The methodological framework, Iterative Classification for Graph-Set-Based Design (IC-GSBD) employs an iterative approach to efficiently narrow down a graph-based dataset containing diverse design solutions to identify the most useful options. Utilizing Geometric Deep Learning (GDL) through PyTorch Geometric, we analyze a small subset of the dataset to train a machine learning model which is then used to predict the remainder of the dataset iteratively, progressively narrowing down to the top solutions.
The current data consists of two MATLAB .m1 files.
The first contains over 43,000 graphs that represent Analog Electric Circuits.2
The second contains over 32,612 graphs representing aircraft Thermal Management Systems (TMSs).3
Each circuit contains an adjacency matrix, node labels representing the various components, and a graph-level label representing the performance of the circuit.4
Most of the libraries used are pre-built into Python, but please ensure that you have the following:
The primary machine learning tools required to be installed are:
Once the files are downloaded, the run.py file is the primary script where you can adjust particular variables such as save paths.
This project is still ongoing, and more engineering architecture data will become available in the near future.
Please cite my dissertation (and the respective papers of the methods used) if you use this code in your own work:
@phdthesis{Sirico2024c,
author = {Sirico Jr, Anthony},
title = {Integrating geometric deep learning with a set-based design approach for the exploration of graph-based engineering systems},
type = {{Ph.D.} {Dissertation}},
school = {Colorado State University},
address = {Fort Collins, CO, USA},
month = aug,
year = {2024},
pdf = {https://www.engr.colostate.edu/%7Edrherber/files/Sirico2024c.pdf},
}
If you notice anything unexpected, please open an issue and let us know. If you have any questions, feel free to discuss them with us.
Footnotes
-
MATLAB is not required to run this script. But, if you wish to manipulate any of the raw data, MATLAB or Octave is required. ↩
-
A. Sirico Jr, D. R. Herber. "On the use of geometric deep learning for the iterative classification and down-selection of analog electric circuits." ASME Journal of Mechanical Design, 146(5), p. 051703, May 2024. doi: 10.1115/1.4063659 ↩
-
A. Sirico Jr, D. R. Herber. "Geometric deep learning towards the iterative classification of graph-based aircraft thermal management systems." In AIAA 2024 Science and Technology Forum and Exposition, Orlando, FL, USA, Jan 2024. doi: 10.2514/6.2024-0684 ↩
-
D. R. Herber, T. Guo, J. T. Allison. 'Enumeration of architectures with perfect matchings.' ASME Journal of Mechanical Design, 139(5), p. 051403, May 2017. doi: 10.1115/1.4036132 ↩