Semester of Graduation

Summer 2026

Degree

Master of Science in Industrial Engineering (MSIE)

Department

Mechanical and Industrial Engineering

Document Type

Thesis

Abstract

Additive manufacturing requires quality assurance methods capable of detecting process anomalies before defects propagate across subsequent layers. This thesis develops a trustworthy artificial intelligence framework for real-time surface anomaly detection in Fused Deposition Modeling (FDM) using laser-derived depth maps and Vision Transformers (ViTs). Conventional convolutional models aggregate global context gradually through stacked local kernels, which can limit their sensitivity to spatially distributed surface deviations. The proposed approach instead tokenizes depth-map patches and applies global self-attention to classify four process conditions: normal printing, under-extrusion, over-extrusion, and empty or void regions. The framework combines token-level attention visualization, vanilla saliency, Integrated Gradients, and t-SNE and UMAP projections to examine where the model attends, which surface regions influence class scores, and how process states are organized within the learned representation space. Classification performance is assessed using class-wise precision, recall, F1-score, one-vs-rest ROC-AUC, bootstrap confidence intervals, baseline comparisons, and inference latency. On the evaluated data set, the model achieves a macro-F1 score of 0.877 and a macro-averaged one-vs-rest AUC of 0.972. The explanation results indicate that model predictions are associated with physically meaningful gaps, ridges, void boundaries, and deposition patterns. The model processes a complete printed layer within the typical inter-layer inspection window on the evaluated GPU, demonstrating the potential of globally attentive models for accurate and operator-auditable FDM monitoring. Future research will focus on multi-sensor integration, cross-platform validation, model calibration, and closed-loop process control for broader real-time deployment.

Date

7-13-2026

Committee Chair

Bappy, Mahathir

LSU Acknowledgement

1

LSU Accessibility Acknowledgment

1

Download

Share

COinS