Trillingsgebaseerde lokalisering van schade in plaatvormige structuren – een breedband-gedreven aanpak via spaarse schatting

Efficient damage identification is a crucial technology in a variety of industries, where it offers economical, environmental, and societal benefits. This thesis is interested in the first-line inspection of plate-like structures, an active research topic with a broad range of practical use-cases. Vibration-based damage identification is a natural choice for this application, which offers practical, cost-effective solutions to inspect both metallic and composite components. Within the landscape of vibration-based approaches, this dissertation identifies a number of trends which have emerged to improve the efficiency of damage localization, by limiting the reliance on high-fidelity modelling, and by reducing the number of sensors and overall inspection effort. This insight motivates this research to continue the work of colleagues on the ``mid-frequency'' approach, which is shown to combine those trends in a novel and promising manner. The adopted approach aims to pair the practicality and robustness of low-frequency modal testing, with the sensitivity and resolution of scatterer imaging. The present research aims to develop this approach into a mature inspection technique that can tackle representative inspection cases on complex structures with real damage. As a first step, this thesis validates the assumptions underlying scatterer imaging by examining low-frequency modal testing data of a car door. While this first analysis is promising, it also demonstrates that several improvements are required to tackle complex cases with real damage. In response, this dissertation develops the frequency-coupled group lasso imaging method, which takes full advantage of the broadband nature of modal testing measurements. In addition, the proposed method gives complete freedom in terms of sensorization, and can be meaningfully tuned using an automated strategy. This thesis demonstrates detection and localization of a hole in a car door, using only a single accelerometer for the largest diameter, and any set of 5 accelerometers for the smaller diameters. These results are shown to be competitive with reference works applying ultrasonic arrays to simpler structures with comparable damage. Through a comprehensive statistical analysis, this dissertation quantifies the performance improvement brought by the coupling of frequencies. On a reference composite plate case, ROC analysis shows that the frequency-coupled method consistently outperforms the decoupled method, is more robust to noise, and less sensitive to sensor placement. In addition, it remains accurate when damage is not on the imaging grid, even at coarser resolutions. To support this statistical analysis, this thesis also develops the DW-ROC, a backwards-compatible extension of the ROC, which is more representative of performance, and adds the ability to rank imaging results according to localization error.