Motivation and research problem

Recent trends towards digitalization have been shaping modern structural health monitoring (SHM) concepts, promoting wireless technologies for SHM with wireless sensor nodes capable of decentralized “on-board” data analysis. In this context, and given the lack of constant sensor data transmission, the reliable and fault-tolerant operation of wireless sensor nodes is of paramount importance, motivating the development of sensor fault diagnosis (FD) approaches for SHM. FD approaches that have been proven useful in related disciplines cannot simply be adapted to SHM, due to the characteristics of the structures being monitored. Moreover, existing FD approaches may oversimplify reality, frequently assuming that sensor faults appear in individual sensors and with single fault types, while the structural condition is considered unchanged. In practice, though, sensor faults may occur simultaneously in multiple sensors (or sensor nodes), sensor faults may be combined, and the structural condition may change, as shown in the figure below.

Research objectives

The project aims to develop a framework for robust sensor FD in modern wireless SHM systems. The proposed framework will include a sensor FD methodology that will address multiple dimensions, i.e. (i) all FD steps (detection, isolation, identification and accommodation) and all common sensor fault types, (ii) real-world sensor fault dimensions (such as single-sensor faults, simultaneous sensor faults as well as single and combined fault types), and (iii) the diagnosis of sensor faults in the event of structural damage. In addition, the framework will include a design concept towards decentralizing the sensor FD methodology into wireless sensor nodes, thus, enabling the performance of self-diagnostics at the edge of wireless sensor networks. In this direction, artificial intelligence techniques will be coupled, enabling decentralized, fault-tolerant sensor data analysis. To provide robustness of the framework, the steps towards detecting, isolating, identifying and accommodating sensor faults will be based on sound mathematical principles. The research hypothesis is formulated as follows: The reliability of modern wireless structural health monitoring systems stands to benefit from a multi-dimensional sensor fault diagnosis framework, enabling robust fault diagnosis of real-world sensor faults “at the edge” of wireless sensor networks, i.e. in an automated/decentralized fashion without user interaction.

Expected outcome

The outcome of the project is expected to enhance the alignment of modern SHM concepts, centering around wireless sensor networks that feature on-board data analysis, with emerging trends on edge computing through integrating robust self-diagnostic capabilities into the wireless sensor nodes themselves. As a result, wireless sensor nodes, typically already capable of performing SHM tasks locally and quasi-independently, are expected to function as fully-fledged, fault-tolerant Internet-of-Things devices with enhanced operational autonomy.