<p>Electron Beam Powder Bed Fusion (EB-PBF) is a pivotal additive manufacturing technology renowned for its ability to fabricate complex components with high precision and efficiency. However, the presence of defects such as porosity and powder spreading can compromise the mechanical properties and overall performance of the final products. Given the increasing use of EB-PBF for critical single-part production, maintaining quality standards through defect detection is crucial, yet conventional non-destructive evaluation methods are often costly or impractical, particularly for large, high-density components. This study presents a real-time monitoring system that uniquely leverages inherent electron emissions data for defect detection, eliminating the need for additional instrumentation while providing spatial defect distribution data. Unlike conventional approaches that rely on computationally intensive image or video analysis, our methodology utilizes time-series (profile) data from both in-situ and post-layer electron emissions, enhancing feasibility for real-time applications. We present a comprehensive comparative analysis of two deep learning models—Convolutional Neural Networks (Conv1D) and Multi-Scale Temporal Autoencoders (MSTAE)—to evaluate their effectiveness in detecting defects across varying severities, including low porosity, medium severity, severe porosity, and powder spreading defects. The models were assessed using reconstruction error metrics such as Mean Absolute Error (MAE), Huber Loss, Coefficient of Determination (R²), and Structural Similarity Index (SSIM). Additionally, Exponentially Weighted Moving Average (EWMA) control charts were employed to monitor and analyze defect detection performance in real time.</p>