| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Additive Manufacturing Modeling, Simulation and Machine Learning
|
| Presentation Title |
Data Bridging: A Novel Pipeline for Efficient Statistical Exploitation Across Multiple Data Populations |
| Author(s) |
Alex Gonzalez, Craig A Brice |
| On-Site Speaker (Planned) |
Alex Gonzalez |
| Abstract Scope |
A novel hybrid design of experiments & machine learning pipeline called “data bridging” is introduced as a data-generation alternative to the current statistical-model-centric environment in manufacturing and materials data science. Data bridging studies can extract discoveries beyond individual studies’ data, and the potential number of new findings scales combinatorially with the number of studies being bridged. While the “big data” mentality within manufacturing and materials science pursues larger data sets, not all data is equally helpful. Data bridging allows us to quantify data utility, and augment knowledge specifically based on these resources.
The data bridging process leverages lessons from design of experiments, meta-studies, and meta-analysis to identify and mitigate common statistical problems such as variable confounding, low statistical power, and over-interpolation/extrapolation among multiple populations of data. Quantitative knowledge of variables is maintained across combinations of datasets, while reducing the time, budget, and data limitations of a single actor. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, Computational Materials Science & Engineering, Additive Manufacturing |