Domain-Adaptive Deep Learning for Manufacturing Prognostics and Surrogate Modeling

ABSTRACT

The rapid advancement of artificial intelligence (AI) and data analytics are revolutionizing advanced manufacturing. As a pillar of AI, deep learning (DL) models play a foundational role in developing digital solutions to prognostics and surrogate modeling for manufacturing processes and systems. The performance of DL models highly depends on the statistical distribution and characteristics of the data used during model training. In manufacturing, significant variations can be introduced into the process and system data by the different materials, operations, and build parameters used. To adapt deep learning models to such data variations toward high-fidelity prognostic and modeling results, the deep learning models should have new structures that accommodate data’s statistical domain shift and auxiliary information coming from the process or system. Toward this goal, my group investigated domain adaptation and designed multiple domain-adaptive deep learning models for prognostics in sophisticated manufacturing processes and surrogate modeling conditionally on manufacturing build parameters. This talk will mainly cover a domain-adaptive U-Net for prognostics in Aerosol Jet Printing and a Multi-Parameter Simulation Generative Adversarial Net for conditional surrogate modeling in Laser Powder Bed Fusion. These models are milestones toward knowledge-informed DL domain adaptation and digitalizing high-resolution 3D printing in future studies.

BIOGRAPHY

Shenghan Guo is an Assistant Professor in the School of Manufacturing Systems and Networks at Arizona State University. Her research centers around knowledge-informed data analytics and AI models. Her research group, Data Analytics & Insights in Manufacturing (DAIM), strives to develop knowledge-informed AI solutions for smart manufacturing processes and systems and aims to advocate for human-centricity in AI and smart manufacturing. She is experienced in statistical quality control, prognostics, and data mining. She has handled multiple research datasets from manufacturing fields, particularly those with complex properties, such as in-situ thermal video and multi-sensory data streams. The current applications of her research include in-situ prognostics, customized additive manufacturing, and manufacturing worker monitoring and training. Her lab hosts an OPTOMEC Aerosol Jet Printer for high-resolution flexible 3D printing, which supports experiments for electronic printing/packaging, human-machine interactions, and AI-assisted fabrication of new materials/structures. Education

Ph.D. in Industrial and Systems Engineering, Rutgers, The State University of New Jersey, New Brunswick, NJ, U.S. (2021)

M.S. in Engineering Sciences and Applied Mathematics, Northwestern University, Evanston, IL, U.S. (2016)

M.S. in Financial Mathematics, The Johns Hopkins University, Baltimore, MD, U.S.

B.S. in Financial Engineering, Jilin University, Changchun, China (2013)

Research website: https://sites.google.com/asu.edu/shenghanguo/home

Google Scholar: https://scholar.google.com/citations?user=fOwlLlEAAAAJ&hl=en&oi=ao

Zoom Link: https://stevens.zoom.us/my/se541a
Zoom Meeting ID: 422 166 1984
Passcode: meet@se541