Todo
update
During the runtime of simulations, the user needs to keep track of the simultation state without being obliged to write out the complete domain to disk. This is achieved by tracking objects, at which the state is evaluated at certain defined intervals and then written to disk.
We support three formats for tracking.
Transient data: simple gnuplot-style A simple ascii file which creates a small header and then writes at each defined time step a column wise representation of all variables and points. This is not efficient and should only be used for simple point-tracking objects. Each process writes its own file.
output = {
format = 'ascii',
},
folder = './tracking/'
Spatial Data: This format stores solution data along with coordinates in single data file for every timestep. Therefore, there would be multiple files depending on the total number of time-steps. It is ideal for tracking line and plane.
output = {
format = 'asciiSpatial',
},
folder = './tracking/'
Writes all participating elements to a treelm mesh and creates for each
required time step a restart file. The output is done in vtk
and pvd
files
that can be visualised with Paraview
.
This is a scalable tracking object which works regardless of number of processes.
output = {
format = 'vtk',
},
folder = './output/'
Note
We strongly encourage you to use the Harvester format. We will demonstrate each format in the examples of tracking point, line and plane in the following sections.
All variables defined in the tem_varSys_module can be tracked. See tem_variable_module for how to define variables. @note Note Examples are given here.
derived variables: all variables defined in the global variable system can be tracked
spacetime: as analytical result, can be defined in lua file
tem_reduction_spatial_module reductions, e.g. difference, sum or average of defined variables
We show the tracking with simple shapes here. Get an overview over all existing shapes in tem_shape_module and examples are given here
So far point, line, plane are implemented in tem_shape_module.
The user specifies the desired parameters in the configuration file (in the scheme table) as
-- if you want to have a analytical solution as reference, you can add
-- a spacetime variable first
-- define single new (spacetime) variable
variable = {
{
name='spacetime',
ncomponents=1,
kind = 'st_fun',
st_fun = luaFun1
},
{
name='spacetime2',
ncomponents=1,
kind = 'st_fun',
st_fun = luaFun2
},
{
name='dens_diff',
ncomponents=1,
kind = 'operation',
operation = {
kind = 'difference'
input_varname = {
'density','spacetime'
}
}
},
{
name='wss_diff',
ncomponents=1,
kind = 'operation',
operation = {
kind = 'difference'
input_varname = {
'wss','spacetime2'
}
}
}
}
-- Tracking
tracking = {
{
label = 'probe_error',
variable = {
'velocity','dens_diff','wss_diff'
},
folder = './tracking/',
output = {
format = 'ascii'
},
time = {
interval = 1,
min = 1000,
max = 4000
},
reduction = {
'l2norm',
'l2norm',
'l2norm'
},
shape = {
kind = 'canoND',
object = {
origin ={0.,0.,0.},
vec = {1.,0.,0.},
segments={100,100,100}
}
}
}
}
Note
More examples for how to use the variable System can be found here.
The tem_reduction_spatial_module takes a set of input quantities and reduces them according to the chosen operation. You can choose between the following operations: - average: Average of all occurring values - sum: Sum of all occurring values - max: Maximum of all occurring values - min: Minimum of all occurring values - l2norm: L2 norm of all values, i.e. L_2(x) = sqrt( sum( x_i^2 )) - linfnorm: L_inf norm of all values, i.e. L_inf(x) = max( | x_i | )
reduction = {
'norm' -- 'average' or 'sum'
}
Note
The table of reductions has to correspond to the number of entries in the tracking variable table. For each variable, specify a reduction as
tracking = {
variable = {'pressure', 'velMag', 'density'},
reduction = { 'average', 'max', l2norm' }
}
Here, a point at the location {0.,0.,0.} is defined as a pressure probe. The interval is set to 1, meaning at each time step, the pressure has to be written. The probing starts at simulation time 1000 and lasts until 4000. The tracking write both pressure and velocityX in a single file. Use ref_value to normalize the output.
# Name: GaussPulse_TestSimulation Date 2011-060-27 Time 11:43 CEST
# Probe type: Point at {0.0 0.0 0.0}
# Quantities: pressure, velocityX
# tMin: 1000, tMax: 4000, Interval: 1
# ref_value: 1.0, 0.01
# iTime rTime Pressure velocityX
1000 1000.0 1.0241818 0.00158287
1001 1001.0 1.0251427 0.00152617
...
4000 4000.0 1.7162511 0.01273673
The shape is defined as a canonical object. The different shapes are basically derivatives of a simple cube. A cube with no length in any direction is a point, a cube with length along only one axis is a line whereas a cube with length along two axes is simply treated as plane by the underlying code. One example each for Point, Line and Plane is shown below.
Example: tracking a point in ASCII format
tracking = { label = 'Velocity',
variable = {'velocity'}, -- options: density, velocity
-- Tracking over a point
shape={kind = 'canoND', origin = {0.5,0.5,0.5} },
time = {min = 0, interval = 1}, -- if no max given: active for all times
output = {format = 'ascii'},
folder = 'tracking/'}
}
Once, the above section is executed within the musubi.lua file, an output in the asciiTransient format is generated in the output folder specified (tracking in this particular example). Now, we open the result file inside tracking folder, the structure of which looks like below:
# Rank of the process: 0
# time velocity_01 velocity_02 velocity_03
1.0000 0.000000000000 0.000000000000 0.000000000000
2.0000 0.000000000000 0.000000000000 0.000000000000
3.0000 0.000000000000 0.000000000000 0.000000000000
4.0000 0.000000000000 0.000000000000 0.000000000000
...
...
1997.0000 -0.006065558329 0.000443400622 -0.000003481789
1998.0000 -0.006063901533 0.000443795703 -0.000003286949
1999.0000 -0.006061596302 0.000443327890 -0.000003534752
2000.0000 -0.006059665212 0.000443720604 -0.000003339771
Here, it can be noticed that 3 velocity components for all time steps at the point (0.5, 0.5, 0.5) are stored. The interval was chosen as 1, so they are tracked for each time step. If the interval was chosen as say 10, then the tracking would have been done after 10 units of simulated time.
We need to set the origin and the direction vector,
as e.g. {origin={0.,0.,0.}, vec={1.0,1.0,1.0}}.
The line is represented as a list of points in the tracking facility.
The number of subdivisions and its distribution of the points on the line
needs to be set, as e.g. segments=500, distribution='equal'
.
Make sure to specify enough segments to take into account all the required
elements on the line.
Elements are added only once, so it might be beneficial to specify much more
segments than actually needed to consider elements on the highest levels as
well.
Example: Tracking a line in ASCII Spatial format
tracking = {
label = 'line_probe',
variable = {'state'}, -- state (=pdfs), pressure, density etc
shape = { kind = 'canoND',
object={
origin = {0.5,0.5,0.5},
vec = { {0.0,0.2,0.0}},
segments = {300},
distribution='equal'}
},
output = {format='asciispatial'},
folder='tracking/',
time = {interval = 5, min = 0, max = 800 } }
}
After, the above tracking is executed within musubi.lua file multiple res files are obtained inside the tracking folder each specific to particular time step. If you open any one of the 'res' files, there will be six data columns: three coordinate and three velocity components. An example is shown below:
# Rank of the process: 0
# coordX coordY coordZ velocity_01 velocity_02 velocity_03
0.51562500 0.51562500 0.51562500 -0.000998448727 -0.000118182367 0.000007560959
0.51562500 0.54687500 0.51562500 -0.000922746216 0.000011023243 -0.000000130462
0.51562500 0.57812500 0.51562500 -0.001083956104 0.000038534363 -0.000003774649
...
0.51562500 0.64062500 0.51562500 -0.001343981219 -0.000131107478 -0.000004381864
0.51562500 0.67187500 0.51562500 -0.001261956611 0.000081027056 0.000000215364
0.51562500 0.70312500 0.51562500 -0.001733303068 0.000060154818 0.000005793207
Out of three coordinate components, only 'coordY' is varying and 'coordX',
'coordZ' remain constant. This is according to our tracking input with
origin = {0.5, 0.5, 0.5}, vec = {0.0, 0.2, 0.0}
.
You can now make a line plot with any one of velocity components with
respect to 'coordY' in gnuplot.
We need to set the origin and the direction vectors, as e.g.
object = {origin={0.,0.,0.}, vec={{1.0,1.0,1.0},{0.0,-1.0,1.0}}}
.
The plane is, just as the line, represented as a list of points in the
tracking facility.
The number of subdivisions in each direction and its distribution of the
points on the plane needs to be set,
as e.g. segments1=500, segments2=300, distribution='equal'
.
Make sure to specify enough segments to take into account all the required
elements on the line.
Elements are added only once, so it might be beneficial to specify much more
segments than actually needed to consider elements on the highest levels
as well.
Example: tracking a plane in harvester format
tracking = {
label = 'plane_probe',
variable = {'state'}, -- state (=pdfs), pressure, density etc
shape = { kind = 'canoND',
object={
origin = {-75.,-15.,0.},
vec = { {150.,0,0.}, {0.,30,0.}},
segments = {300, 300},
distribution='equal'}
},
output = {format='vtk'},
folder='tracking/',
time = {interval = 5, min = 0, max = 800 } }
}
It will result in vtk
files at the given time steps and one pvd
file
that is an animation file over all vtk
files.
Example: tracking the whole mesh
tracking = {
...
shape = { kind = 'all' },
-- or:
shape = { kind = 'global' },
-- or:
shape = { kind = 'globalmesh' },
...
}
Example: tracking the boundary elements
tracking = {
...
shape = { kind = 'property', property = {'boundary'} },
...
}
-- or:
tracking = {
...
shape = { kind = 'boundary', boundary = {'bnd_name1','bnd_name2'} },
...
}
Example: tracking solid elements
tracking = {
...
shape = { kind = 'property', property = {'solidified'} },
...
}
The user-specified tracking shapes have to be translated to a set of elements, on which the defined quantities are calculated and written to disk in a suitable format.
All elements are converted to a set of points.
For each point, the correspoinding treeID is added to the element list of
the tracker as track%elems( iLevel )%elem(1:nElems)
.
Each element will only be added once.
All these elements are then treated in the main loop within the tracker
routine.
Each solver comes with a harvesting tool that allows the postprocessing
of restart data after the simulation has been run.
If this harvesting is to be done on a set of files, this may become a
little tedious.
To ease the post-processing there is a small Python script available in
peons/harvest_series.py
, which takes care of the execution of the
post-processing script for multiple restart files.
It can be used as easily as running
peons/harvest_series.py restart_*.lua
The main argument this script expects is a list of files to work on. There specification may make use of shell globbing patterns as in the example above.
Aside from the files to process there are various other options that
can be specified, like the name of the harvesting executable, the
name of the Lua executable and the template to use to construct the
input file for the harvesting tool.
These options may always be provided as command-line arguments, but
can also be stored in a configuration file. By default harvest_series
will look for a file named series.config to read its configuration,
but you can tell it where to look for it via the -c
option.
In the configuration file you provide the options to use in the form of:
option: value
Thus, for example you could provide the name of the harvesting tool as follows:
harvester: build/atl_harvesting
See the Python ConfigParser module documentation for more details on the syntax of the configuration file. You may provide a list of files to process in the config file. In this case the files should be separated by commas, you may still use globbing patterns in the file name specification.
Have a look at the available options with the -h
option:
peons/harvest_series.py -h