A machine learning and robotic simulation framework for the generation and assembly of structures using discrete irregular elements

Date

2025-6-18

Author

Solmaz, Emire Nur

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

4489
views

0
downloads

Cite This

Architectural structures are built by different techniques including masonry which is associated with modularity. The main idea of masonry is composing discrete units with various options that could be both regular/conventional and irregular/non-conventional to build a structure. In terms of irregularity, existing materials like salvage materials can be suitable for innovative configurations that could be challenging due to the precision problem. This research intends to contribute to the literature that addresses this challenge by adopting a stereotomic approach which is a long-established technique for combining discrete units. Machine learning (ML) and deep learning (DL) methods namely, decision tree, K-means clustering, and convolutional neural networks (CNNs) are used to advance the process of creating the intended configurations, and robotics is used to assemble them in a simulation environment. Masonry wall images are selected to comprise the dataset since they could reflect the properties of both stereotomy and stacking. After gathering and eliminating the images based on quality, a total of 393 images were used for the main dataset. A data preprocessing focusing on an edge detection algorithm is applied to extract the features of the pattern that are used to train the decision tree, K-means, and CNN models. A pipeline consisting of these models, mainly the decision tree model is proposed to detect possible base patterns for the final configurations that are created with a placement algorithm. It is shown in this research that the proposed pipeline based on a series of geometric operations, predictions, and a placement algorithm regarding stereotomic principles gives compelling configurations for a set of discrete pieces that can be further assembled by robot simulation.

Subject Keywords

Discrete elements, Stereotomy, Machine learning, Placement configuration, Robotics

URI

https://hdl.handle.net/11511/115466

Collections

Graduate School of Natural and Applied Sciences, Thesis