An Anomaly Detection Method Augmented by Domain Knowledge and Predictive Residuals Alan Bilgisi ve Öngörücü Kalintilarla Zenginleştirilmiş Anomali Tespit Yöntemi
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/siu71813.2026.11636559
- Basıldığı Şehir: İstanbul
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: Anomaly detection, early warning systems, multivariate time series, predictive maintenance
- Orta Doğu Teknik Üniversitesi Adresli: Evet
Özet
Predictive maintenance in industrial systems requires early and reliable detection of anomalous behavior from highdimensional sensor data. In railway air production units, failures often emerge gradually due to complex interactions between physical variables. In this paper, we propose a physics-informed early warning framework combining domain-driven feature engineering with forecasting-based residual analysis. The proposed approach integrates efficiency-based degradation metrics and LSTM residuals into a stacked attention autoencoder. Experiments on the MetroPT-3 dataset demonstrate that the proposed method significantly improves detection lead time and event-level recall compared to baseline models.