YoungJoo Hyun

YoungJoo's research interests include computer vision and signal processing for manufacturing, with previous work on embedded visual inspection, explainable anomaly detection and vibration signal analysis.

Publications

Encoding time series as images for anomaly detection in manufacturing processes using convolutional neural networks and Grad-CAM

Y. J. Hyun, Y. Yoo, Y. Kim, T. Lee, W. Kim. International Journal of Precision Engineering and Manufacturing, 2024.

• Converts manufacturing force, vibration and sound signals into 2D representations using GADF, GASF, MTF and recurrence plots, so time-series anomaly detection can be done with CNN-based computer vision models
• Combines frequency-domain preprocessing with CNN classification and Grad-CAM to show which regions drive an anomaly prediction, giving operators an interpretable result
• Recurrence plot with ResNet50 reaches 99.6% accuracy on the PHM 2010 dataset, and the approach also exceeds 90% accuracy on real CNC-machining data

Signal separation of CNT-based vibration data using TVF-EMD and BSS for structural health monitoring

Y. J. Hyun, S. Lee, J. Lee, S. Lee, Y. Yoo. Measurement, 122948, 2026.

• Reconstructs multi-channel representations from single-channel CNT vibration signals using TVF-EMD, then applies blind source separation methods such as FastICA and Sparse PCA to recover the individual source components
• Compares EMD, TVF-EMD, TVF-EMD+SPCA, TVF-EMD+ICA and a CNN spectral-mask baseline under mixed-frequency and impact conditions, with SNR/SDR and parameter-sensitivity analysis
• TVF-EMD+ICA gives the best overall separation, reaching 22.61 dB SNR for the 30 Hz component in a 30 Hz + 3 kHz mixture and 21.48 dB under impact conditions