Voxelization Toolkit

Voxelization Toolkit for (IFC) Building Models

Introduction

(Building) models are widely different depending on modeller, phase, intended use, software application and version. Conventional BRep and Mesh geometry libraries can have robustness issues coming from floating point rounding errors and tolerances. For these reasons we provide building model analysis functionality using voxels. Comprehensive analysis regardless of modelling choices is enabled by robustly discretizing the complete building model geometry. The integer math is robust and easy to understand.

Use cases

The toolkit presented here has been used for:

• Reliably compute volume (e.g entire building volume above and under ground; quantity calculation for non-manifold element geometries)
• Evacuation analysis (providing a length to nearest exit for every reachable place in the model; taking into account obstacles)
• Building code compliance (e.g of space dimensions taking into account contained elements)
• Safety analysis (by analysing reachable space and fall hazards)
• Detection of exterior elements (by calculating interior volume using neighbour traversal)
• Model conversion (e.g for acoustic analysis - eroding small elements - using PCA to export slanted surfaces derived from voxel grid)
• ...

Examples

Three examples (including visualization) are elaborated in python code in ./python/voxec

Headroom height

Calculate headroom height in the model taking all geometric elements into account

Evacuation distance

Calculate distance to the exterior for any point in the model taking into account obstacles

Exterior elements

Geometricaly assess exterior elements on multi-aspect models as BIM authoring tools often have erroneous IsExternal properties specified.

Installation

The voxec(.exe) runtime interpreter can be installed using the Anaconda (conda) package manager from the ifcopenshell channel.

conda install -c conda-forge ifcopenshell=v0.7.0
conda install -c ifcopenshell voxelization_toolkit

Quick tutorial

0. Perform the installation steps above
1. Create a new empty folder with:
   • voxelfile.txt; contents of https://github.com/opensourceBIM/voxelization_toolkit/blob/e9ce7e0c5569da67c9953bcd4e6496b5ce1c3699/tests/fixtures/voxelfile1.txt
   • duplex.ifc; contents of https://github.com/opensourceBIM/voxelization_toolkit/blob/e9ce7e0c5569da67c9953bcd4e6496b5ce1c3699/tests/fixtures/duplex.ifc
2. cd to the directory in powershell/bash and run: voxec voxelfile.txt
3. Observe an output is printed like below as the flow of the procedure defined in voxelfile.txt is executed.

[                                                  ]
  severity: notice
  message: executing {statement}
  time: 2024:10:25 14:36:43.283
  statement:
    text: file = parse("duplex.ifc")
[#####                                             ]
  severity: notice
  message: executing {statement}
  time: 2024:10:25 14:36:43.392
  statement:
    text: surfaces = create_geometry(file)
[###########                                       ]
  severity: notice
  message: executing {statement}
  time: 2024:10:25 14:36:44.575
  statement:
    text: slabs = create_geometry(file,include={"IfcSlab"})

4. Now add a line: export_point_cloud(outmost_voxels, "outmost_voxels.obj")
5. and rerun voxec.
6. The exported file can now be visualized for example in Blender.

Usage

Recommended usage is using the voxec runtime interpreter, which takes an input "voxelfile" which describes a series of commands. The grammar of such as voxelfile is:

iden = alpha (alnum | '_')*
statement = iden '=' function_call
quoted_string = '"' (char - '"')* '"'
sequence = '{' quoted_string? (',' quoted_string)* '}'
value = strict_double | int_ | quoted_string | iden | sequence
function_arg = iden '=' value
function_args = function_arg? (',' function_arg)*
function_call = iden '(' function_args ')'
argument_list = iden? (',' iden)*
function_def = "function" iden '(' argument_list ')' statement* "return" iden
start = (statement | function_def)*

For example:

file = parse("duplex.ifc")
slabss = create_geometry(file, include={"IfcSlab"})
roofss = create_geometry(file, include={"IfcRoof"})
slabs = voxelize(slabss)
roofs = voxelize(roofss)
floors_surface = subtract(slabs, roofs)
floor_volume = volume2(floors_surface)
floors = union(floors_surface, floor_volume)
floors_surface = collapse(floors, 0, 0, -1)
num = count(floors_surface)

A list of available commands is provide below.

Available commands:

assert()

Asserts that a certain the voxel grid or integer value specified in input represents a count greater than zero.

name	required	type
input	Y	voxels,integer

collapse()

Reduce the volume of components along the direction vector to 1. Useful to calculate floor area by eliminating the thickness of floor slabs. This is the reverse of extrusion.

name	required	type
input	Y	voxels
dx	Y	integer,real
dy	Y	integer,real
dz	Y	integer,real
until	n	voxels
max	n	integer

collapse_count()

Same as collapse() but return voxel grid of type uint32 with the number of collapsed voxels.

name	required	type
input	Y	voxels
dx	Y	integer,real
dy	Y	integer,real
dz	Y	integer,real
until	n	voxels
max	n	integer

component_foreach()

Apply a function to each connected component individually

name	required	type
input	Y	voxels
function	Y	string
argument	Y	string

constant_like()

Return a constant voxel grid of type type with value value with the same dimensions as input.

name	required	type
input	Y	voxels
value	Y	integer
type	n	string

copy()

Return a copy of input, optionally with a different type.

name	required	type
input	Y	voxels
type	n	string

count()

Count the number of non-zero voxels in input. This is information is present as meta-data on most voxel region types.

name	required	type
input	Y	voxels

count_components()

Count the number of connected components in input.

name	required	type
input	Y	voxels

count_neighbours()

Returns a new u32 grid with for every voxel the number of neighbours counted, with 6 or 26 for connectivity.

name	required	type
input	Y	voxels
connectivity	Y	integer

create_geometry()

Create geometry from a set of IFC building model inputs (uses IfcOpenShell, result is a set of OpenCASCADE solids)

name	required	type
input	Y	ifcfile
include	n	sequence
exclude	n	sequence
optional	n	integer
only_transparent	n	integer
only_opaque	n	integer

describe_components()

Print information on the connected components in input. Writes JSON information to a new file in ouput_path.

name	required	type
output_path	Y	string
input	Y	voxels

describe_group_by()

Takes the component labels in groups and print information on the components selected by these individual labels in input. Writes JSON information to a new file in ouput_path.

name	required	type
output_path	Y	string
input	Y	voxels
groups	Y	voxels
use_bits	n	integer
conserve_memory	n	integer
only_counts	n	integer

dump_surfaces()

Revoxelize input_surfaces and dump information to output_path when intersecting with input_voxels

name	required	type
input_voxels	Y	voxels
input_surfaces	Y	surfaceset
output_path	Y	string

dimensionality_estimate()

Use PCA to estimate the local dimensionality for individual voxels. The three eigen values that are returned in a new grid describe the degree of variance along the principal axes. When these eigenvalues are roughly equal the voxel van be considered part of a 3d component. If there is one eigenvalue considerably greater than the other two, it constitutes a part of 1d component. Etc.

For each individual voxel a neighbourhood consisting of voxels within a topological distance of max_depth voxels is formed. For voxels on the edge of a component this means an assymetric neighborhood is formed, because the neighborhood extends inwards into the component, but outwards there are no neighbours. This affects the distribution of eigenvalues. In order to get a more uniform distribution of eigenvalues within a component, the reposition=1 can be used, which first takes a neighborhood of max_depth voxels and then takes the mean of that to be the seed of a second neighborhood traversal of max_depth_2 voxels. This effectively moves the center of the neighborhood inwards.

name	required	type
input	Y	voxels
max_depth	n	integer
max_depth_2	n	integer
distance_2	n	real
neighbourhood_size	n	integer
reposition	n	integer

export_csv()

Exports a voxel grid to CSV.

name	required	type
input	Y	voxels
filename	Y	string

export_ifc()

Exports a subset of the IFC model in input depending on the whether a revoxelization of input_surfaces (previously computed from input) intersects with input_voxels.

name	required	type
input	Y	ifcfile
input_voxels	Y	voxels
input_surfaces	Y	surfaceset
output_path	Y	string

export_json()

Same as export_ifc(), but don't serialize a full IFC model but rather element identifiers.

name	required	type
input	Y	ifcfile
input_voxels	Y	voxels
input_surfaces	Y	surfaceset
output_path	Y	string

export_point_cloud()

Similar to export_csv() but a different format.

name	required	type
input	Y	voxels
filename	Y	string

exterior()

Computes the exterior of input by starting traversal from a point outside of the bounding box of input.

name	required	type
input	Y	voxels

fill_gaps()

Fills 1-sized holes in input.

name	required	type
input	Y	voxels

fill_volume()

Fills the interior volume of the bounded region in input.

name	required	type
input	Y	voxels

filter_attributes()

Filters the ifc elements in input based on attribute values. Use keyword arguments to specify attribute names and values.

name	required	type
input	Y	ifcfile

filter_properties()

Filters the ifc elements in input based on property values. Use keyword arguments to specify property names and values.

name	required	type
input	Y	ifcfile

free()

Frees storage associated with a voxel grid. Use with caution.

name	required	type
input	Y	voxels

greater_than()

Returns a binary voxel grid where the uint32 values in input are greater than rhs.

name	required	type
input	Y	voxels
rhs	Y	integer

halfspace()

Create a half space (size provided by input) based on the plane equation a b c d.

name	required	type
input	Y	voxels
a	Y	real
b	Y	real
c	Y	real
d	Y	real

intersect()

Returns the intersection of a and b.

name	required	type
a	Y	voxels
b	Y	voxels
if_non_empty	n	integer

invert()

Returns the inversion of input. Should only be used on binary voxel grids.

name	required	type
input	Y	voxels

json_stats()

Writes json statistics of currently defined variables (potentially filtered by variables).

name	required	type
output_path	Y	string
variables	n	sequence

keep_components()

Keep only the connected components in input with size greater or equal to min_size.

name	required	type
input	Y	voxels
min_size	Y	integer

keep_neighbours()

Keeps only voxels with the num_neighbours amount of neighbours. Connectivity should be 6 or 26.

name	required	type
input	Y	voxels
num_neighbours	Y	integer
connectivity	Y	integer

less_than()

Returns a binary voxel grid where the uint32 values in input are less than rhs.

name	required	type
input	Y	voxels
rhs	Y	integer

local_move()

Move the created IFC geometry along their local coordinate axis, can be used to create thicker volumes for doors or windows for example.

name	required	type
input	Y	surfaceset
dx	Y	real
dy	Y	real
dz	Y	real
VOXELSIZE	n	real

local_sweep()

Sweeps the created IFC geometry along their local coordinate axis, can be used to create thicker volumes for doors or windows for example.

name	required	type
input	Y	surfaceset
dx	Y	real
dy	Y	real
dz	Y	real
VOXELSIZE	n	real

mesh()

Create a OBJ triangle mesh from the voxel interior. When use_value is specified create interfaces between segments with different values. When with_components is specified numeric identifiers are assigned to the connected components.

name	required	type
input	Y	voxels
filename	Y	string
use_value	n	integer
with_components	n	integer
groups	n	voxels
with_vertex_normals	n	integer

normal_estimate()

Perform PCA normal estimatation on the voxel grid in input. max_depth specifies the topological connection threshold of the neighbours to incorporate in the PCA.

name	required	type
input	Y	voxels
max_depth	Y	integer

offset()

Returns a new voxel grid where neighbours of voxels in input are set.

name	required	type
input	Y	voxels

offset_xy()

Returns a new voxel grid where XY-neighbours of voxels in input are set.

name	required	type
input	Y	voxels
d	Y	integer,real

outmost()

Returns the outmost (rightmost actually) connected component in input.

name	required	type
input	Y	voxels

parse()

Parse the IFC building model using IfcOpenShell

name	required	type
input	Y	string

plane()

Create a 1-thickness plane from the plane equation in a b c d.

name	required	type
input	Y	voxels
a	Y	real
b	Y	real
c	Y	real
d	Y	real

print_components()

Print information on the connnected components in input.

name	required	type
input	Y	voxels

print_values()

Print information on the values in input.

name	required	type
input	Y	voxels

repeat_slice()

name	required	type
input	Y	voxels
axis	Y	integer
location	Y	real
repetitions	Y	integer

resample()

Resample the voxel grid in input using an integer factor. Use negative values for downsampling.

name	required	type
input	Y	voxels
factor	Y	integer

segment()

Apply segmentation using a voxel grid obtained from normal_estimate().

name	required	type
input	Y	voxels
angular_tolerance	n	real
max_curvature	n	real

set()

Set an individual voxel on grid in input.

name	required	type
x	Y	integer,real
y	Y	integer,real
z	Y	integer,real
input	n	voxels

shift()

Shift the voxels in input along the direction vector dx dy dz.

name	required	type
input	Y	voxels
dx	Y	integer,real
dy	Y	integer,real
dz	Y	integer,real
until	n	voxels
max	n	integer

subtract()

Subtract b from a.

name	required	type
a	Y	voxels
b	Y	voxels
if_non_empty	n	integer

sweep()

Sweeps the voxels in input along the direction vector dx dy dz.

name	required	type
input	Y	voxels
dx	Y	integer,real
dy	Y	integer,real
dz	Y	integer,real
until	n	voxels
max	n	integer

traverse()

Perform traversal on the voxels in input starting from the selection in seed, optionally with along a maximum number of voxels in depth. Operates on 6 or 26 connectivity. Returns a binary voxel grid by default but can also return uint32, in which case, when using 26-connectedness, values are multiples of [10, 10 * sqrt(2), 10 * sqrt(3)].

name	required	type
input	Y	voxels
seed	Y	voxels
depth	n	integer,real
connectedness	n	integer
type	n	string

union()

Returns the union of a and b.

name	required	type
a	Y	voxels
b	Y	voxels
if_non_empty	n	integer

volume()

DEPRECATED use volume2() Returns the bounded interior region of input.

name	required	type
input	Y	voxels

volume2()

Returns the bounded interior region of input. Calculated as the inversal of the exterior.

name	required	type
input	Y	voxels

voxelize()

Voxelizes the geometry obtained from an IFC input file. Use method='volume' to have individual element interiors filled (computationally more complex).

name	required	type
input	Y	surfaceset
VOXELSIZE	n	real
type	n	string
method	n	string

zeros()

Return a new empty voxel grid filled with zeros, same size as input.

name	required	type
input	Y	voxels