Maximizing System Identification Optimal Sensor Placement and Value of Information in Modal Parameter Estimation

Characterizing the short- and long-term dynamic behavior of civil structures and infrastructures is critical to understanding their performance and reliability over time. It is common to describe their structural behavior in the modal space in terms of natural frequencies, damping ratios, and mode shapes. These parameters provide quantities of interest that can be interpreted from a physical-mechanical perspective and can be used for various purposes, such as model updating, damage detection, and performance assessment. However, several challenges exist in the extraction strategies for the modal parameters. First, modal identification techniques often require selecting user-defined parameters, which influence the system identification process and, consequently, the results. Second, the selection of sensors, their position, and their number are crucial: an incorrect choice of sensor or position can result in missing essential information for identification. The quantity has an impact, as each sensor provides information, but also introduces a certain amount of noise into the measurement: in the system identification, it causes uncertainties in the estimation of the structural modes, as it contaminates the signal, covering the data of interest; moreover, increasing the number of sensors leads to higher costs. This thesis aims to develop and perform a strategy for investigating the relevance of the identification algorithm's parameters and sensor placement choices made by the user on the modal identification results. Therefore, system identification is first carried out using data-driven stochastic subspace identification (DD-SSI), varying the user-defined parameters. Then, based on the results in terms of frequency, a sensitivity analysis is conducted using analysis of variance (ANOVA) to understand which parameters have an influence on the results. Subsequently, a study on optimal sensor placement is conducted using a value of information (VoI) based approach that considers the trade-off between the information gained in the system identification and the costs. The objective is to maximize the identification of modes using entropy as a metric to evaluate the quality of the data collected by the sensors while minimizing overall costs, which include the cost of sensors, installation and maintenance costs, and computational costs. This is achieved through multi-objective optimization methods, such as objective function and Pareto front approach. Additionally to this study, the search for the best combination of DD-SSI parameters was integrated, resulting in a three-dimensional outcome: VoI, cost, and combination of DD-SSI parameters. To validate the methods, a numerical case with 4 degrees of freedom was used for system identification by varying the parameters, as well as to understand the influence through ANOVA. The full strategy was tested on the 'Viaduc de Chillon'. From the results, it was observed that the parameters that most influence the DD-SSI results are the number of block rows of the Hankel matrix, the partition of the Hankel matrix, and the system order. Furthermore, different optimal solutions were found depending on the weight factor, which balances information and cost; depending on its value, either the information data or the cost component becomes more significant. The Pareto front solution is also described. The results show the importance of conducting such analyses, as they allow us to obtain high-quality results while substantially reducing costs.