Dense packing for stone masonry wall generation

Wang, Q., dos Santos, K. R., & Beyer, K. Geometric planning for masonry wall construction with natural stones using discrete convolution. Engineering Structures

@article{Wang_2025, 
title={Geometric planning for masonry wall construction with natural stones using discrete convolution}, volume={329}, 
ISSN={0141-0296}, 
url={http://dx.doi.org/10.1016/j.engstruct.2025.119695}, 
DOI={10.1016/j.engstruct.2025.119695}, 
journal={Engineering Structures}, 
publisher={Elsevier BV}, 
author={Wang, Qianqing and dos Santos, Ketson R.M. and Beyer, Katrin}, 
year={2025}, 
month=apr, 
pages={119695} }

Important Tips

👉 A 2D version of the generator is able to generate dense and stable stacked stones given pixelized shapes 2D stacking.

🤖 The generator has been used as geometric planner for a real wall construction work using robots robot wall.

💡 The convention of the packing space is that the X direction is the Height direction.

💡 Make sure the scale of the input stone meshes matches the order of magnitude of the wall’s dimensions.

Dataset

Some pre-generated stone walls can be found here 👉 zenodo.

Set up Environment

Tested on Ubuntu 20.04.6 LTS.

Step 1: Build Virtual Environment

python -m venv ./venv/
source ./venv/bin/activate

Step 2: Install dependencies

pip install -r requirements.txt

Usage

Examples

Packing

python main.py --dataID 19 --stoneRep available_stones --stoneIni stone_

Packing followed by prerotation

python prerotate_stones_flat.py --dataID 19
python main.py --dataID 19 --stoneRep available_stones_rot --stoneIni S

Packing considering multiple poses per stone

python prerotate_stones_flat.py --dataID 19
python main.py --dataID 19 --stoneRep available_stones_rot --stoneIni S --nPose 4

Input and output

The input (stone meshes) are triangular suface meshes in .obj format stored in ./data/example*/[stoneRep]/.

The output files are :

• stone with id m placed in iteration n in its resulting position represented in triangular mesh: wall_1_iteration[n]_valid_refined_[m]_best_pose_random_sequence.obj

• stone with id m placed in iteration n in its resulting position represented in voxel: wall_1_iteration[n]_valid_voxel_[m]_best_pose_random_sequence.ply

• Failured placement at iteration n will generate stone meshes in their initial position: wall_1_iteration[n]_invalid_stone_[m]_best_pose_random_sequence.ply

Features

• wallDimX/Y/Z

Switching the size of the bounding box of the packing space for generating stone walls of different sizes.

• wallMaxX

Using a mask as top boundary of the packing space such that stones cannot be placed higher than this limit. This is useful if you want to enforce a layer-wise packing.

• nPose

Number of candidate poses to evaluate per stone. Candidate poses are generated by uniformly rotating the stone around the x-axis from 0 to 360 degrees.

• seq

The sequence of stones provided to the algorithm can be random or sorted by volume from largest to smallest.

Preprocessing tools

The script prerotate_stones_flat.py rotates stone meshes such that the shortest dimension of the stone is aligned with the X direction of the packing space, the other two main dimensions of the stone are aligned with the Y and Z directions of the packing space.

Visualization

Triangular meshes can be visualized using Meshlab.

Command Line Switches

./main.py [OPTIONS]
OPTIONS:
--dataID INT                    The ID of the dataset in ../data/example[dataID] 
--wallDimX INT                  The dimension of the wall in X direction (default: 70)
--wallDimY INT                  The dimension of the wall in Y direction (default: 70)
--wallDimZ INT                  The dimension of the wall in Z direction (default: 40)
--wallMaxX INT                  The maximum packing height in X direction (default: None)
--stoneRep TEXT                 The folder name of the stone representation 
                                (default:available_stones_rot)
--stoneIni TEXT                 The prefix of the stone file names (default: S)
--nPose INI                     Number of candidate poses to evaluate per stone (default: 1)
--seq TEXT:random/volume        The sequence of stone placement: random or volume (default: random)

Contact

For questions and feedback, please contact:

Qianqing Wang at Earthquake Engineering and Structural Dynamics (EESD)

📧 Email: [qianqing.wang@epfl.ch]