Motivation

Wireless sensor networks have matured to a promising alternative to conventional tethered systems for structural health monitoring (SHM). The reliability of wireless SHM systems largely depends on the tolerance against faulty or incomplete data collected by wireless sensor networks. Current approaches towards fault-tolerant SHM are usually based on classical fault diagnosis that builds upon “analytical redundancy”, eventually aiming at repair or replacement of faulty data. Despite the known advantages of classical fault diagnosis, the complete fault diagnosis procedure is computationally costly, because faulty sensor data must automatically be detected, isolated, identified, and accommodated. The computational efficiency can be increased by using advanced signal analysis tools of applied harmonic analysis to allow working directly with incomplete or faulty data.

Goals and methods

The main goal of this collaboration project is to develop and to validate a methodology for fault-tolerant wireless SHM based on applied harmonic analysis using (i) frame analysis and (ii) deep learning (Fig. 1.). First, frame analysis (or, more specifically, fusion frame analysis) will be coupled with generalized sampling, enabling computationally efficient reconstruction of signals even in the case of faulty or incomplete sensor data. In addition, frame analysis supports sparse reconstruction of signals, where only a subset of sensor data is used. Sparse reconstruction generates lower computational costs while increasing fault tolerance levels of the wireless SHM systems considered, because the robustness of signal reconstruction is not influenced by the absence of sensor data. Second, to increase the level of tolerance against systematically faulty sensor data, deep learning algorithms are proposed, complementing frame analysis by convolutional neural networks, thus reducing the influence of faulty sensor data on the monitoring quality. As an outcome of the project, it is expected to provide a computationally efficient and robust methodology for fault-tolerant wireless structural health monitoring.