Background and motivation

Wireless structural health monitoring (SHM) systems play are crucial in ensuring safety and longevity of civil infrastructure. However, the reliability of SHM systems is highly dependent on the accuracy of sensor data. Faulty sensors may lead to misinterpretation of structural conditions, particularly when multiple faults occur simultaneously. Conventional fault diagnosis systems often lack transparency, limiting trust in artificial intelligence (AI)-based decisions. This project addresses the growing need for explainable artificial intelligence (XAI) in fault diagnosis, supporting the development of trustworthy and interpretable diagnosis tools for SHM. The research builds upon the prior DFG-funded project “Explainable fault diagnosis for smart cities“ and is supported by a one-year completion scholarship funded by the Hamburg Act for the Promotion of Young Researchers (HmbNFG).

Proposed concept

The project aims to develop a decentralized XAI framework capable of diagnosing combined sensor faults and explaining the diagnosis decisions in real-time monitoring environments (Figure 1). Long short-term memory networks are used to model temporal dependencies in sensor signals, enabling identification of combined sensor faults. To ensure transparency in the decision-making process, model-specific and model-agnostic XAI methods, such as integrated gradients and Shapley additive explanations, are integrated. A comparative evaluation framework is also established to benchmark the diagnosis accuracy and interpretability of different XAI methods. The decentralized nature of the concept proposed in this project allows for sensor-level fault diagnosis, reducing reliance on centralized processing and expensive computational resources.

Expected results

The project is expected to result in (i) a validated AI-based framework for diagnosing combined sensor faults in SHM systems, (ii) a comparative study for XAI methods adapted for sensor fault diagnosis, and (iii) an interpretable fault diagnosis tool that supports real-time and decentralized decision-making.