From Design to Production for Metal Additive Manufacturing: A Dual Focus on Model-Based Design Guidance and Workforce Preparedness

Abstract

Effective use of additive manufacturing (AM) design freedom and rapid manufacturing capabilities requires designers to promptly address geometry-related quality issues, such as distortion, to ensure part accuracy. To date, AM part design has been a challenging and expensive process, given the complexity and cost of the AM process, the design freedom, and the limited availability of AM data. As a result, ensuring part design for improved quality hinders the broader adoption of AM technologies and may discourage designers from engaging with AM. The trial-and-error design process, based on time-consuming simulations, experiments, and ad-hoc rules, highlights the need for more efficient, data-driven design frameworks. Data-driven models can provide faster predictions of AM process-induced distortion, generate design guidelines by analyzing critical geometric features, and perform design compensation based on these distortion predictions. My research has three distinct aims focused on improving AM design processes and preparation of students for AM: (1) compensate part designs for minimizing part quality issues for the metal AM process, particularly laser powder bed fusion (LPBF), (2) efficiently predict AM part quality and provide design modifications based on the critical geometric features, and (3) explore engineering students experience in engineering, their engineering identity formation and engagement in makerspace to provide a scalable solution that increase students engagement in design and making activities like AM. To address the first aim, a key dimensional characteristics-based geometry compensation integrated remanufacturing framework for reverse engineering (RE) and AM was proposed. The proposed two distortion compensation algorithms utilized the 3D CAD model and the STL model. Findings indicated that STL-based compensation underperformed the CAD-based approach. The deviation distributions of the four remanufactured parts (two case study parts with two compensation methods) and their corresponding nominal geometries had mean values ranging from 30.0 μm to 48.9 μm and standard deviations ranging from 66.2 μm to 78.6 μm. The second aim is to develop a machine learning-based geometry-driven distortion risk (low, medium, or high risk) prediction model for a broad range of axisymmetric geometries. This model identified critical geometric features of parts that contribute to geometrical deviations and quality prediction, and provided targeted design modification recommendations for parts based on their predicted quality measures. This approach can deliver fast and high-quality predictions for a wide range of parts. Shape descriptors accurately classified distortion risk (with an accuracy of 86.4% for 81 test parts) and recommended design modifications to reduce distortion risks based on the key geometric feature trends for distortion risk classes. The third aim of understanding diverse students' engagement in making, makerspaces, and engineering, as well as their engineering identity formation, revealed emergent themes on making. For example, men and womxn (students who identified as women as one of their gender identities) may hold differing perspectives or exhibit varying interests in makerspace engagement, with men focused on specific technologies in the makerspace, while womxn are focused on space, community, and projects, and students feel more like engineers by the end of the semester. These findings can help develop more inclusive and engaging engineering courses, as well as support the development of the AM curriculum. The ML-based part distortion prediction and design recommendation, as well as the geometric feature-based design compensation model and the exploration of differences in engineering students' experiences and engagement in design and manufacturing, can support the broader goal of transitioning seamlessly from design and geometry capture to AM parts with reduced distortion. Model-based methods can capture the interactions among geometric features that impact part distortions, providing accurate and rapid predictions of distortion, design modifications, and compensation. Taken together, this dissertation reinforces the need for model-based methods and workforce training on design for quality that ensures both design intent and part quality in metal AM.