Explainable fault diagnosis for smart cities
Final Report Abstract
This project has developed a decentralized, explainable artificial intelligence (XAI) framework for identifying sensor faults in “smart” civil infrastructure, encompassing civil engineering structures equipped with wireless structural health monitoring (SHM) systems. Unlike conventional SHM systems that rely on centralized data processing with limited transparency, automated fault diagnosis in this project has been formulated as a stand-alone classification problem, enabling the identification of both single and combined sensor faults using machine learning at a wireless sensor node level, rendering the SHM systems scalable, flexible, and transparent. To perform classification, long short-term memory (LSTM) networks have been trained using time-series sensor data, including both non-faulty signals and signals with artificially injected faults. The LSTM networks have been optimized to run on low-power wireless sensor nodes. To enhance transparency, explainability has been integrated into the models using Shapley additive explanations (model-agnostic) and gradient × input (model-specific). The explainable LSTM networks have been embedded into wireless sensor nodes using Internet-of-Things principles, enabling local fault identification and explanation. For user interaction and visualization of fault types and relevance scores, a dashboard based on the lightweight Message Queuing Telemetry Transport (MQTT) protocol has been developed. The decentralized XAI framework has achieved significant reductions in bandwidth usage and has proven suitable for application in resource-constrained environments. Validation experiments have been carried out using both artificial and real-world data, including a SHM system installed on a pedestrian bridge. The results have shown high classification accuracy (>90 %) and consistent, interpretable predictions from the LSTM networks. In summary, the proposed framework has provided a reliable and transparent solution for smart infrastructure monitoring, offering a promising foundation for future smart city applications.
Publications
-
Automated decision making in structural health monitoring using explainable artificial intelligence. In: Proceedings of the 28th International Workshop on Intelligent Computing in Engineering (EG-ICE). Berlin, Germany, 06/30/2021.
Peralta, J., Fritz, H., Dadoulis, G., Dragos, K. & Smarsly, K.
-
A systematic survey of Internet of Things frameworks for smart city applications. Sustainable Cities and Society, 83, 103949.
Peralta Abadía, José Joaquín; Walther, Christian; Osman, Ammar & Smarsly, Kay
-
An Introduction and Systematic Review on Machine Learning for Smart Environments/Cities: An IoT Approach. Intelligent Systems Reference Library, 1-23. Springer International Publishing.
Peralta Abadía, José Joaquín & Smarsly, Kay
-
Internet of Things Frameworks for Smart City Applications—A Systematic Review. Computing in Civil Engineering 2021, 83-89. American Society of Civil Engineers.
Peralta Abadía, José Joaquín & Smarsly, Kay
-
SENSOR FAULT DIAGNOSIS COUPLING DEEP LEARNING AND WAVELET TRANSFORMS. Proceedings of the 13th International Workshop on Structural Health Monitoring. Destech Publications, Inc..
ABADÍA, JOSÉ JOAQUÍN PERALTA; FRITZ, HENRIEKE; DRAGOS, KOSMAS & SMARSLY, KAY
-
A deeplearning-based approach towards identifying combined faults in structural health monitoring. In: Proceedings of the 11th European Workshop on Structural Health Monitoring (EWSHM). Potsdam, Germany, 06/10/2024.
Al-Nasser, H., Al-Zuriqat, T., Dragos, K., Chillón Geck, C. & Smarsly, K.
-
AIoT- enabled decentralized sensor fault diagnosis for structural health monitoring. In: Proceedings of the 11th European Workshop on Structural Health Monitoring (EWSHM). Potsdam, Germany, 06/10/2024.
Chillón Geck, C., Al-Zuriqat, T., Elmoursi, M., Dragos, K. & Smarsly, K.
-
Identification of combined sensor faults in structural health monitoring systems. Smart Materials and Structures, 33(8), 085026.
Al-Nasser, Heba; Al-Zuriqat, Thamer; Dragos, Kosmas; Geck, Carlos Chillón & Smarsly, Kay
-
Identification of composite sensor faults in structural health monitoring systems using long short-term memory networks. In: Proceedings of the 2024 European Conference on Computing in Construction (EC3). Chania, Crete, Greece, 07/15/2024.
Al-Zuriqat, T., Al-Nasser, H., Dragos, K., Chillón Geck, C. & Smarsly, K.
-
Explainable sensor fault diagnosis for structural health monitoring. In: Proceedings of the 11th ECCOMAS Thematic Conference on Smart Structures and Materials (SMART 2025). Linz, Austria, 07/01/2025.
Al-Nasser, H., Dragos, K. & Smarsly, K.
