FMU-Based Residual Decomposition and Dual LSTM Autoencoder Fusion for Anomaly Detection in Collaborative Robots FMU Tabanli Kalinti Ayriştirma ve ?Ikili LSTM Özkodlayici Birleşimi ile ?I şbirlikçi Robotlarda Anomali Tespiti
34th Signal Processing and Communications Applications Conference, SIU 2026, İstanbul, Türkiye, 7 - 10 Temmuz 2026, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/siu71813.2026.11636980
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: anomaly detection, collaborative robots, complementarity analysis, LSTM autoencoder, residual decomposition, score fusion
- Eskişehir Osmangazi Üniversitesi Adresli: Evet
Özet
This paper presents a score-level fusion framework combining physics-informed and data-driven Long Short-Term Memory (LSTM) autoencoders for anomaly detection in collaborative robots, evaluated on 1.1 million real sensor samples from a UR10e robot. A residual LSTM autoencoder operates on 12-channel intrinsic/extrinsic residuals from an FMU-based inverse dynamics model, while a raw LSTM autoencoder processes 24-channel direct sensor data. Evaluation with four synthetic fault types (motor drift, collision, encoder glitch, sensor noise) shows that weighted-average fusion reaches high detection performance and yields a substantial F1 improvement over the best individual model. Window-level complementarity analysis reveals that roughly one-quarter of anomalies are detected only by the residual model and one-quarter only by the raw model, confirming genuine complementarity between the two representations.