Session: 16-01-01: Poster Session: NSF-Funded Research (Grad & Undergrad)

Paper Number: 99733

99733 - Design De-Identification of Thermal History for Cross-System Process-Defect Modeling of Metal-Based Additive Manufacturing 

One of the biggest limitations in the broader adoption of additive manufacturing (AM) techniques is the in-situ defect detection for part certification. It is crucial for users to detect process anomalies in an effective and timely manner since the offline counterpart methods have proven costly and time-consuming. However, building a robust in-situ process monitoring model is also very challenging due to the high part complexity, large amounts of training data required, and the highly variable part designs and printing parameters in diversified AM applications. These obstacles pose serious challenges for AM users, especially those working with small-to-medium sized manufacturers (SMMs).

One potential solution is to aggregate process data from multiple AM users. Transfer learning allows for AM process data from different users in related domains to be combined to create a larger and more diverse training set. The aggregated training data can then be leveraged to develop a more accurate, robust, and generalizable machine learning model for anomaly detection. Furthermore, these models would require less training data from each user than traditional independent machine learning models. Unfortunately, the major practical obstacle in aggregating process data from multiple AM users is the data privacy concerns that arise from sharing process data outside of the user’s organization. AM process data contains critical product design information, which heavily involves the intellectual property (IP) of the individual user. Sharing these data outside the user’s organization can potentially expose the AM users to the risk of IP theft.

This work proposes an Adaptive Design De-identification for Additive Manufacturing (ADDAM) methodology for masking the design information in the AM thermal process data, while simultaneously retaining the quality related information for process monitoring. This methodology will allow for the secure sharing of AM process data among multiple users, which establishes the foundation for data aggregation and transfer learning modeling, leading to the privacy-preserving cross-system anomaly detection models with improved IP security and model robustness. The technical contributions include: 1) the development of process data privacy and design de-identification framework for AM applications; and 2) the development of the ADDAM algorithm with measurable privacy and utility for AM process data.

A support vector machine (SVM) classifier is used to evaluate the performance of the proposed ADDAM algorithm in terms of both privacy gain and utility loss. A real-world case study is used to validate the proposed method based on the fabrication of two cylindrical shaped disks using the directed energy deposition (DED) process. The results demonstrate a significant improvement in the privacy gain, while simultaneously limiting the loss in anomaly detection performance.

Presenting Author: Durant Fullington Mississippi State University

Presenting Author Biography: Durant Fullington is a second year M.S. and Ph.D. student in the Industrial and Systems Engineering Department at Mississippi State University. He previously received his B.S. in Industrial and Systems Engineering from Mississippi State University in 2021. His current research focus includes the development of privacy-preserving de-identification models and cross-system certification models for metal-based additive manufacturing applications. In addition, his other research interests include machine learning, deep learning, and transfer learning techniques for additive manufacturing applications.

Authors:

Durant Fullington Mississippi State University
Wenmeng Tian Mississippi State University