EXPERIMENTAL EVALUATION OF A THREE-CHANNEL HYBRID DEEP LEARNING FRAMEWORK FOR INDUSTRIAL PROCESS ANOMALY DETECTION
- Authors
-
-
Yusupbekov N. R.
Tashkent State Technical University Named After Islam Karimov, Tashkent, Uzbekistan
Author
-
Avazov Y. Sh.
Tashkent State Technical University Named After Islam Karimov, Tashkent, Uzbekistan
Author
-
Rashidov G. Kh.
Tashkent State Technical University Named After Islam Karimov, Tashkent, Uzbekistan
Author
-
- Keywords:
- Kalman filtering, industrial anomaly detection, LSTM neural networks, autoencoder reconstruction, CNN–LSTM classification, time-series monitoring, hybrid deep learning, process fault detection.
- Abstract
-
This study presents an experimental evaluation of a hybrid deep learning framework for anomaly detection in industrial process monitoring using real multivariate sensor data obtained from a bio-oxidation system. The proposed architecture integrates Kalman filtering for noise suppression, LSTM-based forecasting for temporal deviation detection, autoencoder-based reconstruction analysis for identifying hidden structural anomalies, and CNN–LSTM classification for multi-channel feature fusion. Experimental results demonstrate that Kalman filtering significantly improves signal quality, enabling stable model training. The combination of prediction error and reconstruction error provides complementary anomaly indicators, enhancing detection robustness. The proposed system achieved good discriminative performance with a ROC–AUC score of 0.76 and demonstrated high sensitivity in detecting abnormal operating conditions. A sensitivity-oriented threshold strategy ensured complete anomaly coverage, which is critical for safety-sensitive industrial applications. Event-level evaluation confirmed that all anomalous process events were successfully detected. The results indicate that the proposed hybrid monitoring framework provides an effective solution for real-time industrial anomaly detection, early fault diagnosis, and intelligent decision support in complex technological systems.
- References
-
1.
R. E. Kalman, “A new approach to linear filtering and prediction problems,” Journal of Basic Engineering, vol. 82, no. 1, pp. 35–45, 1960.
2.
S. Hochreiter and J. Schmidhuber, “Long short-term memory,” Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997.
3.
H. Wang, M. Liu, and Y. Zhang, “Anomaly detection in industrial time-series data using deep learning,” IEEE Transactions on Industrial Informatics, vol. 16, no. 3, pp. 1928–1937, 2020.
4.
J. An and S. Cho, “Variational autoencoder based anomaly detection using reconstruction probability,” in Proc. ICML Workshop on Anomaly Detection, 2015.
5.
Y. Malhotra et al., “LSTM-based encoder–decoder for multi-sensor anomaly detection,” arXiv:1607.00148, 2016.
6.
G. Hinton and R. Salakhutdinov, “Reducing the dimensionality of data with neural networks,” Science, vol. 313, no. 5786, pp. 504–507, 2006.
7.
F. Chollet, Deep Learning with Python. Manning Publications, 2018.
- Downloads
- Published
- 2026-03-02
- Issue
- Vol. 2 No. 2 (2026)
- Section
- Articles
- License
-
This work is licensed under a Creative Commons Attribution 4.0 International License.
