164 C. Rainieri and G. Fabbrocino Fig. 15.1 Students developing their classwork with the support of their laptop Starting from the analysis of sensor specifications, the question of appropriate attachment of the sensors is introduced, showing how it can affects the bandwidth and even alter the measurements. Applicative examples are proposed, showing that accelerometers can be mounted by a variety of methods, including magnet, adhesive, and stud or screwed bolt. Each of these methods has advantages and limitations. The importance of stiff connection and adequate preparation of the surface is remarked. As previously mentioned, sensor location is responsible for the observability of structural modes. The fundamental reasons, which may require moving the sensors from one location to another, are presented. In particular, the problems related to the installation of sensors in correspondence or very close to nodes of the structural mode shapes, to poor spatial resolution of measurements and unfortunate choice of sensor layout are presented by means of applicative examples. Typical sensor layouts for different structural schemes are also presented (Fig. 15.2). Even if they represent only crude guidelines for the design of appropriate sensor layouts, they are useful to simplify test planning in the presence of regular structures. In order to apply the previously mentioned notions, at least one field application on a real structure (Figs. 15.3 and 15.4) or a scaled model (Fig. 15.5) is proposed. All relevant details are provided in order to allow the student to autonomously set up their first experimental test. After having collected their first dataset, taking into account the importance of the appropriate setting of sampling frequency and total record duration [2], they are committed with data validation and pre-treatment. In particular, a specific application is proposed, where the students visually inspect the individual time histories and estimate the corresponding probability density functions and power spectral density functions in order to detect anomalies, such as excessive noise, noise spikes, clipping, and so on. Then, they apply basic linear algebra tools (least squares) or filters to remove eventual spurious trends in the data. Once validated and pre-treated data are available, the students have to develop their own modal parameter estimation software. Among the OMA methods available in the literature, the peak picking method [5] and the Frequency Domain Decomposition (FDD) method [6] are proposed because they are characterized by a simple theoretical background and ensure a fast implementation once auto- and cross-power spectra are estimated. When these methods are implemented and applied, the main difficulty is related to the need of dealing with complex numbers and complex-valued vectors. In our experience, it is important to recall the basics and the different representations of complex numbers before illustrating the OMA algorithms.
RkJQdWJsaXNoZXIy MTMzNzEzMQ==