Session Sheet - 2026 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2026): Big Data Analytics

2026 Annual International Solid Freeform Fabrication Symposium (SFF Symp 2026): Big Data Analytics
Program Organizers: David Leigh, University of Texas at Austin

Monday 1:30 PM
August 3, 2026
Room: Zlotnick Ballroom 6
Location: AT&T Center

Session Chair: Yan Lu, National Institute of Standards and Technology; Jia Liu, University of Florida

1:30 PM
A Risk‑Informed Digital Twin Framework for Qualification of Metal Additive Manufacturing Processes: Clay Mullins¹; ¹GKN Aerospace
     Additive manufacturing (AM) qualification remains a barrier to broader adoption of metal AM in aerospace and defense. Traditional material allowables frameworks assume material homogeneity and statistical stability, assumptions that do not hold for AM processes where local material properties depend on time‑resolved thermal and process history. As a result, qualification often requires extensive testing while still providing limited insight into spatially varying part performance.This presentation describes a risk‑informed framework for an AI‑enabled Digital Twin to support qualification of metal AM processes, demonstrated using Laser Metal Deposition with wire (LMD‑w). The framework leverages process log data together with inspection and mechanical test outcomes to enable uncertainty‑aware assessment of material performance and defect risk. Rather than replacing existing qualification practices, the Digital Twin is positioned as a decision‑support tool that supports traceability, quantifies uncertainty, and enables risk‑based determination of testing sufficiency.

1:50 PM
Modeling Surface Roughness on Fatigue Performance of Additively Manufactured Ti-6Al-4V with Machine Learning: Shehzaib Irfan¹; Nabeel Ahmad¹; Daniel Silva¹; Alexander Vinel¹; Shuai Shao¹; Jia Liu²; ¹Auburn University; ²University of Florida
    Fatigue life in as-built additively manufactured parts exhibits inherent variability due to surface roughness. This uncertainty of fatigue life poses a significant challenge in the widespread adoption of additively manufactured parts in mission-critical components. In this study, we investigate how the deep notches on the surface of additively manufactured Ti-6Al-4V affect the fatigue life of the specimens and their spread. We quantify the impact of surface roughness as a stochastic process and incorporate it into a fatigue prediction model via machine learning. The approach uses a physics-based model as its foundation to describe the relationship between stress and fatigue life, and incorporates Gaussian processes to model the random effects of surface-finish features, thereby quantifying the scatter in fatigue life. It has also been benchmarked against another recently proposed method, resulting in an improvement in the prediction mean absolute percentage error from 39% to 22%.

2:10 PM
Reliable In-Situ Flaw Detection of Thin-Walled Lattice Structures: Alex Riensche¹; Abdalla Nassar¹; Christopher Apple¹; Anil Chaudhary²; Alex Istrate²; Angelo Visco²; Ted Reutzel³; Jan Petrich³; Gregory Colvin⁴; Vishal Musaramthota⁴; Robert Ghobrial⁵; Ryan Peitsh⁵; Hui Wang⁶; Rebekah Downes⁶; Garrison Hommer⁷; Joy Gockel⁷; Craig Brice⁷; ¹ASTRO America; ²Applied Optimization; ³Penn State ARL; ⁴Honeywell; ⁵Lockheed Martin; ⁶Florida State; ⁷Colorado School of Mines
    The ability to build complex, thin-walled structures using laser-based powder bed fusion (PBF-LB) additive manufacturing exceeds our ability to economically inspect for flaws using conventional non-destructive techniques, e.g., computed tomography, ultrasonic, and 2D radiography. This work presents high-resolution, long-exposure-near-IR and illuminated-visible imaging combined with a machine learning framework trained on CT data to demonstrate in-situ flaw detection. The approach targets detection of geometric deviations and voids exceeding 300 um with 95% confidence level and 90% probability of detection. Progress towards this objective alongside time and cost savings are presented.

2:30 PM
Machine Learning Modeling for Real-Time Melt Pool Monitoring in Laser Powder Bed Fusion Additive Manufacturing: A Hybrid Approach: Inioluwa Emmanuel¹; Zhuo Yang²; Ho Yeung²; Xinyao Zhang¹; ¹FSU; ²NIST
    We benchmark AI/ML methods for real-time melt pool monitoring in laser powder bed fusion (LPBF), where inference latency and limited labeled data constrain deployable model design. Using 1,200 balanced melt pool images of Nickel superalloy 625 from the NIST AMMT platform, we frame anomaly detection as binary image classification and compare five models: three transfer learning architectures (ResNet50, EfficientNetB0, MobileNetV2), a Random Forest on EfficientNetB0 feature embeddings (hybrid), and a Random Forest on raw pixels (baseline). Each model is evaluated on accuracy, precision, recall, F1, AUC, training time, inference latency, and CPU and GPU usage relevant to open-architecture LPBF machines. The hybrid EfficientNetB0 plus Random Forest achieves F1 0.9451, accuracy 0.9458, AUC 0.9904, and 1.15 ms per-image inference, outperforming pure deep learning baselines on both accuracy and latency. Pairing pretrained convolutional features with classical ensembles offers a robust, deployable route to real-time melt pool anomaly detection under data-limited LPBF conditions.

2:50 PM Break

3:30 PM
Domain Adapted Large Language Models for Additive Manufacturing: Peter Pak¹; Amir Barati Farimani¹; ¹Carnegie Mellon University
    This work presents AdditiveLLM2 a collection of multi-modal, domain adapted large language models built upon the instruction tuned variants of open weight models (Gemma 3, Qwen 3, Gemma 4) using a relatively small dataset of around 50 million tokens. The dataset (AdditiveLLM2-OA) consists of open-access additive manufacturing journal articles with data extracted for the domain adaptive pretraining and visual instruction tuning processes. Various stages of the developed model are evaluated with the Additive-Manufacturing-Benchmark which consists of additive manufacturing domain specific tasks compiled published resources. AdditiveLLM2 exhibits proficiency in both language and vision based tasks, achieving accuracies upwards of 90% in general additive manufacturing knowledge. This domain adaptive pretraining and instruction tuning strategy outline an accessible specialization method for large language models to a domain such as additive manufacturing.

3:50 PM
Cross-Printer Reproducibility Modeling Using a Deep Auto-Encoded Multi-Domain Gaussian Process: Logan Heck¹; Krystel Castillo-Villar¹; Adel Alaeddini²; ¹Texas Sustainable Energy Research Institute / The University of Texas at San Antonio; ²Southern Methodist University
    Additive manufacturing (AM) enables complex geometries but suffers from inconsistent dimensional accuracy across identical printer units, limiting industrial scalability. This study proposes a Deep Auto-Encoded Multi-Domain Gaussian Process (DAE-MDGP) framework to predict dimensional deviations in stereolithography (SLA) with emphasis on cross-printer reproducibility. The approach integrates an autoencoder with Gaussian Process regression to learn a data-driven, nonstationary kernel capturing nonlinear interactions among process parameters and machine-specific behavior. Unlike fixed-kernel or single-domain methods, the learned kernel models similarity across multiple printers. The framework is evaluated using thirty prints of a complex part across three SLA printers under varied parameters. Results show DAE-MDGP achieves the lowest prediction error across multiple metrics, outperforming baseline and conventional models. These findings demonstrate improved predictive consistency across printers, enabling better parameter selection and more reliable AM production.