data.

The rotate and zoom functions works perfect - just what I was looking for. I also figured out where the beep came from - sound@ Wink

jjgermis · Posted: Wed May 23, 2012 9:42 am Post subject:

jjgermis · Posted: Wed May 23, 2012 10:09 am Post subject:

The original example I got from Dan without any functionality is given below. This offers the basic framework for the mesher (I am working on).

Many ClearWin functions needs a call-back function. As I unterstand it, one needs a function as parameter. To do so one could either:
1.) define the function as external (this I assume is Fortran 77 style); or
2.) as an interface (this seems to be 95 onwards).

Question
Is this the correct understanding of external/interface in the context of a function as a parameter?

DanRRight · Posted: Thu May 24, 2012 12:17 am Post subject:

If use Fortran90 style I usually put all callback functions into the module after CONTAINS and USE it, then you do not need to declare them as external

jjgermis · Posted: Thu May 24, 2012 3:31 pm Post subject:

For a test I plotted the same function in Matlab (or Octave). Changing the view requires some time and everthing is much slower than the same example using OpenGL.

Initially I thought the example is nice but nothing for me. In the meanwhile I know that this is exactly what I need. For each triangle I have the (x,y) coordinates and the vector potential, i.e. (x,y,z) for each node and the triangle node information. Between the glBegin() and glEnd() I only have to specify three vertices (i.e. call glVertex3f) instead of four as in the given example. Moreover, I can do the following:
1.) for the mesh only I can use glVertex2f; and
2.) for the vector potential solution glVertex3f.

jjgermis · Posted: Sun May 27, 2012 10:08 am Post subject:

Dan, I tried you advice and it was easy to plot the mesh! Below is an image of the result.

You mentioned that you used examples you found in the FTN95 Help. Functions in the slab animation are similar. However, I tried this in the past and could not get it to work. I now realized that the example in the online help requires an include (that is not provided). In the FTN95 example directory one will find another slab example.

Question:
What options do I have to print/save the image. I would prefer encapsulated postscript, but jpeg will do as well.

Uploaded with ImageShack.us

jjgermis · Posted: Mon May 28, 2012 7:45 am Post subject:

Hi Dan, below is the subroutine I use for reading the mesh. The basic format is as follows:
1.) line 1: nde
2.) line 2: nelm, nnode, testa, nlines
3.) nelm lines: (node(i,j),j=1,nde), ITYPE(i), (NEIGH(i,j),j=1,NDE)
nodes of triangle, element type, neighboring triangles
4.) nlines lines: (LINE(i,j),j=1,2)
begin en end nodes of each line segment
5.) nnode lines: x(i), y(i)
node coordinates

Definitions:
nde: number of nodes (always thee for triangular mesh)
nelm: number of elements
nnode: number of nodes
testa: testa=0 -> no vector potential at node
testa=1 -> vector potential at node
node: triangle nodes
itype: type of material (is an integer)
neigh: neighboring nodes of a triangle
lines: line segments
x: x-coordinates
y: y-coordinates

JohnCampbell · Joined: 16 Feb 2006 Posts: 2615 Location: Sydney

Why do you read NEIGH ?
You should find this automatically;
For each edge of a triangle;
find the other triangle that shares this edge. ( there should be only one)
if it exists, then select the other triangle's other node
else nominate this edge as an edge to the region.
Edges could be defined as ordered nodes, n1:n2 where n1 < n2

An outcome of this analysis would be to identify:
all external edges in the model
all edges at a boundary between materials ( other triangle has different itype)
( else internal edge, else duplicte connection edge )

This would be a useful subset of edges to draw, confirming some basic geometry. You could draw the 3 (or 4) types of edges with a different colour to validate the model.

John

DanRRight · Posted: Tue May 29, 2012 7:27 am Post subject:

Do you make it in 3D or just 2D? Do you color it? Post here, it should be looking very nice

Does animation in my mesh example work? With Salford own OpenGL examples i do not need to add anything, i only commented last paragraph in opengl.ins which always caused my OpenGL not to work at all (including i think up to right today)

As to saving image -- i just copy it from the screen. I also have saving function somewhere programmed into the code (John Horspool suggested here in the forum how to do that exactly, as i remember) but could not find it

jjgermis · Posted: Tue May 29, 2012 12:08 pm Post subject:

Hi John and Dan,

thanks for your comments!

Indeed, I actually never (directly) use the neigh variable - it was part of the format I use to come into when I started with the FEM code. I do have somewhere a subroutine outline, which determine the boundaries as you mentioned (I need to search at home for the subroutine - will post it). Since I work with rotational machines this allows me to automatically detect the inner/outer radius as well as the periodic boundaries.

Dan, your example works perfect - including the animation. The actual mesh is 2D - since I only work with 2D field problems (in the screenshot above you can see the function in your original code). The vector potential solution is then visualised in 3D, but here most people prefer the field lines in 2D. I prefer the 3D view since can then visually check if the solution is fine (actually some time ago a found a bug in the solution that was only clear by viewing the 3D plot!).

At present I also do a screenshot. It would however be nice to have an menu with save as...

jjgermis · Posted: Tue May 29, 2012 6:59 pm Post subject:

Hi John,

in a previous thread I posted the code for the very simple mesher I had. Once the triangles are known, the following code calculates the outline and the neigbors. The fpl is only a type declaration for all the mesh/solution variables - this makes handling the data much easier.

DanRRight · Posted: Tue May 29, 2012 10:15 pm Post subject:

here is a link about saving opengl screen.

http://forums.silverfrost.com/viewtopic.php?t=2103&highlight=opengl

Nice would be to see saving OpenGL screen the way it is done in regular graphics

DanRRight · Posted: Wed May 30, 2012 11:04 pm Post subject:

Don't forget to make small code examples while you learn OpenGL so that more people will come after your steps. Hopefully there will be the whole community of OpenGL programmers from which we will learn quicker then from all manuals. May be some day someone will create Visual OpenGL in Visual Clearwin so that we will stop programming and will just assemble our programs with mouse clicks or finger taps like from lego pieces.