Hybrid LSTM-CNN Framework for Predictive Anomaly Detection in Operational Technology Networks
Ślązak, Julia (2025)
2025
Tietojenkäsittelyopin maisteriohjelma - Master's Programme in Computer Science
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
Hyväksymispäivämäärä
2025-12-01
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-2025120111123
https://urn.fi/URN:NBN:fi:tuni-2025120111123
Tiivistelmä
Industrial automation has become increasingly significant since the mid-20th century, and today, many industrial systems are shifting toward remote operation. This transition imposes stringent requirements on the reliability and performance of communication networks that interconnect industrial devices. These networks produce large volumes of heterogeneous data, which can be leveraged to gain insights into system behavior and detect anomalies. This thesis investigates the application of advanced signal analysis techniques combined with Machine Learning (ML) models to enhance anomaly detection in industrial environments. The research aims to bridge theoretical approaches with practical implementation, contributing to improved monitoring and operational resilience in modern industrial networks.
The work involved designing and implementing a real-time anomaly detection system tailored for industrial communication networks. A dataset was collected from a real-world testbed, preprocessed, and analyzed to extract relevant features. Following an extensive literature review and theoretical research, the data were validated using statistical methods to ensure reliability. Based on these findings, the system architecture was developed to support real-time inference, integrating anomaly detection into a monitoring pipeline suitable for industrial applications. Performance metrics such as accuracy and precision were evaluated to confirm the feasibility and effectiveness of the proposed solution.
The results demonstrate that ML-based approaches can significantly improve anomaly detection compared to traditional threshold-based methods. The implemented system achieved high detection accuracy while maintaining real-time processing capabilities, proving its applicability in operational environments. However, limitations related to data diversity and sensitivity were identified, suggesting the need for broader datasets and privacy-preserving techniques in future work. Overall, this study confirms the potential of intelligent anomaly detection systems to enhance the reliability and security of industrial networks.
The work involved designing and implementing a real-time anomaly detection system tailored for industrial communication networks. A dataset was collected from a real-world testbed, preprocessed, and analyzed to extract relevant features. Following an extensive literature review and theoretical research, the data were validated using statistical methods to ensure reliability. Based on these findings, the system architecture was developed to support real-time inference, integrating anomaly detection into a monitoring pipeline suitable for industrial applications. Performance metrics such as accuracy and precision were evaluated to confirm the feasibility and effectiveness of the proposed solution.
The results demonstrate that ML-based approaches can significantly improve anomaly detection compared to traditional threshold-based methods. The implemented system achieved high detection accuracy while maintaining real-time processing capabilities, proving its applicability in operational environments. However, limitations related to data diversity and sensitivity were identified, suggesting the need for broader datasets and privacy-preserving techniques in future work. Overall, this study confirms the potential of intelligent anomaly detection systems to enhance the reliability and security of industrial networks.