Enhancing In-Situ Monitoring in Additive Manufacturing with Conv1D Autoencoder: A Deep Learning Approach to Analyze Total Electron Emission 1D Signal

This paper presents a ConvlD Autoencoder within a CNN framework for real-time anomaly detection in EB-PBF manufacturing, using TEE data for improved quality control. Our model is trained to identify deviations signaling defects and refine quality assurance. The approach includes data acquisition, preprocessing, and a detailed setup of the detector plate for optimal TEE signal capture. The efficacy of this model in detecting a range of defects is validated by setting an anomaly detection threshold at the 95th percentile of training error. Results confirm the model's capability to identify anomalies, indicating its utility in advanced manufacturing monitoring systems. This proof of concept paves the way for future model enhancements and data expansion, intending to advance EB-PBF process reliability and efficiency through deep learning.