Session: CIE–31: Graduate Student Poster Symposium

Paper Number: 148478

148478 - Diffusion Modeling Based Causal Data Fusion for Predictive Additive Manufacturing Digital Twins 

The distinct geometries, multi-scale material structures, and functional complexities inherent in Additive Manufacturing (AM) demand a comprehensive understanding of its diverse physical phenomena. With the proliferation of big data, digital twins (DTs) augmented with sophisticated machine learning techniques enhance the comprehension of AM processes through crucial causal relationships, notably Process-Structure-Property (PSP) relationships. Nonetheless, the dynamic intricacies of AM and the vast quantities and varieties of data present a significant research lacuna, particularly regarding 1) the integration of causality and 2) the development of sophisticated methods for accurately inferring representations for each modality and their interactions during data fusion.  To address the gap, the study proposes a novel methodology employing a denoising diffusion probabilistic model (DDPM), termed causal data fusion for AM via DDPM (CDF-AM-DDPM). CDF-AM-DDPM facilitates the unprecedented prediction of PSP causal relationships utilizing multi-modal, multi-scale AM data and the generation of newfound, synthesized PSP features based on learned distributions. The illustrative case study showcases a DDPM model adeptly integrating energy density and melt pool size characteristics layer-wisely for laser powder bed fusion, while capturing new, hidden causal linkage features. By leveraging DDPM's predictive capabilities and concentrating on the causal aspects of PSP relationships, this research offers more nuanced and comprehensive insights into AM processes. Consequently, the proposed methodology has the potential to improve simulation, prediction, and optimization of AM processes, thereby catalyzing advancements in manufacturing quality, efficiency, and innovation via predictive DTs. 

Presenting Author: Fatemeh Elhambakhsh Arizona State University

Presenting Author Biography: Fatemeh Elhambakhsh is a second-year Ph.D. student in Data Science, Analytics, and Engineering at Arizona State University. She gained her master's degree in industrial engineering from Iran University of Science and Technology in 2020. Her research interests include additive manufacturing, design optimization, AI, machine learning, and data analytics in manufacturing and design. Her current research focuses on utilizing deep generative modeling to construct and improve digital twins of additive manufacturing processes.

Authors:

Fatemeh Elhambakhsh Arizona State University
Hyunwoong Ko Arizona State University